Podcasts about Hopper

united states pokemon pok world records hopper goldin nostalgix

Play Episode Listen Later Aug 10, 2026 13:49

Hopper apologizes for a mistake in the last episode then goes over three more World Records for Pokemon based video games!YouTube:https://www.youtube.com/@GameGallerynetLink to 10.0 Silver World Record:https://www.instagram.com/p/DbvxgyVorqP/Link to 9.8 Silver World Record:https://goldin.co/item/2000-gbc-game-boy-color-pokemon-silver-version-usa-sealed-video-game-mht9sd?queryId=eyJxdWVyeUlkIjoiYzAzMTFmMDAzMjAyMDhlODczZmYzMGVkNWMwNzAxYmUiLCJjYXJkSW5kZXgiOjF9Linke to 9.8 Aka World Record:https://goldin.co/item/1996-gb-game-boy-pocket-monsters-red-jpn-unopened-video-game-made-in-jq3p9y?queryId=eyJxdWVyeUlkIjoiZTVkNzU1ZThkOTM2OWE1ZGI5NjdhOWQ1MWQ2NjAzZmQiLCJjYXJkSW5kZXgiOjExfQ%3D%3DLink to Pocket Monsters Blue World Record:https://www.instagram.com/p/DbyQlXGFOIg/?img_index=1Link to Pokémon USA variant guide:https://www.videogamesage.com/forums/topic/6629-pokemon-generations-i-to-iii-us-box-printvariant-guide-with-contents-and-population-surveyLink to A Few Games:https://www.afew.games/Solds/Images by Nostalgix, Goldin & arkscollectiblesFor edu. & entertainment#Pokemon

#225 Pokemon Records

Play Episode Listen Later Aug 9, 2026 18:53

Hopper goes over(4) NEW ALL-TIMEWORLD RECORDS:$1.2M Pokemon Red$450K Blue$300K Yellow$200K CrystalFor education &entertainmentImages owned byNostalgix.ig

records pokemon hopper

Rev. Dr. Harden Hopper Preaches on Matthew 14:22-23

rev james harden matthew 14 hopper preaches first united methodist church

Play Episode Listen Later Aug 9, 2026 16:08

Rev. Dr. Harden Hopper preaches on Matthew 14:22-23.First United Methodist Church of MariettaGiving link: https://onrealm.org/mariettafumc/-/form/give/nowChurch website: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://www.mariettafumc.org/

Stop Plucking Talent, Start Developing It with Spence Shelton

unSeminary Podcast

Play Episode Listen Later Aug 6, 2026 42:28

Welcome back to another episode of the unSeminary podcast. Today we're joined by Spence Shelton, Lead Pastor of Mercy Church, a rapidly growing multisite church in Charlotte, North Carolina. Since launching in 2015, Mercy has expanded to multiple campuses while maintaining a relentless focus on reaching people, developing leaders, and sending them into ministry. In this conversation, Spence shares how churches can intentionally build a culture of multiplication by developing leaders from within, creating healthy sending pathways, and stewarding opportunities for church revitalization. Stop “plucking” leaders. Start growing them. // One of Spence’s biggest challenges to church leaders is resisting the temptation to solve leadership shortages by hiring talent from elsewhere. Just as churches prioritize reaching lost people rather than simply attracting Christians from other churches, they should approach leadership development the same way. Instead of constantly recruiting experienced leaders, churches should intentionally identify, disciple, and develop people already sitting in their congregations. This approach takes longer, but it creates leaders deeply shaped by the church’s culture and mission. The power of the shoulder tap. // Many future ministry leaders never consider vocational ministry simply because no one asks them. Spence encourages pastors to intentionally “tap people on the shoulder” when they observe leadership potential. Some of Mercy’s key leaders—including executive pastors and campus pastors—came directly from marketplace careers like banking, manufacturing, and physical therapy. Rather than waiting for seminary graduates to apply, Mercy helps people connect their existing gifts and experiences to ministry opportunities. Create a clear leadership profile. // To make leadership development intentional, Mercy created a “Lead Pastor Profile” built around three categories: character, competency, and calling. While designed for future church planters, the framework serves as a roadmap for developing leaders throughout the church. By clearly defining what healthy leadership looks like, staff members know exactly what qualities to identify, cultivate, and reproduce in others. Character matters before charisma. // Within the character category, Spence believes self-awareness is one of the most overlooked leadership traits. Effective leaders understand both their strengths and limitations and are secure enough in their identity in Christ to build teams around areas where they are weaker. Without self-awareness, talented leaders can make excellent first impressions while creating long-term organizational damage. Character requires observation over time and not just impressive interviews. Multipliers build movements. // When evaluating leadership competency, the ability to multiply is critical. Can this person develop other leaders? Have they demonstrated the ability to move beyond accomplishing ministry personally and instead equip others to lead? Churches often unintentionally hire capable “doers” rather than leader-builders, limiting long-term growth. Multipliers create healthy ministries because they reproduce leadership instead of accumulating responsibility. Alignment creates unity. // Calling isn’t simply about theological agreement, but also includes ministry philosophy. Spence stresses the importance of ensuring leaders share the church’s convictions about how ministry is carried out. When someone consistently wants to reshape the church into something fundamentally different, the healthiest response may be helping them pursue the ministry God is calling them toward elsewhere. Clear alignment creates healthier teams and stronger long-term partnerships. Let people discover their gifts by leading. // Churches shouldn’t expect to know someone’s leadership potential before giving them opportunities to serve. Mercy intentionally creates pathways where people can lead community groups, coach leaders, and gradually assume greater responsibility. Some will flourish as long-term small group leaders. Others will demonstrate broader leadership gifting and eventually plant churches. The key is creating a culture where trying, learning, and growing are celebrated rather than feared. Revitalization begins with humility. // Spence also shares lessons from Mercy’s experience helping revitalize struggling churches. Successful partnerships require patient relationships, deep humility, and genuine respect for the legacy of existing congregations. Rather than viewing church buildings as opportunities to expand Mercy’s influence, the team approaches every conversation as the stewardship of Kingdom resources. By honoring the faithfulness of previous generations, Mercy has seen older congregations joyfully become part of new seasons of ministry fruitfulness. Faithfulness compounds over time. // Spence closes with an encouragement for church leaders: don’t underestimate what God can do through long-term faithfulness. Leaders often overestimate what can happen in five years while dramatically underestimating what God can accomplish over twenty. Healthy multiplication isn’t built through shortcuts. It grows steadily through consistent obedience, patient leadership development, and a commitment to stay the course. To learn more about Mercy Church, visit mercycharlotte.com or follow Spence Shelton or the church on social media for additional leadership resources and updates. Thank You for Tuning In! There are a lot of podcasts you could be tuning into today, but you chose unSeminary, and I'm grateful for that. If you enjoyed today's show, please share it by using the social media buttons you see at the left hand side of this page. Also, kindly consider taking the 60-seconds it takes to leave an honest review and rating for the podcast on iTunes, they're extremely helpful when it comes to the ranking of the show and you can bet that I read every single one of them personally! Lastly, don't forget to subscribe to the podcast on iTunes, to get automatic updates every time a new episode goes live! Thank You to This Episode’s Sponsor: Risepointe Do you feel like your church’s or school's facility could be preventing growth? Are you frustrated or possibly overwhelmed at the thought of a complicated or costly building project? Are the limitations of your building becoming obstacles in the path of expanding your ministry? Have you ever felt that you could reach more people if only the facility was better suited to the community’s needs? Well, the team over at Risepointe can help! As former ministry staff and church leaders, they understand how to prioritize and help lead you to a place where the building is a ministry multiplier. Your mission should not be held back by your building. Their team of architects, interior designers and project managers have the professional experience to incorporate creative design solutions to help move YOUR mission forward. Check them out at risepointe.com and while you’re there, schedule a FREE call to explore possibilities for your needs, vision and future…Risepointe believes that God still uses spaces…and they're here to help. Episode Transcript Rich Birch — Hey friends, welcome to the unSeminary podcast. So glad that you have decided to tune in. You know, this week is a great conversation and was recommended by one of my favorite church leaders in the country. And so you’re going to love this conversation. It’s going to be super helpful. And I do think it’s going to challenge you and provide you some steps forward. Rich Birch — Excited to have Spence Shelton with us. He is from Mercy Church, a multi-site church with three locations, if I’m counting correctly, in Charlotte, plus Church Online. God is leading Mercy Church into a season of a bunch of expansion and investment in ministry facilities and outreach over a two-year discipleship initiative called All to Him. They have a goal to launch the Mercy Family of Churches in in late 2025, which I think is maybe launched, or we’ll see where that goes, and is aiming to form their own collective within the next five years. Super excited to have you on the show today, Spence.Spence Shelton — Man, such an honor to be with you, Rich. Man, this is, I’m a big, going to say it, I’m a big fan of yours, man… Rich Birch — That’s kind of you to say. Spence Shelton — …because you have produced so much helpful content for me as a leader to lean into and grow little by little. And I can only hope that I can contribute a little bit more to the great work you’ve been doing, man.Rich Birch — Come on, that’s amazing. Honored to have you with us and to learn from you today, Spence. And mutual friend in Andrew Hopper…Spence Shelton — That’s right.Rich Birch — …who I love, fantastic leader. And so shout out to Hopper. Thankful for what God’s doing in his church as well. So Mercy Church started in 2015. Give us a picture of kind of where things are at today. If we were to come to the church, what would we experience? Kind of talk us through the last season, kind of the last couple of years. What’s that look like?Spence Shelton — Yeah, man. So if you came to Mercy today, what you’d find is a church, you might be surprised to learn how big it actually is because the church is, you know, a couple thousand people, but it’s across seven services and three campuses. And there’s some of some of that’s intentional. Some of that is, hey, we’re just kind of following the Lord where he’s led in this. And we didn’t really plan for it, you know, so. Spence Shelton — Our average service size is not that big, but we’ve the Lord’s been really gracious. If you came what you’d find is especially at our broadcast campus you’d find an old church building that you’d be thinking: wait are all these people how are they you know filling in here, you know. Rich Birch — Right. Spence Shelton — And you’d see cars parked uh all through a neighborhood and everything else and you’d see a retrofitted old church building, and that’s just the story of God’s grace on us – a building that was given to us and a church family that decided, hey we want to join you guys and they gave us their building. Really cool moment that like a legacy moment of faith in our church. And a couple other spots where we’re seeing similar sorts of things and it’s just it’s awesome, man. Spence Shelton — So we feel blessed. We feel like we are um doing what God has called us to do because we’re seeing that kind of momentum in the ministry that we’re doing. But none of our facilities are going to wow you, I promise. I actually love training church planters because, you know, so many times they come in and they’re like, Oh, ah we can do this. Rich Birch — Right. Spence Shelton — I’m like, yeah, you can do it. You know? Rich Birch — That’s normal like us.Spence Shelton — So we are praying and asking the Lord for better wins. Yeah. But you can do it, man. If you can do it here, you can do it anywhere kind of thing. So Lord’s been really gracious to us. And it’s a, it’s been a real joy these past 10 years.Rich Birch — It’s so cool. It seems like you guys are in a bit of a pivotal season or an inflection point season where you’ve got a bunch of things on the horizon, launching Mercy Family of Churches… Spence Shelton — That’s right. Rich Birch — …the 50 Missionary Goal by 2030, a collective. Talk me through those. What is the future look like? What’s driving the urgency? Help us understand kind of what the future looks like.Spence Shelton — Yeah, man, I think this happens for any church, both leader and in just the life of a church. You come to those milestone moments. We hit it last year with 10 years, and it was not only a milestone in terms of age. There was some just real growth that had been happening to the church that was forcing some good conversations around long-term you know expansion so that we can continue to facilitate what we believe God, actually, I should say we could steward what God is doing in our church.Spence Shelton — And so we launched an initiative that we called All to Him. I was teaching through the book of Romans all year last year. And so it comes out of Romans 12:1 and 2, because Christ gave his all for us, we give our all to him. That is our living sacrifice that we do. Spence Shelton — And so we said, we got some goals as a church, man. We want to see our church, we got some reaching goals. We got some training up goals. We got some sending out goals. Because when I planted Mercy Church, one of the things that was on my heart: if we’re going to come to Charlotte, North Carolina, we were going to be a sending church. Which means we’re going to build a sending hub here where this thing’s going be like an aircraft carrier that just sends out missionaries all over the place into the battles that the Lord has for us across the world.Spence Shelton — That was just on my heart. Instead of going to, you know, a place that might be, might have a, I don’t know, like a north Pacific Northwest or Northeast, something like that. If we’re coming to the Southeast to the Bible Belt, we’re doing so intentionally. Very intentional.Spence Shelton — So that’s been on our heart um ever since day one and this All to Him initiative. You know, we were hoping that by 2030, we double the size of our church. That’s because Charlotte is growing by 150 people a day.Spence Shelton — So like, we’ve got to, you know, this isn’t like it’s just because it’s something we want to do. It’s a calling, a good stewardship of the moment that we’re in. And we want to we’re launching a seminary certificate program for our members of our church because we want to train them up. Right.Spence Shelton — And we got a goal to see 50 missionaries go on the mission field by 2030. We just planted a couple of weeks before the recording of this, we planted our next church here in Charlotte, and we’ve got plans for several more.Spence Shelton — We are a large family of churches about to become a collaborative. We’re, we’re ah you know, 2028, 2029, somewhere around there, we’ll officially form that collaborative if the Lord continues to do what he’s doing. All because, man, we are convinced that God has brought us here to reach people, train them up, send them out, which I hope does not sound novel to any church leader listening to this.Spence Shelton — I hope it just sounds like, yeah, man, that makes sense.Rich Birch — Right.Spence Shelton — It’s like, well, then what’s the the magic or what’s the the secret sauce? It’s just putting your hand to the plow. And now here we are 10, 11 years in, and we’ve been steadily doing that same work. And I think the Lord is, um you know, our buddy Andrew would use the flywheel. That’s his favorite version of that illustration. But you keep putting your hand to the same thing and eventually the Lord starts to bear fruit. And that’s what we’re seeing now.Rich Birch — Yeah, so good. And yeah, you know that and you’re a kind leader, you know that for some leaders, the reason why we’re talking to you today is unfortunately, that is more novel than it should be. And so we want to dig into that a little bit.Spence Shelton — Sure..Rich Birch — Particularly, let’s maybe start with that middle section, this training goals, you’re working on trying to train people up. When you’re thinking about the kind of leadership vision you have as a church, you’re thinking about, hey, we’re going to raise up leaders both you know here to and then also sending.Rich Birch — I think we we can fall into this dichotomy around, okay, so do we find these leaders from other places and attract them to come here? Or, do we grow them up internally?Rich Birch — Do we find a way to develop them internally? So it’s this idea of plucking talent rather than developing it. Talk to me through why is training so hard in so many churches? Why do we struggle with that? And we seem to default towards, let’s just find somebody from over there and have them come and help us here.Spence Shelton — Yeah, man, I think that mentality starts all the way down to the idea of reaching people that don’t know Christ versus just sheep swapping from one church to the other church. Right.Spence Shelton — So you can grow your church by transfer growth. Um, which I actually am one who, Hey man, if somebody on the sideline in one church and then they plug into your church, I I’m all for it And if the same thing happens with Mercy, all for it. Spence Shelton — But our aim, what we all know as pastors and ministry leaders is our aim is man, let’s reach lost people and then disciple them. Right. Rich Birch — Yes.Spence Shelton — The same thing I think should be true with leadership development. For some reason, when it comes into developing leaders, we tend to go a different direction and think, well, let’s just go find where some leaders are and get them to to plug in.Spence Shelton — It’s like, we don’t do that even with discipleship. Why are we switching our mindset? And so I want to just kind of, I have to, I was convicted and wanted to call our church and I would call others to the same mindset that you have with discipleship. Let that carry over into your leadership and your leadership development as well.Spence Shelton — And I think that’s how we’re going to grow, just as how we’re going to grow the kingdom by reaching lost people, how we’re going to grow the leaders in the kingdom, the number of leaders we have, it’s going to be developing them in our church. And I just… I don’t know that we’re even thinking that way and asking the question, is the guy who’s a 27-year-old physical therapist in my church, has anybody ever bothered to tap him on the shoulder and ask, hey, man, could God be calling you into ministry?Rich Birch — That’s good.Spence Shelton — Because I watch you and no we’re not doing the shoulder tap, man. And so um right now, I just I love it. Just a little bit of the story of our church… Rich Birch — Yeah. Spence Shelton — …our executive both of our executive pastors and two of our three campus pastors were all in the marketplace. One of our executive pastors was in banking. Another was a plant manager. We have two campus pastors who were both, I use physical therapists because they’re both former physical therapists. And you know, we’re sitting there having these conversations.Spence Shelton — You know what makes a great pastor is someone who has sat with person after person told them, hey, you should do this to feel better. And then they come back to you later and say, hey, I don’t feel better. And you say, did you do the thing I told you to do? And they say, no. And you say, well, you should have done that.Rich Birch — Right.Spence Shelton — And a patient, long suffering physical therapist turns out to be a great pastor, but somebody’s got to tap him on the shoulder and say, hey, man, I think God might be calling you into this. And so my my encouragement to anybody who’s a in ministry right now, kind of in that leadership development role, and you’re like, I don’t know, we’re gonna have to go outside to find some talent.Spence Shelton — I promise it’s going to take longer the way I’m talking about. But if you will seek the Lord, ask the Lord, and then start shoulder tapping and giving some opportunities for guys to lead, for men and women to lead there in your ministry and whatever the leadership roles might be, I think there might be more people in there than you realize. Because I believe that God provides for the church that he’s growing, that he’s called you to lead. Rich Birch — Yeah, I love that. Spence Shelton — And so I think the same thing is true with leaders as well.Rich Birch — Yeah, it’s so good. I love that. And I you know I love the even the vision for, hey, maybe there are some people within your church. We see this statistically, like even if you’re thinking about campus pastors, 9 out of 10 campus pastors are coming from within, which means we’ve got which means we’ve got to get good at this. Spence Shelton — Yeah.Rich Birch — And yeah, if your church has any kind of unique slant on the world, you’ve got to find people who have come up through that culture to ah to ultimately push that culture forward.Spence Shelton — Yes.Rich Birch — I love this this marketplace to ministry transition. How do we help people do that? That’s a real key transition. I would imagine on your end, you’ve got to help connect the dots for somebody. I love that there, how you even with the physical therapist, we’re able to connect the dots between, hey, here’s what you’re doing. Let me talk to you about the kind of role that we’re thinking about. How do you make that transition? Have you developed anything to try to help make that, you know could you know, connect those dots for people as you’re having those conversations?Spence Shelton — Yeah, I would say we don’t I don’t have a conversational map that I’m doing.Rich Birch — Sure.Spence Shelton — I don’t have that yet, though I think that’s just like many things, trial and error, and I’m still in the still cooking to formulate that map.Spence Shelton — But what I do have that we’ve worked through with our staff is something we call, it’s interesting, we actually call it the lead pastor profile.Rich Birch — Okay.Spence Shelton — The reason we call the lead pastor profile is because everything we’re doing, we’re building towards, developing towards planting more churches. And you’ve got to have a lead pastor to go and plant those churches. Rich Birch — Yeah. Spence Shelton — Now, the thing is actually when you when you look at it, it’s actually just a ministry leader profile, you know? Rich Birch — Right. Spence Shelton — Everything on there, but there may be a couple of things that are nuanced specifically to lead pastors, but it’s kind of like when you go into Timothy and you see what are the characteristics of an elder? Well, every single one of them is just a good Christian, except then you got this thing able to teach. You know what mean? Rich Birch — Right, right.Spence Shelton — So it’s a similar thing is like most of our profile, nine points on it and divided into three buckets, it’s mostly just what makes a good ministry leader. Well, then what I’m doing is what we are doing is we’re taking that and we’re training our whole staff through it and saying, Hey, here’s what we are looking for. Here’s what we’re trying to develop. And then here’s everything. So I do a training monthly with our staff on these points. Rich Birch — Right.Spence Shelton — And then I do it and I say, yeah, I have to verbally say it, never underestimate or never overestimate, I guess just that it’s going to happen without you saying it, always say what you want them to do, you know, just good old preaching point there. But man, Hey, I want you guys to take what I’m teaching you and go teach your volunteers. Rich Birch — That’s good. Spence Shelton — And go teach those that. And I want you to teach your leaders. And then I want your leaders to be able to teach them. And if it doesn’t work, I want you to come back and tell me so that we can get better at making this a multiplying thing. Right. Spence Shelton — So we just got to have a lot of checks and balances along the way. So we’re hopefully training. And it took a while to pull this together in a way that it makes sense to all of us. And now that we’re kind of running it, we can see the value in it of, hey, this is what we’re looking for. And now you can go shoulder tap and say, I know what I’m looking…So the the kids director can go, I know what I’m looking for.Spence Shelton — Now I can go and start shoulder tapping because I know what, you know what i mean? Now that know what I’m looking for, it’s…Rich Birch — Totally.Spence Shelton — Those are probably the the beginnings of a conversational map to to walk somebody through, but at least you know what you’re looking for. You can start to have those conversations. And if you have the relational equity, those conversations can really go somewhere.Rich Birch — Love it. I’d love to double click on that. I love this lead pastor profile idea. I think working to get the clarity I think is in is a great, you know, a great task for all of us to wrestle through.Rich Birch — We think, well, we want to develop our leaders, but then we don’t have any definition of what do we mean by a leader, like at its more most basic level.Spence Shelton — Yes.Rich Birch — So we don’t need to get into all nine, but are there a couple of them that are either counterintuitive or maybe we miss, or you find yourself coming back to time and again, that are like real dividing points or like ones that kind of stick out to you in the in the framework?Spence Shelton — Yeah. All right. So obviously all nine matter to me or I wouldn’t have had them all. So I’ll do my best.Rich Birch — Yes, exactly.Spence Shelton — I’ll do my best.Spence Shelton — So we divide him into three buckets, character, competency, and calling. So each one of those has three points under it. And I’ll click on one of them, maybe in each, and you can tell me if you want to go further.Rich Birch — Sure. That’s great. Yeah.Spence Shelton — But in character, the one that I find to be the most overlooked is self-awareness.Rich Birch — So true.Spence Shelton — I find that to be the most overlooked characteristic of a good leader is one who is aware of their strengths and aware of their weaknesses and is settled in their identity in Christ so that their weaknesses don’t crush them.Rich Birch — That’s good.Spence Shelton — You know, they can instead build a team around them, you know. Because the classic thing with a lead pastor is to say, well, I’m good at vision and bad at details. It’s like, OK, man, maybe we need to do something a little more thoughtful than that, because that resonates with me for sure… Rich Birch — Right. Right. Spence Shelton — …that you ask the guys around me for sure. But what what? what about are you better at vision casting or vision creating? And what are the evidences of that in your ministry? You know, so for me, I’m better at vision casting than I am vision creating. That’s probably my my 9 to 10 is actually sharing it.Spence Shelton — But I got to have some help and and love having help in the other side of that. And so there are plenty of other things like that. But when we have someone who’s not aware of their strengths and weaknesses, usually that’s that’s like a stop point for us.Spence Shelton — And so if we’ve got someone we think, oh, man, they they come across because here’s the thing. Unless you have these points, somebody will can make a great first impression and a lot of long term damage in your ministry. Rich Birch — That’s good. Spence Shelton — Not because they have bad intent, but because they’re not aware of what they’re… And you give them a lot of responsibility and maybe they’re not actually, they don’t have the eyes of Jethro in Exodus 18 to know how to build out a ministry.Rich Birch — Right. Right. Spence Shelton — You just thought they did because they gave off a good first impression.Rich Birch — Right. Spence Shelton — So that’s my biggest biggest one in that character, or not biggest one, but that’s a key one, I think, in the character is…Rich Birch — Right. A key one that sticks out.Spence Shelton — …man, are you self-aware? And so whatever we use, we love different, whether it’s the personality assessments, leadership assessments, but we also, and we just, we wanna develop from within because that gives us the ability to see.Rich Birch — Right.Spence Shelton — We can see what their strengths and weaknesses are. Rich Birch — Yes. Spence Shelton — Because we’ve had people eyes on them for months and years and you can’t really trade that out.Rich Birch — Yeah, and character takes a long time of up-close observation to really get a sense of where where people are at.Spence Shelton — Yes.Rich Birch — Because, and you know we know this is, this is you know we don’t have to think long. There are lots of examples of leaders who—and all look great—but there’s like a 2% you know, problem in their character that ended up showing itself in the most kind of ugly of ways.Spence Shelton — That’s right.Rich Birch — So I know all nine are super important. I get that. appreciate that. But we can’t be here for three hours. So competency, what was what’s one in there that that that tends to say we’re just trying to whet people’s appetite to kind of help them think through it a little bit.Spence Shelton — Yeah, man. For competency, the one that I would, the the one that, yeah, that spend a lot of time on when it comes to competency, what I would say is the multiplier. Rich Birch — That’s good.Spence Shelton — I want someone who is a proven multiplier before I move them into an increased um level of leadership and responsibility. I want to see it, man. And if we, otherwise, again, what I’m trying to weed out is someone that talks a good game or someone that just has a, there may be someone I think a lot of leaders that might actually be really strong in the gift of hospitality have found themselves in the arena of administration simply because they are, like generous, kind, easy to talk to people. And they’re great. You want to be around them. They’re probably encouragers, that kind of thing. Spence Shelton — And it’s like, wait a minute, have they multiplied? Do they actually understand what it takes…Rich Birch — That’s good.Spence Shelton — …to take someone from leading self, to use the old Ramsharan leadership pipeline, from leading self to leading others, have they actually, do they understand what it takes? Rich Birch — Yeah, that’s good.Spence Shelton — And do they understand what it takes to go from leading others to leading leaders? And have they, let’s let’s just look behind them. If you wanna go the the Bible route, you just take 2 Timothy 2:2…are they able to train up faithful teachers who can then teach others also? That to me is, especially when we get into who’s going to go out and plant a church…Rich Birch — Right.Spence Shelton — …and I’m not sending somebody out that’s not a builder. Another way to say multiplier would be builder, but I say multiplier because I’m specifically thinking around leadership and leading those who will lead others. Rich Birch — Yeah. Spence Shelton — If we want to build a movement of churches for the sake of the glory of God, and we want these churches to go and continue to build, there’s got to be multipliers leading this. Rich Birch — Yes. Yes.Spence Shelton — You know, gotta be. So when it comes to competency, that’s like a a deal breaker for me for who we’re going to send out the plant. And so I want to develop it starting in our student ministry, you know, and working our way through or starting in our college ministry, working our way through.Spence Shelton — I don’t want to wait until we just try to, otherwise we’ve got to go pluck out from the, you know, ether, somebody that we hope is a multiplier. And I don’t want to do that. I’d rather raise them up.Rich Birch — Well, and we see this temptation in churches when they get to, say, 1,000, and then they’re they’re growing you know they’re they’re eyeing 2,000 beyond. There’s a there’s a dangerous temptation that can come in, particularly for our staff teams. I know we’re talking broader than just staff teams, but particularly for our staff teams, we can ah we can forget the Ephesians 4 call to equip others to to ministry.Spence Shelton — That’s right.Rich Birch — And we just become doers. We hire doers rather than multipliers… Spence Shelton —Yes. Rich Birch — …because they get stuff done. And we’re like, oh man, they’re solving problems for us. But that is can then erode a church long-term very quickly. I’ve seen that so many times where they end up with, you know they slow down.Spence Shelton — That’s right.Rich Birch — They’re like, why are we slowing down? It’s well, because you’ve stopped multiplying. You’ve stopped doing what we’re called to do. I love that. Rich Birch — So then the third category calling anything in there that would, that steps that kind of is similar. Again, I know all nine are are really important, but, but any of those that that kind of jumped to the, to the front.Spence Shelton — Maybe I’ll give you one that has we’ve learned through some some painful trial and error, Rich Birch — Okay. Sure. Spence Shelton — And that is, we call it alignment. And what we mean by that is: as you develop a leader, when it comes to their long-term place in your church and leading maybe even beyond your, so in our case, beyond our church to going in and leading ah a church that’s going be a part of the family of our churches, it’s really important that not only are they theologically aligned, but they are philosophically aligned in terms of how we do church.Spence Shelton — So no shade on any other way of doing church. What I’m saying is so that we can continue to be united in one voice internally, so at Mercy Church with how we do discipleship, and then so that we can be united in how we can help and train one another as a family of churches, we got to be aligned not just around our theological convictions, but on our ministry convictions, how we do it.Spence Shelton — I would say to those of you out there who have a promising young leader, but it seems like they just don’t want to do, they want to change your church into their, a different kind of church.Rich Birch — Right.Spence Shelton — The best thing you can do is set them free. Rich Birch — Yes. Spence Shelton — And I don’t mean like, that’s not like code for firing, by the way, that genuinely that genuinely is like, help them find a way, an off ramp to whatever it is they sense God calling them to. If they’re not, you know, they might be just, they need more season of development. That might be all it is. But when it comes to appointing people towards and, and, setting people into key leadership. So take an elder role at Mercy Church, community group coaching, shepherding, that kind of thing where you’re in now, you are shaping, not just participating. You’re really shaping the DNA of our church. Spence Shelton — We’ve got to be aligned with what we’re doing. It also makes for a much more enjoyable experience. I’m not saying we’re not thinking. We’re always thinking over here about how to do it better and trying to be faithful and everything else.Rich Birch — Yes.Spence Shelton — Of course. But I’m saying, man, you want to feel like you’re locked arm and arm going together. And when conviction comes that goes a separate way, it’s like best to separate. So we’ve we’ve learned that the kind of the hard way. I think a lot of lessons are hard.Rich Birch — Sure.Spence Shelton — And so I offer this as our missions pastor likes to say, we have paid some dumb tax. Rich Birch — True. Spence Shelton — And so wherever we can save another person from paying it, we’d love to do that for you.Rich Birch — Yeah, that’s so good. I love that. and And yeah, I think alignment’s a massive deal and and calling it out. And I would think being in a church that is um so focused on sending, you’re you’re hoping that that’s a big part of what’s kind of the outcome of the church. Spence Shelton — That’s right.Rich Birch — That, you know, having an an outlet for that sending in a way that makes the most sense. Hey, here’s a leader who seems to have the skills and, you know, and and and then you can celebrate, hey, they seem to have some unique ideas that they want to go and see happen somewhere else.Spence Shelton — Yes. Yes.Rich Birch — That’s fantastic. Rather than having to like fight that. I think that’s cool. That’s really good. Rich Birch — So, you know, help me ah pick this apart a little bit. So one of the interesting things I find in this whole kind of leadership pathway, leadership pipeline conversation is there are, so I think we all can be influential. I think in one way we are all called to be leaders, but not necessarily everyone that we’re discipling is necessarily going to be a lead pastor. They’re not all necessarily going to. Rich Birch — Now I understand we’re kind of discipling towards that, but help me understand what are you doing internally with folks that you’re like, I just, they’re just not going to make that jump eventually. How do you have that conversation? I know that’s a hard conversation. What does that look like?Spence Shelton — Right.Rich Birch — Or am I just looking at this wrong? Which could be the case too. Happy to be wrong. Help help me think through that.Spence Shelton — Well, we’ve got to have some freedom that comes from distinctive categories. This will help us a great deal. Helps me in pastoring a great deal. First, we got to have the category of spiritual gift. The Lord has given everyone a spiritual gift… Rich Birch — Yep. Spence Shelton — …and we got to celebrate that. Rich Birch — Right. Spence Shelton — And we got to be careful not to turn the leadership pipeline into, or that concept of leader development, turn that into a a grading of gifts. Like you’ve got a, you know, you’re, you need to graduate from one gift. Maybe that’s a better way to say it. Graduate from one gift to the other, to the other. It’s like, no, no, no. Rich Birch — That’s good. Spence Shelton — What we’re not talking about is graduating out of your gift set. God has given you that. And the best thing for you and for his church is to get really good at how he’s given you or how he’s gifted you. Rich Birch — That’s good. Spence Shelton — And when we when we conflate those, when we only talk about just you know getting better, it’s like, man, we got to watch out. Instead, let’s say, how can we get better at the giftings God has given us so that we can steward it really well?Spence Shelton — And then we find ourselves able to say, and when it comes to leadership, and here we are as church leaders, we’re now inside of a category. We’re inside of a category of leadership. And how can we develop leaders who are gifted by God as leaders? Well, the good news is maybe take some pressure off everybody, you’re not really going to know if someone’s gifted in leadership until you let them try to lead. Rich Birch — True, yes. Spence Shelton — So you got to let them try to lead, man. And then as they lead and you give them a couple of swings, two, three, four, whatever, and you coach them along the way, hey, here’s what we’re looking for in a leader. Here’s how you can grow from leading self leading others. Take the pressure off. Spence Shelton — This might be how God has gifted you, we don’t know. But the best way to see, based on what we’ve seen so far, is to let you try. So here’s a community group, and here are the things you’re gonna need to be a good community group leader, and we’re gonna coach you and support you.Spence Shelton — Let’s go. And let’s see what the next six months brings, right? Rich Birch — Yep, that’s great.Spence Shelton — Hey, this went really well. We wanna take you out of this community group leader seat and put you in as a coach. Now there’s a different set of skills you’re gonna need to be a coach, but man, it seems like your people responded to your leadership. What do we think about you as a coach? Spence Shelton — Each set of the way, man, do I have the, the way I always say it to our staff—they’re sick of me saying this, so it’s great to talk to you because…Rich Birch — I’m not sick of it. Yeah.Spence Shelton — …[inaudible] fresh person to say this too—is that clarity in expectations creates confidence in execution. The clearer you are with what you are asking someone to do, the more confident they will be in doing it. Another way to say it.Rich Birch — That’s good.Spence Shelton — So that’s just my my big thing there on on this. Let’s get over now that we’re, and what we’re going to find is some people are going to lead a community group faithfully for 20, 30 years at our church. Rich Birch — Right, right.Spence Shelton — Praise the Lord. What we can’t see is that we can’t see that a failure. Rich Birch — No. Spence Shelton — Others are going to lead for six months to two years. And we’re going to man, this, this person’s a leader and, and we’re seeing it. They already, they’re planting. Spence Shelton — I had someone come up to me the other day at church and tell me, Hey man, I just wanted to tell you, we’re so excited. You challenged me about five years ago to lead a community group. And we just planted our fifth group. And I went, like I went, I was, my mind went so far. I didn’t even have time to be, to be celebrating. Rich Birch — Right. Spence Shelton — Cause I was like, does anybody know that there’s a guy out here… Rich Birch — Right. Yes. Who’s just cranking out groups. Spence Shelton — …with the with an apostolic gift just planting groups every year what’s going so so. Rich Birch — Yes. Yeah. Yeah. Yeah. That’s great. Spence Shelton — I was like hey, but so that that was a a great moment the Lord gave me just of, hey this is this is going really well and there will be some that will lead a group and things won’t go well and they won’t go well and you’ll keep trying different efforts. So to my community group leader friends, hey once someone tries and tries and tries, would you maybe tap their shoulder and say hey can we try something else. You know. Rich Birch — Good. Spence Shelton — Like you are not a failure. We just got to find the gift that God has given you. Rich Birch — Right. Spence Shelton — And the more that’s in the conversation of your church, that’s in the atmosphere of your church—We’re just looking for the way God’s gifted you. And we’re going to try and make you the best you can be at how he’s gifted you.—The easier the conversation is going to be of, I don’t think it’s community group leadership. I don’t think it’s this. I don’t think it’s that. Because we have a culture of, Hey, trial and error, you know?Rich Birch — Right. That’s good. That’s so good. I love that. And I love this has been a great kind of look under the hood of this whole multiplication issue. I think a lot of church leaders out there have the desire. They have the like, oh, we, man, I would love to be a church that has that kind of vision. But then what are the mechanics underneath of that? How do we actually do that?Spence Shelton — Yeah. Yeah.Rich Birch — Super helpful. Pivoting in this totally, well, it’s a related but different direction. Your church has been involved in a number of revitalizations. You talked about on the you know top end, you had a you know facility given to you. And I this is like a super inappropriate, with just a few minutes left, curveball to throw in there. What’s your shorthand thing for us to learn about that? Rich Birch — You know that we’ve seen, particularly from a multisite point of view, this area has grown hugely in the last 10 years. It was 10 years ago, 5% of new campuses came about because multiple revitalization. The most recent study showed showed it was 40%. Spence Shelton — Wow.Rich Birch — I suspect Warren’s out talking right now. I suspect the number is even going to be higher. What are the few things that you’ve learned about, you know, those kinds of conversations and relationships that have allowed it to work for you? And it’s been helpful. It’s been a joy giving part of the story at Mercy. Talk us through that.Spence Shelton — I’m going to do so using another point in my lead pastor profile just to have some fun here.Rich Birch — Love it.Spence Shelton — So one of the things we look for in a leader is confident humility. Rich Birch — That’s good. Spence Shelton — Confident in terms of, man, they believe God has called them to something. You know there’s a there’s a settled confidence there. But the humility to be patient and wait on the Lord and see what he has and to consider others, Philippians, to consider others more important than yourself. Right? This matters a great deal when it comes to this area of revitalization.Spence Shelton — What a leader needs to know that there’s this temptation, and I had it, of why would you not give us your church building? Rich Birch — Right.Spence Shelton — Things are going well for us and struggling for you. Why would you not do that?Rich Birch — Yes.Spence Shelton — Well, it’s that there’s an unintentional, I’m call it a holy ambition that may be a little bit creating some blinders. Okay, let’s use the positive side of that. The the negative side would be there’s some arrogance there. Rich Birch — Right. Yeah.Spence Shelton — But the positive would be holy ambition. But what you gotta know is, man, often these facilities that are housing these churches, the ones that are remaining, a lot of the times that there are people that have been there for decades and for a generation or two, and they put blood, sweat, tears and money into where they are. Spence Shelton — And we had to take the the approach of slow by my standards, still fast by theirs… Rich Birch — Yes. Spence Shelton — …slow by my standards. Conversation after conversation that began with , in both of these cases, began with the church that needed revitalizing coming to us. We didn’t go to them. Spence Shelton — They came to us and said, hey, we’re trying to figure out what’s next. We hear you might be able to help. How can you help?Spence Shelton — And we had to present them. We had to consider that a stewardship process. of a kingdom resource, not a how do we advance the name of Mercy Church.Spence Shelton — And so it’s happened two different ways for us right now. One is a campus of our church. You know, we it took months and months, but they wound up voting to hand us the keys, join our congregation, and we held their hand the whole way through. And it was a really beautiful thing. And it’s where our broadcast campus meets today. Spence Shelton — There’s another one that happened where we decided the best thing was for that to be a, because of the neighborhood it’s in in East Charlotte, to be a church planting incubator where we could plant a church every 18 months to two years. Rich Birch — Love it. Yeah.Spence Shelton — And so that’s what we’re doing right now. And there’s a church plant in there. And that was the best stewardship of that building. And after the conversation with their people, but we kind of, we basically put that with them and say, I had a lot of lunches and dinners that I just I’m going to tell you, if you want to get into this game and you want to do it in a way that’s going to honor the Lord, you got to be ready for the patient handholding that I would say is honoring a legacy of faith these folks are are taking.Spence Shelton — In many ways, the greatest step of faith taken in our church has been a couple of other churches handing us keys…Rich Birch — Yeah, 100%.Spence Shelton — …to something that the Lord was an outpost for the gospel at one point. And now in both of those cases for us and in the others that we’re starting to see that are still too soon for me to talk about, it’s just, man, now there’s an outpouring happening in both of those places. And you got like, man, there’s a couple of folks that are from the original congregation of what is now our, where our broadcast campus now is. And they, they, they’re greeters. They have tears more mornings than they don’t on Sunday mornings.Rich Birch — Yeah, yeah, for sure.Spence Shelton — It’s beautiful, man. Rich Birch — Yeah, it is.Spence Shelton — I mean, they’re all in. They are way bigger fans of Mercy Church than I am. And I’m the lead pastor of the church. Rich Birch — Yes, right. Spence Shelton — You know what I mean? And they just, they are like, they love, because what they actually are is they, they took a step of faith, a huge step of faith, trusting the Lord. And the Lord has blessed them and renewed a a vigor in their ministry. Spence Shelton — I saw a couple of them, mid 80s, and it was about 11 a.m. weekday and they were leaving the church building. I was pulling up. I just put windows down. Hey, what are you doing? It’s like, well, we just finished prayer walk in the neighborhood a couple of times…Rich Birch — Wow.Spence Shelton — …and it’s getting too hot now. We’re going inside. And I’m like, these are the people I want, man. Rich Birch — Yes.Spence Shelton — Like just they’re retired and they have chosen they’re going their day prayer walking. It’s like, well, you told us to pray. And I’m like, yeah. Yeah. Rich Birch — That’s amazing. Yeah, that’s incredible.Spence Shelton — Thank you for leading the way you have the time and this is what you do with it. So, man, there are if if all we see is just, Oh a facility to advance my cause. And we miss what has actually been the sweetest thing for our church has been our very young church got some grandparents.Rich Birch — Yes. Spence Shelton — It’s beautiful, man. Beautiful. Rich Birch — Yeah. Yeah. That’s so good. Spence Shelton — So that’s just a little bit I learned from that ah from those couple of runs we’ve had. Rich Birch — Yeah, that’s amazing. And yeah, that underlines, so it’s definitely the majority—I think it’s three quarters, if I remember the most recent statistics on these sort of revitalizations—are coming about because the joining church is coming to the lead church, not the other way around.Spence Shelton — That’s right. That’s right.Rich Birch — And so you have to position yourself in a way that that confident humility is a great kind of benchmark to be thinking about, friends. How do we you know hold ourselves in a way that we’re here to serve other people, we’re here to care for them in a way. And I say to church leaders when we talk about these things, I’m like, you have to handle these relationships well enough so that then they’ll tell other people, hey, wow, we loved working with Mercy Hill.Rich Birch — Man, there we love working with Mercy Church. Well, we love working with mercy Mercy Hill too. Spence Shelton — Sure. That’s right.Rich Birch — We love working with Mercy Hill, Mercy Church, too many Mercies. Wow, what a great you know interaction that was been. And wow, that’s a life-giving thing.Spence Shelton — That’s right.Rich Birch — And I loved that. That is what ultimately we’ll see more of these happen long-term. Not because you did some, you know, so too many people have these like, You know, like they use merger and takeover and like this kind of aggressive language, which just doesn’t help in the conversation at all.Spence Shelton — No, it doesn’t help at all. Yeah, that’s exactly right. You got to have that humility. And if somebody’s sitting there wondering, okay, man, well, how do I how do I wait on that without, I want to go get it. I want to get started. I’d say two or three things to do if you’re looking and you believe your church might be a good church candidate for helping another church revitalize, whether that’s being a campus of yours or or something to that effect. Spence Shelton — I’d say first start praying and asking the Lord. And that if you’re not fasting, then you’re not praying yet. Rich Birch — Yeah. Love it.Spence Shelton — So I don’t want to hear that you’re praying if you’re not fasting on this. Secondly, man, this is why networks matter. So I would say whatever your denomination or stream is, whatever, however you’re kind of tied to other churches in your community. If you have no relationships with other churches in your community, and then you want to go and be the revitalizer, that’s going to come, that’s probably going to come across a certain way.Spence Shelton — So There are a number of—years and years and years—of meetings that I’ve gone to, conversations I’ve sat in, in order to genuinely try to say, how can we help? How can we help?Rich Birch — Yeah.Spence Shelton — Whether that’s sending a worship leader to go serve at a church one Sunday morning or helping one church with a a building fundy kind of like thing they got to upfit or something, you know.Spence Shelton — Basically, I’m trying to be a team player in the Kingdom here in our little corner of the world in Charlotte. And then that’s going to earn the respect and the trust for when those conversations can happen in the relational network, those conversations can happen. Spence Shelton — So I am one who just in my nature, I do not like meetings. I don’t like them. Rich Birch — Sure.Spence Shelton — They feel like necessary evils. Rich Birch — Love it.Spence Shelton — And I’m trying to get better about that. But they are actually what I do love and value is trust and relationships. And again, confident humility. I have to be more about the Kingdom than I do about my own church. And a rising tide lifts all boats, you know.Rich Birch — Yeah, it’s so good. Well, this been a great conversation today, Spence. I really appreciate that. And thanks for going with me there with that little you know detour at the end.Spence Shelton — Of course, the joy.Rich Birch — That was fantastic. But anything you’d love to share just as we wrap up today’s conversation?Spence Shelton — I would say too—I don’t know what’s spurring this, but something that has helped me so much in my ministry to the leader out there, especially the young leader, but maybe to all of them—you will always overestimate what you can do in five years and you will drastically underestimate what you can do in 20. Rich Birch — That’s so true.Spence Shelton — And there maybe a way to say it is you will overestimate what God will do and underestimate what God will do. So this is a little plug of me saying, stay the course, stay the course where you’re at. The Lord is going to do a work in honor of your faithfulness to the work. Stay in it, man. Stay in it.Rich Birch — Yeah, that’s so good. Just to add to that, I’m doing some research on working on a book on churches that break the 2000 barrier. And one of the interesting telltale signs are senior leaders and senior leadership teams that have been in the same church or the same community for 20 plus years. Spence Shelton — There you go. Rich Birch — So going into their third decade, it just takes time. Spence Shelton — There it is. Rich Birch — You know, it’s not it’s it’s not you know’s not overnight. Rich Birch — Well, I really appreciate this. This has been a great a great conversation. Spence, if people want to track with you or with the church, where do we want to send them online? Spence Shelton — Yeah, man, you can go to probably social media is the the day and age that we live in. So @SpenceShelton is my social media handle. It’s Instagram and and those kind of things. Our church is just @MercyCharlotte. And you can go on our website, mercycharlotte.com. Find everything you need there. Spence Shelton — We’re launching a podcast. Praise God, let’s go. You can find all of it, though, just through our Instagram handles, which is usually a good first landing point for folks. And however we can help, man, feel free to reach out. Nothing super shiny about what we got going on over here, but we love being in the game for the Kingdom. However we can help.Rich Birch — All the best as you serve in Charlotte. Thanks so much, man.Spence Shelton — Thank you, brother.

The Inference Engineering Masterclass — Philip Kiely & Ali Taha, Baseten

Latent Space: The AI Engineer Podcast â€” CodeGen, Agents, Computer Vision, Data Science, AI UX and all things Software 3.0

Play Episode Listen Later Aug 3, 2026 101:29

Watch the full episode on YouTube:We first covered Baseten last year when DeepSeek mania was at peak hype. Now they have raised a monster $13B round and become one of the new cohort of AI Infra decacorns that are (with Nvidia, Intel, and the semis complex) chief beneficiaries of the Inference Inflection. We return to Baseten at the peak of the 2026 edition of Open Weights debate. Ali has published a viral breakdown of Kimi K3:And since you last saw him, Philip has spoken at AI Engineer and written the definitive book on Inference Engineering spotted all over SF:Three years ago, inference engineering barely existed as a category.Today, it is one of the most critical disciplines in AI. Inference engineering inherently tackles a different question than standard model training: “How do you turn those weights from training into a product that is fast, reliable, and affordable at scale?” Focusing on these creates an entirely new optimization problem.In one recent GLM-5.2 experiment, quantizing more of the model actually preserved its benchmark quality while increasing throughput by 20%, because the errors introduced in different layers could cancel each other out.Inference is no longer just the final step after training. It is becoming its own engineering discipline, with its own research problems, infrastructure, and increasingly specialized roles.In this episode, Baseten's Philip Kiely and Ali Taha join swyx and Vibhu to explain what actually happens after a new open model is released and what it takes to turn “we generated a token” into a fast, reliable, production-ready API.We go deep on cache-aware routing, disaggregated prefill and decode, quantization, speculative decoding, KV-cache movement, model parallelism, GPU kernels, and the race to make frontier models up to 10× faster. Philip and Ali explain why inference optimizations can still produce gains of 20%, 100%, or even 200%; how quantization errors can cancel one another out; why identical weights can behave differently across clusters; and how Baseten grafted a Kimi vision encoder onto GLM-5.2 without changing the underlying language model.The conversation then expands beyond LLMs into NVIDIA Dynamo, mega kernels, Rubin, AI-specific chips, local inference, video generation, diffusion versus autoregressive models, and the enormous compute barrier to generating coherent long-form video. Finally, we explore the convergence of training and inference, continual learning through persistent KV cache, and the emerging loop where models help optimize the infrastructure that runs them.We discuss:* What happens when a 200,000-token request enters an inference system* Cache-aware routing and reusing previously computed KV cache* Why prefill and decode are increasingly handled by different GPUs* When dedicated deployments become cheaper and more reliable than shared APIs* How speculative decoding uses a smaller model to accelerate a larger one* Tool calling, structured outputs, and what LLMs actually do* What it takes to support a new open model on day zero* Grafting Kimi's vision encoder onto GLM-5.2* Retrofitting inefficient model layers with components from other architectures* Why models sometimes collapse into repeating the same token* How hardware, kernels, and race conditions create nondeterministic failures* Preserving model fidelity while making inference faster* How quantization errors can cancel each other out* Why inference optimizations still deliver gains of 20%, 100%, and 200%* How optimized serving can make a model up to 10× faster* NVIDIA Dynamo, KV-aware routing, and distributed model serving* Speculative decoding the speculative decoder* Why local AI is about making models less dumb while data-center AI is about making them less slow* Tensor, expert, and pipeline parallelism across GPUs* Hardware-aware model design, auto-tuning, and the case against mega kernels* Rubin and why inference is becoming a systems problem* Whether modern GPUs are evolving into programmable AI ASICs* Why enormous models like Kimi K3 require GB300-class hardware* Why open-source video generation still trails Veo, Kling, and other closed models* The quadratic attention bottleneck behind long-form AI video* Autoregressive video, real-time generation, and compounding quality drift* Why future video systems may combine autoregressive and diffusion architectures* Training for inference and inference for training* Continuous post-training, deployment, evaluation, and improvement loops* How GLM-5.2 helped optimize the kernels serving GLM-5.2 itself* Why faster networking could unlock dramatically faster decoding* Continual learning, KV-cache compaction, and persistent model memoryShow Notes* How to build a day-0 API for Kimi K3* 22580: From GPT2 to Kimi3, ExplainedPhilip Kiely* LinkedIn: https://www.linkedin.com/in/philipkiely* X: https://x.com/philipkiely* Inference Engineering: https://www.baseten.co/inference-engineering/Ali Taha* LinkedIn: https://www.linkedin.com/in/aliestaha/* X: https://x.com/waterloointernTimestamps00:00:00 Introduction and the 200K-Token Prompt00:03:18 Dedicated Deployments, Speculative Decoding, and Tool Calling00:11:26 Launching Production-Ready Open Models00:19:06 Model Retrofits, Failure Modes, and Nondeterminism00:28:22 Quantization and Canceling Errors00:32:15 The Race to 10× Faster Inference00:40:48 Dynamo, Speculation, and Local vs. Data-Center AI00:50:18 Model Parallelism, Auto-Tuning, and Mega Kernels01:00:55 Rubin, GPUs vs. ASICs, and Custom AI Chips01:10:03 Giant Models and the Limits of GPU Memory01:12:42 AI Video, Quadratic Attention, and Autoregressive Generation01:21:47 Audio, Images, and Diffusion Models01:27:32 Training, Self-Optimizing Models, and Continual Learning01:40:06 Closing ThoughtsTranscriptIntroduction: Baseten, Waterloo Intern, and Inference EngineeringSwyx [00:00:00]: Okay, we're here in the studio with Philip, old friend from Inference Engineering, the book, as well as Baseten and everything that you've done, you and I have done before, as well as Ali. Welcome.Ali [00:00:15]: Pleasure to meet you.Swyx [00:00:15]: Waterloo intern.Ali [00:00:16]: Waterloo intern, always.Swyx [00:00:17]: When did you get “Waterloo intern” as a handle?Ali [00:00:19]: As a handle? Oh.Ali [00:00:20]: I think the rebranding happened mid-March. When I saw it was open, I was like, “I have to take it. Up for grabs.”Philip [00:00:26]: The problem is that Ali is really good at his job and is not gonna be an intern much longer.Philip [00:00:30]: So we have to figure out who's gonna get the handle.Ali [00:00:33]: Well, I'll pass the torch over to the next intern.Swyx [00:00:34]: Oh, okay. It can be, like, you just pass it to another Waterloo grad.Ali [00:00:37]: To another Waterloo intern. No, bruh.Philip [00:00:39]: Yeah.Ali [00:00:39]: Intern.Swyx [00:00:40]: Intern, yeah.Ali [00:00:40]: And no.Philip [00:00:41]: You gotta get an intern from Waterloo.Ali [00:00:42]: Yeah, I've gotta get an intern from Waterloo.Swyx [00:00:44]: Right.Ali [00:00:44]: But they have to follow the path.Swyx [00:00:45]: Oh, it could, but it could come from Baseten, so it's like whoever Baseten gets from Waterloo.Ali [00:00:48]: Right.Swyx [00:00:49]: Has the title of Waterloo.Ali [00:00:50]: It stays in the ecosystem.Philip [00:00:51]: Exactly.Ali [00:00:52]: Halfway through the internship, you either get it or you're out.Philip [00:00:55]: You should also do, like, a big graduation ceremony where you change the handle.Ali [00:00:59]: Just say it.Philip [00:00:59]: For everybody.Swyx [00:01:00]: You guys are good at ceremonies, clearly. We had a nice launch of the book, very successful. But before we get into all that, I wanna start off with a fun question for you. Okay, you're an expert inference engineer. What happens when I send a long query, say two hundred thousand tokens into Baseten's inference? What's the process of query through GPU model routing, balancing, all that? What is all the stuff that we don't think about?Long Context Requests, KV Cache, and Cache-Aware RoutingPhilip [00:01:26]: With a long query specifically, the first thing that I'm gonna ask is, “Have you sent me this query before, or at least part of it?” and I really hope you have, because it's gonna be a lot easier for me and a lot cheaper for you. So the first thing that we're gonna look at is some cache-aware routing, where we're going to see, we probably have a number of instances, a number of replicas up serving whatever model you're hitting. We want to send this one to something with, number one, available prefill workers, and number two, ideally some cached input already there so that we can skip prefill on at least part of these two hundred thousand tokens. If you're doing two hundred thousand tokens, it's probably coding or a multi-turn agent or something where you would expect to have that cached. If you don't, we're gonna have to send it to a prefill worker. We've at least on certain models disaggregated prefill and decode, so you're going to have one set of GPUs that's solely going to process the input, create the KV cache, and get you your first token, and then that's going to be passed over to a separate set of GPUs, which is going to run decode. We're going to iteratively make those tokens. We're probably going to have some speculator model in front of that. I'm going to assume that you're doing coding, and because of that, our speculator model, which assumes you're doing coding, is gonna have a high draft token acceptance rate. If I'm wrong and you're asking me to summarize every Harry Potter book, it's gonna be slower. And then we stream that output to you and account for it, charge you, a couple of pennies and say, “Hey, would you like to send another one?”Swyx [00:03:04]: Except Baseten doesn't charge by pennies.Philip [00:03:07]: Well, yeah, we charge. I'm assuming that we're talking about the public model APIs. If you are setting up a dedicated deployment, then yeah, it's not pennies.Public APIs vs. Dedicated DeploymentsSwyx [00:03:18]: Yeah, one of the key differentiators when I was talking with Baseten initially was that people who want very high volume just need to rent by the box, ‘cause then it's up to you to figure out how to saturate the box.Ali [00:03:31]: And more often than not, it's, like, way cheaper if you're pushing, like, millions of tokens per hour, if you just pay per hour instead of pay per token.Philip [00:03:37]: Yeah, they do. I think that we've increasingly seen a lot of demand for the pay per token APIs, just because everyone wants to try open models, and then once they find a use case that's really sticky, then they move over to dedicated.Swyx [00:03:51]: Is there a best practice on when it's time to swap over?Philip [00:03:54]: Couple reasons. Yeah, reliability, that's a big one, right?Ali [00:03:57]: Like, if they have a very specific use case, they want you to train something specifically for them, like they want their own spec dec, for instance, for their own traffic.Swyx [00:04:04]: Spec dec is speculative decoding.Speculative Decoding and Custom SpeculatorsAli [00:04:05]: Speculative decoding, yeah.Swyx [00:04:07]: You have to explain.Ali [00:04:07]: Sorry. Like, speculative decoding is like, if you have a huge model, right? And so the model is going to be generating one token at a time every single turn, every single forward pass. So we attach, like, this little, like, parasite, like this layer that goes on top of the model, and this model just has to predict. It does three very fast autoregressive forward passes, and it will predict, like, three certain tokens, and then you do one forward stage over the entire original model in order to see if those predictions were correct or not, and then you accept them or you reject them. Now, this draft model is traffic specific, so if you, like, Philip said, if you're summarizing Harry Potter books, I can train exclusively that draft model on Harry Potter books, and I can guarantee you that I'm gonna accept the three tokens every single time. And so with that case, I increase your decode speed. I wouldn't be able to provide this to you if you're a shared endpointSwyx [00:04:53]: YeahAli [00:04:53]: ‘cause I have no idea if you're doing Harry Potter, if you're doing coding, if you're doing English. We don't know. Also, there was a thing in the book that mentioned that if they really cared about a specific threshold, chapter four, I think. Do you remember that?Philip [00:05:06]: Yeah. The things that you can do is you can set a specific, like, batch sizing, a specific, like, parallelism strategy if you're trying to optimize for, like, throughput versus latency. You can. Maybe a NVFP4 quant doesn't pass your benchmarks and you wanna run a model at higher precision, you could do that. There's just a bunch of reasons why you might wanna have your own endpoint and the biggest one, of course, just being, like, you don't have to deal with someone else doing a hundred million tokens of benchmarking traffic at the endpoint when you happen to be trying to serve your users.Swyx [00:05:40]: Yeah. I think one thing that is. That is a classic journey. Like, it's people is asking the, what happens when you type Google into the browser. Tool calling, is that just, you're generating JSON or is there more complication beyond that?Tool Calling, JSON, and Structured OutputsAli [00:05:58]: Certain customers that we have, they have their own post-trained models, and so they demand a tool calling that's not just, like parse a file or go find the weather. It's something that's very specific and you have to do post-training on this. And if the post-training on the model is not good or if the quantization after the post-training to get the inference to be fast, the model will struggle reading the JSON file and reading the tool calling. But it doesn't require its own like sandbox. It's not like it's going to use that tool calling to like escape a sandbox or like it doesn't have to be contained. It can just be a normal dedicated deployment. The challenge with tool calling more and more seems to be that the companies want certain tool calling which is a very sensitive thing to train. And because you're dealing with all of the JSON outputs, if it doesn't like close the end of the request in a very certain manner, you end up with a model that did the tool calling and like the thinking and so as a result of that, it didn't see the result and just hallucinated the result as it decoded. That seems to be the most challenging thing with tool calling, not really the sandboxes model.Philip [00:06:56]: Yeah, that's a challenge on the training side and then on the inference side, there's work that you can do to scope the possible output. So we published this at this point close to two years ago, the solution to this problem which is you make a state machine and you use that to constrain the output to a specific format. So this is the structured output problem. If you remember backSwyx [00:07:27]: Yeah, the specific grammar is,Philip [00:07:29]: Yeah, exactlySwyx [00:07:30]: GML had this thing.Philip [00:07:31]: Yeah. So it's like the old-school “make sure this is only JSON”, return only JSON orSwyx [00:07:38]: YeahPhilip [00:07:38]: Grandma's gonna die type of prompts.Swyx [00:07:39]: Is it BNF grammar? At some point OpenAI had released a thing that was like, yeah, if you want to constrain your output, write BNF grammar, back as NOR.Philip [00:07:47]: In our inference system, it's just a specified output format. And you get the guarantee that your output's gonna be structured along that format. And so applying that to tool calls can like help cut down on. You can still call the wrong tool or call no tool. It doesn't solve the certainty problem but it at least solves the output structuring problemSwyx [00:08:10]: YeahPhilip [00:08:10]: Within tool calls.Swyx [00:08:12]: And MCP is just another form of tool, right.Philip [00:08:14]: Yeah, exactly.Swyx [00:08:15]: As far as there's no special thing there.Philip [00:08:16]: The thing I'm always like explaining to people is the LLM is not capable of doing anything. It's only capable of making suggestions of what to do and then if those suggestions are formatted in a certain way and applied to a system that knows what to do with them, then an action occurs.Swyx [00:08:32]: Yeah. Part of the fun stuff is, this is solved outside of tool calling too. Like in an agent loop if the output is not correct or you're right, like reasoning, tool calling was done in the reasoning trace, just be like, “Oh, I don't know what to do. Let me just try again.” And it might get there after a few tries. And on your point of training, sometimes this is harder in smaller models, so you don't have the same exact quality outputAli [00:08:56]: Right.Swyx [00:08:57]: When you just swap from a big model, right?Ali [00:08:59]: Yeah. I will say that, before, I think we need to go back to inference engineering proper.Ali [00:09:04]: But, I had expected that something would replace JSON because it's hard to stream JSON ‘cause JSON must be complete and you must have open and close brackets and everything. So it's hard to parse something or validate something while it's being streamed. So people invented all sorts of things that are like, I forget the name of some of these alternatives, but it's something like TOML, something like YAML. But JSON seems to be dominant still.Philip [00:09:30]: The JSON outputs aren't that long, right? Like you could have a long-- ‘cause tool calls also contain the arguments in them and perhaps for a certain tool you might pass like a very long argument. But my impression of the median tool call is that it's a relatively small number of tokens, right? So I would expect that speculators are generally fairly good at something as formatted as JSON. And so you would have like a pretty fast decode step there and that the streaming wouldn't be as valuable, but maybe I'm wrong about that.Ali [00:10:02]: I think you're also bounded by the software or that the model is gonna integrate with if the software is built with JSON for the tool calls or if the company that you'- if your customer says that this is how our software works and our tools are interfaced with JSON, you can ask them to like, change their software and say like, “Yeah, this is gonna be better for the model.” but like with the right training shouldn't be that much of a difference. Also more profitable if it outputs more tokens probably.Swyx [00:10:25]: Depends on your business model.Swyx [00:10:27]: It really depends. But I will say that, as a writer with like experience a lot with generated output, I do try to move from text to JSON text which is very long JSON, right? Like there's paragraphs in every field because I'm trying to structure it, right?Philip [00:10:44]: Right.Swyx [00:10:44]: I want you to first make factual statements, then make opinions then make bullet point summaries, have dates, have entity references have your sources for references, all these things. Anyway, so these are things that like I think people who really experiment with structural output have to really care about. But, let's, let's recurse up the stack a little bit. Before we started recording, you mentioned something really cool, which is that there's a lot of engineering that-- inference engineering that goes on when a new model provider releases a new model, right? So let's call it GLM-5.2, Kimi K3. I had previously assumed, especially if it's like, well, GLM 5 to 5.1 to GLM-5.2, like that you've supported them before. Is it that much work?What It Takes to Support a New Open ModelAli [00:11:26]: It's a lot of work.Swyx [00:11:28]: Yeah. Okay. So like, a lot of people, all you guys, right whenever a new model launch like, people rush to say like, “Oh, Hugging Face supports this, Fireworks supports this, Spacetime supports this,” and I'm like, “Yeah, of course we support it.” But what goes into that? What goes intoPhilip [00:11:40]: I think it's more than just support it too, right? It benefits the consumer a lot. Like I think it was with Kimi K2.5 or GLM-5.2 the latest, there was an inference war, right? X provider is at 90 tokens a second. The next day we're at 150. The nextSwyx [00:11:55]: I kinda kicked that off with the GLM-5.2.Swyx [00:11:58]: I wrote a Twitter article about. It got like half a million views,Ali [00:12:02]: Based on being numberSwyx [00:12:03]: YeahAli [00:12:04]: Or it's for something else.Swyx [00:12:05]: Yeah. Which,Ali [00:12:06]: Oh my GodSwyx [00:12:07]: Which then got everyone really excited about, hey, how can we, bend tracks a little bit further and,Philip [00:12:14]: There's a difference between support the model, as in I can make a token out of this model, and support a model, as in I have a production-ready API from this model.Philip [00:12:26]: Getting to the point of I can make a token out of this model is not that hard because generally the, open source inference engines, vLLM, SGLang of the world oftentimes even receive weights ahead of time, maintainers do, or the people making the model merge PRs to ensure support. So you generally can, just get it working on the standard open source stack without too much pain in most cases. The challenge is, every inference company is gonna have own proprietary stack. Some open source components, some in-house stuff. And for any arbitrary model, there's going to be some new stuff. Sometimes you get lucky, like K, two five to two six was, like, pretty similar.Quantization, Speculators, and Production ReadinessAli [00:13:16]: Yeah. It was pure continued post-trainingPhilip [00:13:18]: YeahAli [00:13:18]: If I remember correctly.Philip [00:13:19]: Even in those cases, there's still stuff you have to do. You have to redo the quantization work. You're taking the model from. Generally, these models are not released in NVFP4, and we want them to be in NVFP4 for maximum Blackwell compatibility. So we have to perform that quantization, and, calibrate the quantization to make sure that we're not causing any regression in the model's intelligence. And then we also have to train the speculator, as we've talked about. Generally, we have. We have ZDR, zero data retention on our model APIs, so we don't know exactly the traffic that people are sending us, but we know what's popular. We know that coding use cases are popular. We know that agents, agentic use cases are popular. So we can get public data sets that are representative of that traffic and train general speculators. Now, with speculators today, you need to train the speculator using the base model itself because you're getting hidden states out of the model from running inference on these specific prompts, and that is the training data you use to create the speculator. So there's that process which you need the real model weights for. And then there's of course just the process of, standing up all the infrastructure behind it, loading all this stuff, testing it. And then when there's a new model with a newer architecture, I think that, like, the DeepSeek models tend to be the most challenging as they have, like, the most novel architectural stuff going on, model after model. But every new model has something. Kimi K2 had. Oh, sorry, GLM-5.2 hadAli [00:14:53]: Sparse attention.Philip [00:14:54]: Yeah,Ali [00:14:54]: YeahPhilip [00:14:54]: the DSA.Ali [00:14:55]: Right. Which is brought from DeepSeek.Philip [00:14:57]: Yeah. AndAli [00:14:59]: So you can copy-paste then?Philip [00:15:01]: It kindAli [00:15:01]: I don't know how this works.Philip [00:15:02]: So, like we had to, like, build support for that into our runtime. And you're right, like it is really interesting the way that all of these open source labs borrow from each other. For example, like GLM-5.2 doesn't have vision. So something that, Haley, a guy on our team, if we could take a look at this, he, like, grafted the Kimi vision encoder onto GLM-5.2.Retrofitting Vision into GLM-5.2Ali [00:15:27]: We'll be training the projector.Philip [00:15:28]: Exactly. So if you think about, like, the encoder, there's the encoder, which is the part that looks at the image and turns it into latent information, and then there's the projector which likeAli [00:15:38]: You can say latent space. It's okay.Philip [00:15:41]: And then there's the projector that maps it onto, the model itself, and then there's the model weights. You don't wanna mess with the model weights because you run a chance of making the model dumber at something else for the purpose of giving it vision. So instead, Haley started with just a projector, which is only a handful of millions of parameters.Ali [00:16:02]: That would be, yeah.Philip [00:16:02]: Yeah.Ali [00:16:03]: Can you show the training one?Ali [00:16:04]: Like the way it groksPhilip [00:16:05]: YeahAli [00:16:06]: Very interesting.Philip [00:16:06]: And maybeAli [00:16:07]: That right therePhilip [00:16:07]: Maybe Ali, you should take it from here. You've got a betterAli [00:16:10]: Ooh, double the sandPhilip [00:16:11]: Understanding of this than I do.Ali [00:16:11]: Yeah. You can see, like, he. The way he trained this is really cool. At the beginning, he was training it using just like, “Here's a picture of a mountain. Can you describe what's in this mountain?” And that caused it just like the first, learning walls. Like here you can see this all we're trying to teach it is to translate the encoded. Like it's already taken the encoder from Kimi K. It's taken the image. It'Philip [00:16:31]: Yeah. FrozenAli [00:16:31]: FrozenPhilip [00:16:32]: With adapter.Ali [00:16:32]: Exactly.Philip [00:16:33]: Yeah.Ali [00:16:33]: So the brain is frozen and the eyes are frozen. It's just we're tryingPhilip [00:16:37]: AlignAli [00:16:38]: Interconnect between the eye and the brain, right? So the projector. And so you take the tokens and then he's like, “Oh, can you describe what's in this image?” And he's like, “Oh, it's a mountain,” or it's a person or it's a human, whatever the case is. But that didn't cause complete understanding. So he changed it such that every image was associated with a data set of questions. Like, does this image have a white male? Does this image have birds in the top corner? Does this image have a scientist in it? All of that stuff. And it would have to answer questions correctly. And using not just training on describing an image, but being able to answer question, another question, answer over time. Like you can see the grokking, which is like genuinely insane, that retrofitting vision into a large LLM can learn to that extent. And even for images that it doesn't perform well on, for instance, if you ask it a picture of like Stephen Hawking, “Who is this?” Maybe it doesn't get it, but it will say something like, “This is Albert Einstein.” Like it still understandsPhilip [00:17:25]: Close enoughAli [00:17:26]: That this is a scientist who is a man who has, some significant achievements, all that stuff. So that's like really cool.Philip [00:17:32]: Yeah. So, we've covered Hao Tian before, who the author of the LLaVA paper that did this, a while ago. And I think that's very foundational work for anyone who hasn't done vision work before.Ali [00:17:41]: Same with the CLIP and MetaCLIP, where you go from just captioning to building out questionsPhilip [00:17:47]: RightAli [00:17:47]: Off the image and how much better you can get performance.Philip [00:17:50]: Right. Right. Right. Yeah. But what's, what's so exciting about this is if you look at a model like this. Now, this is a little bit more of a research project. It's not. It got to 56% on MMLU Pro, I think. So not quite frontier. But if you're running this model, you haven't suffered any loss on your GLM-5.2 quality. If you don't have an image, it'll just behave exactly the way it used to. And ultimatelyAli [00:18:14]: Which in the inference code you literally do not include the other part, right?Philip [00:18:18]: Yeah. You would just skip the encoder if you don't have an image input.Ali [00:18:22]: Okay.Philip [00:18:22]: Just confirming.Philip [00:18:23]: YeahAli [00:18:23]: Does it affect a lot on the overall inference side? Like you're not adding much, you're adding a very small vision encoder. These are typically likePhilip [00:18:30]: They're super fineAli [00:18:31]: Less than a billion parameters, right?Philip [00:18:32]: Yeah. It's, - There's a little bit less standardization among vision encodersSwyx [00:18:37]: YeahPhilip [00:18:37]: So the support matrix can be a little bit, sparser. But overall, yeah, it's a pretty, it's a pretty minor component of the overall system. And ultimately what you get out of the system is all of a sudden you have Kimi Vision, GLM weights, and DeepSeek attention all in one model.Open Source Model Grafting and Franken-MergesPhilip [00:18:56]: And that's, I think, a lot of the power and beauty of open source, is that you can take all of these different components and combine them together into a system that's better than anyoneSwyx [00:19:05]: YeahPhilip [00:19:05]: Can be individually.Swyx [00:19:06]: People used to say that you would also do Franken-merges where you would take likePhilip [00:19:10]: YeahSwyx [00:19:10]: Layers from each model.Swyx [00:19:11]: Does anyone do that anymore?Ali [00:19:13]: Well, to your point previously when you were mentioning like, the work that goes into supporting a model when it first comes out, like GLM-5.2 or MiniMax M3 or whatever the case is. Sometimes you do have to like, you do have to switch out some things. Like, for instance, the MiniMax M3 head uses full attention, and with full attention you end up with this like insane bottleneck in spec dec ‘cause you're doing auto-regressive token generation for three tokens, and you're doing this like N squared over all of the tokens that are in your sequence. Your KV cache is like very large because it's not sparse, it's not top K. So we find it better to like, okay, we're gonna replace this, we're gonna replace this layer with a layer from another model that's using like GQA, for instance. And then just with the right training, you can get it to have the same acceptance rate. So it is very possible to retrofit layers from other models and very much needed. If a layer is like inefficient, the training just becomes the challenge, like how do you ensure that you train it properly? Which again to your earlier point is like the mesh between training and inference. As in like you need very good training in order to do fast inference. That's like, I feel like more and more becoming true.Swyx [00:20:21]: Yeah. Anything else on the support side when you say like get it to fully production ready?Loop Detection, Race Conditions, and Non-DeterminismPhilip [00:20:26]: Yeah. I think that there's also a question of just, we can test a model to a pretty extensive degree, but we're trying to get it out quickly and then you see a bunch of other people test it and you get interesting results. There was an issue with, GLM briefly where we had some like mode collapses where it would just output the same token over and over again for certain prompts on certain temperatures. Like once you expose an endpoint to the real world, there's going to be, so many more varieties of things given to it that you're able to, discover and patch things. So it's not just a, day zero process, it's then like for the first week, for the first month, if a model remains popular, like how do you both fix bugs and then continue to push the envelope on performance?Ali [00:21:21]: What do you mean you don't want your model outputting S?Swyx [00:21:24]: Is there loop detection on that stuff, by the way? It still happens like quite a lot, which is surprising.Ali [00:21:30]: We have like we, in our endpoint, like if a model was to output the same token like four plus times, we just cut the generation. We say like, “Oh, sorry, this-- Like try again,” or like we will reprocess the request. ‘Cause we know then, like if it, like if, yeah, it's four times the same token, it's probably collapsed.Swyx [00:21:45]: Yeah. Is there a way to opt out in case I really want that?Ali [00:21:48]: You want that?Ali [00:21:50]: I think there's a way that we have to handle it. I'm not exactly certain, but I feel like in certain models, like when they output something like you can imagine, like a table for instance, and so they want, they wanna draw like 12 dashes and 12 dashes. Yeah, I think there's a way for that to happen. I think we only do it on certain tokens. Like we exclude certain special characters.Swyx [00:22:07]: Yeah.Ali [00:22:07]: So we only do it on like certain like S is the most common almost. GLM-5.2Swyx [00:22:11]: OhAli [00:22:11]: And I think it was DSV 4 as well. Like you'd just have like looping issues where like you literallySwyx [00:22:17]: ItAli [00:22:17]: Just have like S.Swyx [00:22:18]: Yeah. Is there a special, something special about S? No, just randomlyAli [00:22:21]: It just seems to be the one token involved.Swyx [00:22:23]: Yeah. And it'Philip [00:22:24]: Is thereSwyx [00:22:24]: And it's only temperature 0Ali [00:22:27]: NoSwyx [00:22:27]: Even at other temperaturesAli [00:22:27]: Even at like 0.9 or whatever, it will still, it will still collapse.Swyx [00:22:30]: That's weird, right?Ali [00:22:30]: It's, it is an inference problem to be honest, like a software problem. Like oftentimes, the image you run will-- like NVIDIA will release an image for instance, and if we will upstream the changes from their latest TensorRT-LLM image into our stack, we'll find that it fixes it. Or oftentimes this will only happen in an inference engine that you're using like SGLang. But if you were to switch to vLLM, that isn't the case. So it seems to be like an extremely like deterministic software issue and not really a model issue. It's not like a weights problem. Like I'- we'll say like, “Oh, it's a problem with the quant. We did PTQ wrong,” right? But that isn't, that doesn't make sense because the same weights used with a different inference engine does not repeat the problem. And sometimes it's, the kernels that are being used in the backend have like these very subtle sometimes race conditions, where if you were to use this model hosted on one cluster, you will never get this problem.Swyx [00:23:19]: Oh my God.Ali [00:23:19]: But if you host it on a different cluster, you will. And the reason is the KV cache transfer from a node to node in that one cluster is using a slower interconnect than the node to node in another cluster. So that exposes the race, whereas in another cluster it doesn't. So then you end up just like, okay, this model is not gonna be hosted on this cluster. We're gonna host it on, another cluster because that cluster exposed that problem. But then it ends up with like, okay, is it the software? Is it the model weights or is it the hardware?Swyx [00:23:42]: There is a thing about this with temperature 0 still not being deterministic, right?Ali [00:23:46]: Right.Swyx [00:23:46]: Mostly because of hardware. Even at temperature 0 same model, you won't always get the same output.Swyx [00:23:52]: Even-- But I'm surprised by the race condition one because, I thought PyTorch was a graph that like guarantees that you at least, execute things in the right order.Ali [00:24:02]: Well, yeah, true. Like I'm not, I'm not saying that there is. Like well, you have things like PTL optimizations where like you can start a kernel before the end of the previous kernel, and that's like ‘cause you want to do that because there'sSwyx [00:24:12]: It's like pipeliningAli [00:24:12]: Expense. Exactly.Swyx [00:24:13]: Yeah.Ali [00:24:13]: But it'- But you don't do it cleanly. Like you overlap a little bit of the execution. No, it is very possible that the kernel itself, like that one block that is supposed to be running in this instance of time, that kernel itself has a race condition. For instance, like a missing barrier. Like often if you're designing a kernel and you want it to make it to be very fast, if you don't test it extensively, you'll, you'll have certain threads access data points from registers before they've been written to by other threadsSwyx [00:24:36]: YeahAli [00:24:36]: For example, because like your barrier is wrong or your synchronization was wrong. But yeah, like the testing itself is very difficult in those like, andSwyx [00:24:42]: And there's no like borrow checkerAli [00:24:45]: What does that mean?Swyx [00:24:46]: Like Rust. Like the. If you're trying to have like memory safety It sounds like a comparable problem.Ali [00:24:52]: Well, yes, but you're working in CUDA, right, NVIDIA GPUs. Like- You just need a higher level language like modular Maybe that's what modular is supposed to do. I don't know.Quantization Quality and Vendor FidelityVibhu [00:25:00]: How do you see keeping quality of the model? So you talked about all these steps of, okay, you gotta do quantization, train your own speculative decoderAli [00:25:07]: RightVibhu [00:25:07]: Run on different hardware. Looking at other model providers, okay, you kicked off a inference speed race on the consumer end. What goes into keeping quality the same across them, right? Sure, you can run benchmarksAli [00:25:22]: YeahVibhu [00:25:22]: But, like, how do you determine how much quantization are there standards? What goes intoPhilip [00:25:27]: There's a few things on quality. Most inference optimizations are lossless. KV caching, for example. You are just recomputing or preventing recomputing the same values. Speculation, of course, if a draft token is wrong, it gets rejected. The main lossy optimization is quantization. And that really comes down to, number one, data format, number two, which parts of the model you choose to quantize, which layers, and number three, like doing a lot of calibration on the quantized weights, to ensure that you're preserving all the outliers. There's other tricks that you can do, though. A big one is long context, ‘cause one thing you asked at, right at the beginning is, “Oh, what's gonna happen if I send a 200,000 token request in?” So with a long input sequence, you need to, store a lot more information. You need to process a lot more tokens. And so even if a model has a context of a certain length, you might, as an inference provider, choose to build an API with a shorter context length, and of course a full length one as well. Because if someone doesn't need the full million token context, for example, you can get them better performance. I don't know if that's exactly like quality of the model. The way that I think about quality is to what degree are we faithfully serving the original model? If you think of a golden implementation of a model that performs exactly the way the model is designed to perform, I think of quality as how close are we getting to that, 100% fidelity of the model.Philip [00:27:13]: You can also, of course, think about quality from the training side and how do you push yourself past 100%. But when I think about purely inference optimizations, it's getting faster while staying as close to that 100% fidelity mark as possible. And certainly our standard internally is that, like you should not be able to tell the difference between our API and a, official API. I think Kimi in particular does a good job of vendor benchmarking hereAli [00:27:41]: YesPhilip [00:27:41]: Where they haveAli [00:27:42]: They released an actual vendor benchmark.Philip [00:27:43]: Exactly, yeah.Ali [00:27:44]: ‘Cause they accused, some people, Amazon? There was some provider that was not doing very well on Kimi's benchmark.Philip [00:27:50]: Yeah.Philip [00:27:51]: So, with Reflect we probablyVibhu [00:27:52]: This was a long time ago, right?Philip [00:27:54]: No.Ali [00:27:54]: Yeah, like threeVibhu [00:27:55]: They alsoAli [00:27:55]: Four, five months agoVibhu [00:27:57]: This also happened with, I don't remember which model, but they pulled out quite a few, and then they started a whole chart about this. It might have beenPhilip [00:28:03]: Kimi Vendor Verifier.Ali [00:28:04]: Yeah.Philip [00:28:05]: Yeah.Ali [00:28:05]: Yeah, ‘cause you, ‘cause you'd be pissed, right? Like if you'Philip [00:28:07]: Yeah.Ali [00:28:07]: If like if I'm a consumer and I'm using like Amazon's endpoint for instance, and I've used Kimi and I'm like, “Oh my God, like this is bad,” I'm not gonna say, “Oh, Amazon quantized the model in a bad way.” I'm gonna say, “Oh, Kimi sucks.” Right?Philip [00:28:17]: Yeah.Ali [00:28:17]: So it seems like that makes sense.Philip [00:28:19]: Yeah, they care. They care.Vibhu [00:28:21]: Justifiably.Ali [00:28:21]: Yeah, justifiably.Vibhu [00:28:22]: This is probably a stupid question, but just checking, has anything improved from main quantization?Philip [00:28:28]: Yeah.Vibhu [00:28:28]: Like, is quantization always strictly worse?Ali [00:28:30]: Well technicallyVibhu [00:28:32]: NoAli [00:28:32]: It's a lossy. QuantizationPhilip [00:28:33]: YeahAli [00:28:33]: Is a lossy, it's a lossy implementation.Philip [00:28:36]: Speed improvesVibhu [00:28:36]: Speed improves.Ali [00:28:37]: It the number, likeVibhu [00:28:38]: No, I' always look for inverse scaling laws.Philip [00:28:40]: Yeah.Ali [00:28:40]: Yeah.Vibhu [00:28:40]: This is something I learned from Noam Brown, where like things that normally act in one direction sometimes do.Philip [00:28:45]: Well, technically when you run a benchmark, because these models are deterministic, sometimes your,Ali [00:28:52]: YeahPhilip [00:28:52]: NVFP4 quant is like, two basis points higher than yourAli [00:28:56]: No, it's noise. It's noise.Philip [00:28:57]: Yeah, exactly. I'm like, yeah, it's, it's within. That's why I always say within margin of error.Philip [00:29:01]: And I stopped saying that because everyone assumes that what is, well, within some margin of error, we're barely inside of that to the worst, so we're saying. But yeah, sometimes it's just like, gives you a higher output score. But like Ali said, that's noise. To my knowledge, you're not necessarily making the results better. You're just trying to, again, like keep your fidelity as close to 100% to the original model.Layer Selection, KL Divergence, and Better QuantizationAli [00:29:27]: There is, to your point, research that we did on MP. I don't know if you are able to pullPhilip [00:29:31]: YeahAli [00:29:32]: A tweet we did. One of our research interns, Joshua, I think it's a tweet on how we have 20% better quantized GLM-5.2 than NVIDIA. Essentially what we found throughout like this month research is, okay, quantization is a lossy. It's. You're compressing the data from, occupying 16 bits to occupying, four bits, for instance. And so you're losing some information, and you're trying to minimize that. And so when I say that I'm gonna quantize the model, my job becomes how do I find the layers that I can quantize, and how to find the layers to not. For instance, with image models, I don't quantize modulation layers, and I don't quantize out projections because those two are. Like out projection is what you see as the user. Modulation is what the model sees or understands. Right, exactly. And so to his paper, do you have the. It doesn't have the. Yeah. It's a long paper. I don't know if I can findVibhu [00:30:25]: If there's a part to search or it's probably in the thread.Ali [00:30:28]: It's probably in the thread.Vibhu [00:30:29]: Yeah.Ali [00:30:29]: But the long and the short is it is very possible that quantizing more of the model makes the results. Like if I have a model that I quantize layers one, five, and 10, and another model where I only quantize layers one and It is possible that the model in which I quantized more information is going to perform better because the quantization errors have canceled out. And so what Joshua showed in his mathematical proof where he had like a verifier in, is that you can predict which layers are going to have quantization errors that will cancel out with each other, and you choose to quantize those layers. And so the result of doing this mathematical quantization is you end up with a model that's 20% more quantized than another provider, so you get 20% more throughput of it because there's more layers than running an NVFP4, and your quality is better than that other quant because the layers that you chose to quantize have their errors cancel out, like one layer skewed to the right one layer skewed to the left, one layer skewed to the right. Your final logits distribution is more similar to the original distribution of the model, so you have better fidelity. And so the way we proved this was with KL divergence. So instead of just scoring on the benchmarks, we scored the KL divergence between the logit distribution of the quantized model and the logit distribution of the original full precision model, and we showed that with this technique we get. If your probability distribution on the logits which token it wants to select is more of the same as the original model, you're probably gonna end up staying true to the original model. So yeah, so it seems like previously before this, it seemed like the industry was, well, the more you quantize, the worse it's gonna be, ‘cause the more loss you introduce. That's not exactly, not necessarily true. So yeah, doesn't improve it, but can cancel out.Philip [00:31:57]: I think it might be this, but reminds me a good bit about pruning where you can prune off certain layers.Philip [00:32:03]: But very interesting. Didn't know this was a whole paper you guys put out.Ali [00:32:06]: It's. Fun fact, it was originally 72 pages, this paper, and then we decidedPhilip [00:32:11]: WowAli [00:32:11]: We can't tell. We couldn't release it. So it's now 45.Swyx [00:32:15]: Still 39 pages, so very substantive. We talked about evals and all these things and, like what's possible in terms of speedup? Like it's like probably like the numberInference Speedups and BenchmarkingSwyx [00:32:25]: Thing that people do wanna care about, and it's something that you wrote about in your post. Like official API is 70 tokens per second, and you push it up to 90. Is that like a normal thing?Philip [00:32:36]: So what's cool about working in inference, the reason that I think inference is going to be a useful place to do engineering for a long time, is that if you look at highly optimized domains like, say, finance, if you're in finance, you measure how much better you got in basis points. It's like, “Oh, I got five basis points better, like twentieth of 1% better,” that's huge news because everything is so optimized. When we publish optimizations, it's 20%, it's 100% it's 200%. So there's still probably like a lot further to go, honestly. Like you'll, you'll know that inference is pretty much solved when researchers start publishing about how they got 1% faster at something.Swyx [00:33:19]: Which by the way, because I am from the finance background, in the ‘70s, that was the margin at the time. When you did quantitative finance research, you would findAli [00:33:27]: And like 20%, tens of percent.Swyx [00:33:29]: That's. Yes.Philip [00:33:29]: Yeah.Swyx [00:33:30]: And now it'Philip [00:33:31]: Tiny fractionsSwyx [00:33:32]: For those people interested, look up Andrew Lo's paper. He had a really interesting illustration of quant, stat arb, distribution, narrowing down from like those kinds of 20% differences in the ‘70s, down to nothing today, which is very cool.Philip [00:33:48]: Exactly, and we're at the beginning of the same type of thing. Now benchmarking is hard. I think anyone will tell you that, and benchmarking provider speeds is hard because there's so many variables that go into it. What hardware are you using? How much load do you have on the system? What's the exact nature of the prompts and input and output sequence lengths? All that stuff. But overall, when you start stacking these improvements, you're looking at multiples. You can look at it. The most common form, of course, is TPS, tokens per second, which is bad naming by us in the industry, ‘cause there's two tokens per second. There's tokens per second, the throughput number, and the latency number.Ali [00:34:31]: TTMT, yeah.Philip [00:34:32]: Like total tokens per second out of the, out of the GPU as a throughput number. Most people only care about tokens per second as the latency number, which we should call ITL, intertoken latency, but we don't.Philip [00:34:44]: Anyway, so you can imagine a standard API without many optimizations for a 1 trillion parameter model operating somewhere in the 30 to 50 tokens per second range for reasonable traffic profile. And we generally see the goal of, pushing to 10X that. But, not necessarily day zero, but by stacking enough optimizations, if you have, say like four optimizations, each of which doubles performance. Or sorry, three optimizations, each of which doubles performance, then you stack that up, that's an 8X gain. That's the order of magnitude that we're working with in this space. We're trying to make things substantially faster, not just go from like 70 to 90.Swyx [00:35:38]: Are you saying you've. You have done that?Philip [00:35:40]: So let's say you have as a reasonable baseline, 30 or 40 tokens per second. You can achieve 10X that. So like on GLM-5.2, if you run it unquantized, perhaps on H100s even, and you're just using an off-the-shelf inference engine with no particular optimizations, no speculator, nothing extra around like KV routing, no disaggregation, you're, you're probably, yeah, looking at that like 30 to 40. You think that's like a reasonable baseline?Swyx [00:36:12]: Right. Right.Philip [00:36:12]: To get to something like 10X, there's a lot of trade-offs that you're making. If we're running at more like a 300, 400 tokens per second range, you are using the best hardware possible. You have a optimized speculator. You have done all of your quantization work. You are Seeing a pretty high cache hit rate. You are running with a reasonably small batch size and a parallelism configuration that is tuned for latency versus throughput, but it is possible. So the spreads that you see if you, like, go on artificial analysis or you go on OpenRouter and you look at, the worst provider to the best provider, oftentimes can hit that range. 10X is of course very aggressive. It's oftentimes maybe more of a four to six times improvement. But that's the performance that makes us really excited, is when we can get these huge gains, not just go from 70 to 90 tokens.Stacking Optimizations: NVFP4, Speculation, and DisaggregationAli [00:37:19]: It's also, like, hardware dependent. Like, ifPhilip [00:37:20]: YeahAli [00:37:20]: If you have a thing where you're serving it on just, like, a node of H100s and then you throw, like, you shard the model across, like, four nodes of B200s. Like, you can definitely increase the speed with just throwing more hardware at it. Like, normalizing for the same exact hardware and the same number of GPUs.Philip [00:37:35]: Yeah. Then you're looking at, like, a two to 4X improvementAli [00:37:38]: Right. RightPhilip [00:37:38]: Depending on the inference optimizations. So yeah, it's. Some of it's, what's the call, and some of it's who's the driver.Vibhu [00:37:46]: If you break down the two to 4X, say the example is run GLM-5.2Ali [00:37:51]: YeahVibhu [00:37:51]: On B200sAli [00:37:53]: YeahVibhu [00:37:53]: Single node, right? What's, like, the cost trade-off for effort to get, like, the last bit of juice out versus what should people just think of, right?Ali [00:38:01]: Spectre quantization. Yeah.Vibhu [00:38:03]: Spectre quantization.Ali [00:38:04]: That's, that's, that's like 95%. LikeVibhu [00:38:06]: And how far does that get you? And how easy is that for the average person to do? So say right I wanna throw the weights of GLM-5.2 on a node of B200s, how easy is it to find speculative decoder- decoder model or already quantized model? How much work goes into it?Philip [00:38:23]: If you're doing it up front, it's quite a lot of work. If you're doing it today, there's going to be people who have published things that you can just, you can just grab some NVFP4 weights. You can grab a speculator. Yeah, if we're thinking about, like, what are the 2Xs we're stacking, going from, BF16 to NVFP4 is, it's not quite a 2X, right? It's like. I think it's about, like, 30 to 40%, from 16 to 8, and then another 30 to 40% multiplied from, 8 to 4. So that doesn't quite get you a 2X, but, like, roughly a 2X. Speculator, roughly a 2X. Disagg on top of that if you're able to get enough hardware and put enough traffic through it, another roughly a 2X. And then you add in some, double-digit percent increase from having just a better runtime with, the latest kernels and stuff behind it. And that's how it stacks up.Ali [00:39:21]: YeahPhilip [00:39:21]: So building each of those, like, building the, quantized weights is, for someone who really knows what they're doing, hours to days of work. Building the speculator, again, like, hours to days of work. And the, disagg setup, hours to days. Well okay, but like once you haveAli [00:39:39]: Once set up. Once set up. YeahPhilip [00:39:40]: Yeah, getting disagg working for the first time, I'm saying, of course, is very difficult.Philip [00:39:44]: The marginal implementationAli [00:39:48]: Like, if you're just grabbing, like if you are a person, like just a normal consumer who has access to, like, a node of B200s and you're wondering, “How can I just host it myself?” You don't need to quantize the model yourself. There's always gonna be, like, an open source quantized checkpoint. NVIDIA's gonna push one out if no one else does. You. Usually, the providers will have their own spec dec that they've trained as well. You don't need to train your own spec dec. You can just use that as well.Philip [00:40:09]: Yeah. Like, GLM-5.2 has its own MTP.Ali [00:40:13]: Right. Right.Vibhu [00:40:14]: What's multi token prediction?Philip [00:40:15]: Yes.Ali [00:40:16]: I'm justVibhu [00:40:16]: Can you explain that?Ali [00:40:16]: I'm just an expert.Ali [00:40:18]: I can do it for you in case I get it wrong?Vibhu [00:40:20]: No.Vibhu [00:40:21]: Yeah, you should correct if we're wrong, but their multi-token prediction can be used for self-speculative decoding.Ali [00:40:27]: I'm not sure. I'm not gonna correct that.Vibhu [00:40:28]: Okay. I'm semi-confident in thatAli [00:40:30]: Okay. YeahVibhu [00:40:30]: But someone can check. But it's useful to paint the story of, okay, not just the average person, but say a company wants to switch from serverless inference I wanna throw this up on. I wanna rent some GPUs, throw it up. These are the steps you take to do significantly faster than just put it behind vLLM.Ali [00:40:48]: Right.Vibhu [00:40:49]: I was waiting for a mention of Dynamo.Vibhu [00:40:51]: I feel like, that's supposed to be the baseline that you measure against.Dynamo, KV Routing, and Disaggregation ToolkitsPhilip [00:40:55]: I would think of Dynamo as less of a box system and more of a toolkit for building with. So when we talk about doing aware routing, when we talk about doing KV offloading, when we talk about doing, PD disaggregation, Dynamo fundamentally is. By the way, Dynamo is an open source library from NVIDIA.Ali [00:41:17]: We've done a pod with KylePhilip [00:41:18]: OkayAli [00:41:19]: Kyle Cranin.Philip [00:41:19]: Cool. So then your listeners know then that it supports all the different inference frameworks. And it is multi hardware, which is interesting.Ali [00:41:28]: But it's just a router, it's not like an optimizer layer.Philip [00:41:30]: Yeah. All it does, like, what Dynamo is good at, it is a library for moving information around your cluster, around your hardware. So if you have, KV cache on one place and you need it to be somewhere else, Dynamo coordinates NIXL for you to move that around.Philip [00:41:49]: That doesn't mean that, like, out of the box, you just say, “Pip install Dynamo,” and then you get, like, a massive performance speed up. It's more of a developer toolkit.Ali [00:42:01]: Yeah. I would have said it would. It comes with a set of defaults that you can then swap out.Philip [00:42:06]: It does. If the industry at large, I think, was, like, rolling out all of these deployments, standard, then I think it would be, like, a credible baseline. But, we've got to, we've got to benchmark against, like, what we're seeing in the wild.Speculative Decoding Methods: Medusa, EAGLE, n-Gram, and Spec-SpecVibhu [00:42:23]: I did wanna talk a little bit more about PD disagg, because that is probably, like, number three after quantized and speculative decoding. In your book though, I was just gonna pull out the book.Philip [00:42:31]: Yeah.Vibhu [00:42:32]: Like section 522 on Medusa, 523 on EAGLEPhilip [00:42:35]: YeahVibhu [00:42:36]: 524 on gram.Philip [00:42:37]: It's 55, would be disaggregationAli [00:42:42]: Yeah. Well, no, I just wanted to dwell a little bitPhilip [00:42:44]: YeahAli [00:42:44]: The other. Like, so what do you choose to include? What do you choose to not to include? Because there was all these other techniques.Philip [00:42:51]: Yeah.Ali [00:42:51]: Are these still relevant? Because I think they came out, like, a year and a half ago maybe.Vibhu [00:42:55]: Medusa is quite old.Philip [00:42:56]: Yeah, Medusa's old.Ali [00:42:58]: It was old.Vibhu [00:42:58]: But is it in the book as a good, here'sPhilip [00:43:01]: BaselineVibhu [00:43:01]: Baseline vanilla understand it?Philip [00:43:02]: Like you should know this.Vibhu [00:43:03]: Like I read the paper, I'm like, “ it makes so much sense.”Philip [00:43:05]: Yeah.Philip [00:43:05]: So with the book, I had a couple goals. One was to give people just a working vocabulary for the space as a whole, and the other was to give them some intuition about how each of these techniques works. As I mentioned in my AI Engineer talk, which is the first public addendum to this, the speculation space has moved much faster than everything else. So yeah, even at the time that I wrote the book Medusa, I very much included as a way for people to understand how the space evolved rather than what the most modern technique is. And now of course, there's DFlash, dSpark. There's, there's newer techniques even than EAGLE, although EAGLE is still very commonly used.Ali [00:43:51]: SpecSpecta.Philip [00:43:52]: Yes. Speculative decoding.Vibhu [00:43:54]: What canAli [00:43:56]: Oh, it's a paper by Tri Dao and it's like, it's doing speculative decodingVibhu [00:44:00]: HuhAli [00:44:01]: For the speculative decoder.Philip [00:44:02]: Oh, in spec- oh my God.Ali [00:44:02]: It's literally just an another. It's like, yeah, that's the most simple way to explain it, and it seems like he got trivial speed ups there. But it seems that the complexity with training, it's almost like in our mind at least, it's almost as complex as training GANs. Like it's like a very delicate balance and oftentimes you, it's just but yeah, it's literally speculative decoding on speculative decoding.Vibhu [00:44:21]: Speculative.Ali [00:44:22]: Yeah. We saw this paper.Vibhu [00:44:24]: It's interesting, right?Ali [00:44:24]: Yeah.Vibhu [00:44:24]: I wouldn't even expect it to be very particular to train, I wouldAli [00:44:29]: Right.Vibhu [00:44:29]: The naive part of me is like, okay, train speculative decoder.Ali [00:44:32]: But like, and it makes sense, like the whole idea of speculative decoding is you. It's like, it's like almost like the iPhone auto predict version but for a normal model, right? Like you're just, you're just, generating three tokens and you're like, okay, I'll do prefill on them. And so you save those three turns for your original model. Now your speculative decoder is doing three turns of auto regression, so why not just have an even smaller model?Ali [00:44:53]: The other question there is what are the size of speculators? So say forPhilip [00:44:58]: Right. It's like a billion parameters.Ali [00:45:01]: Like for MiniMax, it's. Yeah. It's like one layer. It's like one 60th of the original model usually.Philip [00:45:06]: Yeah. I think we should do a paper when we get back to the office.Philip [00:45:10]: SpeculativeAli [00:45:11]: SpeculativePhilip [00:45:11]: Decoding.Ali [00:45:13]: No, it's, it does seem like how, when do you stop? But then it also seems like if you're able to train spec-spec decode for instance, right? Like if you're able to have a small model that is accurately predicts what the intermediate speculator is gonna predict, that is able to predict what the original target model's gonna predict, then why not just use that smallest model directly, right?Vibhu [00:45:34]: Yeah. This isAli [00:45:35]: Like it seems likeVibhu [00:45:35]: Adjacent to the routing problem.Ali [00:45:36]: Right.Vibhu [00:45:36]: Yeah.Ali [00:45:36]: Right.Philip [00:45:37]: The thing with speculators is one of the practical constraints on using them is that you do have to run a small model on the same hardware that you're running the big model on. There is a orchestration and resource competition problem inherent in that, and that is one of the constraints on speculation in general, is that draft tokens cost resources to create and cost software complexity to manage. And so if you have like infinitely recursive speculators, you add in quite a bit of that complexity on the actual implementation within the inference engine as well, not just in the training process.Vibhu [00:46:17]: I was gonna say, I would wonder if you could do similar, like distillation and pruning of, it's the same thing, it's just a model. Can we not just distill a lot of the weights, quantize the speculator, out of my domain? The question that also comes up is, this is all for big server workloads, right? How much of this applies to, say I have this MacBook, I wanna run Gemma really efficiently. Similar problems, not the same?Local AI vs. Data Center InferencePhilip [00:46:45]: Pretty different. I talked to Selo, about this on his podcast a couple weeks ago. The difference between inference engineering for the data center and for production workloads versus inference engineering for local AI, is that we start with fundamentally like different constraints and different goals. With local AI, it's how do I fit this model onto my hardware and then make it less dumb? And with data center influence, it's how do I load this model and then make it less slow? And we care about less dumb, and they care about less slow. But the local AI inference engineering ecosystem, I think has a lot for us to learn from in the data center space. They are experts in various forms of quantization, including dynamic quantization that we just don't touch, in the pruning, in the distillation, in the, layer removal. There'Ali [00:47:42]: Layer removal matters less.Philip [00:47:43]: Yeah. There'Ali [00:47:44]: No one loves pruning really.Philip [00:47:45]: Yeah. Well, but the, but they doVibhu [00:47:46]: Which is surprising, right? But that's, that's a whole different thingPhilip [00:47:48]: Just to fit something on the laptop.Ali [00:47:50]: Right.Philip [00:47:50]: So yeah, it's a, it's an interesting, it's an interesting space. Not necessarily that like their techniques make sense for us to do in the data center, because we have different resources and different goals, but more that the process as well as the openness of that field is something to, admire.Ali [00:48:12]: Yeah. Like to your point, like, certain optimizations that would. Like for instance, Turbo Quantum Sharper, like it made such huge hype on that and we did like a whole deep dive on Twitter and like said, what is it? How does it work? Why is it good or not? And it took off and it was implemented on local devices because your memory bandwidth is so slow on like a MacBook, for instance. But try putting the same thing on like an NVIDIA GPU on a B200 Turbo quant would not be. Like, it would not be used. Like, NVIDIA - Like, NVIDIA made it clear that this is not a good optimization, and we've seen it firsthand where the overhead of doing dequantization, quantization of, in the kernel itself with turbo quant kernel, each end is much slower than the time that you save from doing the bandwidth. ‘Cause on the B200s, you have like 3.5 terabytes per second. You don't need decrease the storage that much. You don't need to do, FP4 KV cache. You don't need to use a requant. There's, there's, there's better optimizations to be made. But on Edge devices, it's extremely important, it's extremely useful. So, seems to be, like, different optimizations there, but then they're all uniquely combined with like all you wanna quantize the model, you wanna do speculative decoding, like certain common prefixes with bothPhilip [00:49:18]: Principles.Ali [00:49:19]: Yeah, exactly. Exactly. Exactly.Philip [00:49:20]: They also do a lot of work on, model parallelism, especially over, heterogeneous topology, where you have, some sparks and they are wired together with, Ethernet, DGX sparks.Ali [00:49:35]: Yeah, this is the Exo Labs guys.Philip [00:49:36]: Yeah. You have, a nu

god amazon ai english google hollywood future training building race video writing fun local mit single open iphone harry potter attention run silicon valley tool focusing speed networking principles pc speech scale couple engineering burning roi masterclass mma pleasure wikipedia chip reflect limits grandma albert einstein expert models intel depending images similar eagle mercury openai fireworks compare inter pipeline nvidia decoding speculation halfway cosmos hardware users api preserving nic generally mp intern smaller layers continuous gpt open source mythos waterloo llama pd apis spectre gram rubin stephen hawking layer clip macbook amd llm 10x medusa sparta fps hopper pip blackwell gpu expense agi cpu kimi tp cache midjourney baseline dsa franken prs dynamo gpus kv spec adjacent continual orpheus tps oof spacetime 2x asics wan speculative t4 4x broadcom itali taha 100x veo gans rl json ethernet rsi asic kling tpm inference ep1 ep2 selo feynman modulation cuda 20b what it takes mlp kiely like you tensor speculators nvidia gpus bnf 25b ampere retrofitting pytorch aie sparse nics ptl yaml mtp itl 8x cerebras jamba minimax dsv glm hbm ai engineer loras tom l zdr sft rubins x6 justifiably tmas martin casado andrew lo gml ptq dgx failure modes nvl nvlink noam brown t4s tp2 tp1

Rev. Dr. Harden Hopper Preaches on Matthew 14:13-21

rev james harden matthew 14 hopper preaches first united methodist church

Play Episode Listen Later Aug 2, 2026 14:58

Rev. Dr. Harden Hopper preaches on Matthew 14:13-21.First United Methodist Church of MariettaGiving link: https://onrealm.org/mariettafumc/-/form/give/nowChurch website: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://www.mariettafumc.org/

CLCC - SUMMER OF REVIVAL WEEK#5 (VISION) / PASTOR NEIL HOPPER

Play Episode Listen Later Jul 29, 2026 45:53 Transcription Available

CLCC - A Summer of Revival - Week#5 Vision Without a vision people cast off restraint. Without a vision people perish. We want you to ask the Lord for a vision for yourself and your family, from God's perspective. We are promised in Acts 2:17 and your sons and your daughters shall prophesy, and your young men shall see visions, and your old men shall dream dreams. Vision is God's idea! What God shows you, you can have faith for. https://www.instagram.com/cedarlake_cc/

god lord vision acts revival hopper pastor neil clcc

Raine on Your Parade: Stranger Things - Dear Billy

Raine On Your Parade

Play Episode Listen Later Jul 26, 2026 120:18

Okay back to Stranger ThingsIn this episode we come back to California where some people are telling Mike “trust us with your girlfriend, we're the government” and his head explodes. But instead of leading with the note *FROM* El confirming these are indeed the “good” guys, they just mention Dr. Owens… eh. Also, El throwing shade with that *FROM*. And also FROM is a good show, watch it. El's the best character in this episode and she's not even in it. In Hawkins Max is diagnosing herself with “death” via “psychic torment” and gives herself about seventeen hours to live. Lucas has come around just in time to get murdered by Steve with a lamp in the school hallway. The next morning Max decides to write all day while trying to ignore everyone so... she has turned into… me? Nancy forges some documents to be alone with Robin again and Robin is like, “You like Tom Cruise, who looks like Steve” (NAH…) and Nancy is like “Girl that's old, I like someone else now and it's not you, seriously, I am so agitated whenever you talk or breathe or smile… sigh… what was I saying?” In Alaska, Murray and Joyce get a hotel and there ARE two beds. He says one year old babies are idiots and it's kind of hilarious. Then they go give a stranger $40,000 and drink the drink he offers them.ARE Y'ALL NEW HERE? And by “here” I mean, on EARTH! Joyce, you're naïve in a sweet, wholesome way. You're excused. BUT MURRAY “DON'T ENTER MY HOME WITHOUT PROPER I.D.” Bauman… I expected better, you're stupider than a one-year-old. Hopper and Enzo are in the Russian playground and Enzo is like “Hopper and Joyce, sitting in a tree…” Hopper is like “she's NOT my girlfriend!” and Ezno is like “right, she risks her life for you because of ‘friendship'” And Hopper says, “not just friends… BEST friends” and Will drops in from a helicopter and shoots him. (mentally) In California Jonathon is annoyed that the government agents are lazy and gets an idea. Will and Mike discuss their friendship. Mike is like “we are a team” and Will is like “this means a normal heterosexual amount to me.”Jonathon comes in and says he has a plan to order pizza from a talking pot plant. Max hands out gift cards in the event of her death. Vecna does some laundry and Robin loses her shit because of a pink shirt and a bra. Somehow being crazy works in Robin's favor for once (or like always?) and they get to talk to Freddie Krueger in the Hannibal Lector basement of the insane asylum. Hopper does a quick little marvel movie and then eats peanut butter with his fingers. Joyce and Murray pass out and get kidnapped while Yuri calls Enzo to rub it in his face that they all got played. Everyone in snowy areas gets captured. All the boys in California get shot at and escape via a flying pizza van driven by Cheech and somehow also Chong. Robin and Nancy learn about the *POWER OF MUSIC* (for real) And Victor tells them about a haunting the 50s that killed his family. But they too get caught for lying and then run out of there Cinderella style leaving all their shoes behind. Max goes to the cemetery to apologize to her abuser for his death that he got himself into (sort of)… and honestly I can't even make jokes, these parts are epic. Max is put under a trance and talks to her brother who is dead (and/or in Australia filming this) But it's actually Vecna and he chases her into his own mind where she almost dies by floating.BUT Kate Bush and the power of friendship save her (and it's not even cheesy) and that was quite the episode.

california australia earth girl russian dear stranger things tom cruise parade owens yuri hopper chong power of music cheech vecna hannibal lector in alaska freddie krueger

Rev. Dr. Harden Hopper Preaches on Matthew 13:31-33, 44-52

rev james harden matthew 13 hopper preaches first united methodist church

Play Episode Listen Later Jul 26, 2026 16:01

Rev. Dr. Harden Hopper preaches on Matthew 13:31-33, 44-52.First United Methodist Church of MariettaGiving link: https://onrealm.org/mariettafumc/-/form/give/nowChurch website: ⁠⁠⁠⁠⁠⁠⁠⁠⁠https://www.mariettafumc.org/

Riot Girls

Smarty Pants

Play Episode Listen Later Jul 24, 2026 23:49

In her 25 years as a music journalist, Jessica Hopper has profiled the doyennes of modern rock and pop music: Björk, Kacey Musgraves, St. Vincent, Liz Phair, Robyn, and many more. Her reviews run the gamut from the latest Nicki Minaj album and the “mobile shopping mall that is the Vans Warped Tour” to the only album by D.C.'s first all-women punk band, released three decades after they broke up. The new second edition of The First Collection of Criticism by a Living Female Rock Critic expands on the 2015 one. That the provocative (and mostly accurate) title still works six years later points out that rock criticism has even fewer women in it than rock music does. Hopper joins us on the podcast to discuss her writing, from her beginnings as a local Chicago critic to her expansive oral histories of Hole and the women who transformed Rolling Stone in the 1970s. This episode originally aired in 2021.Go beyond the episode:Jessica Hopper's The First Collection of Criticism by a Living Female Rock CriticRead “Building a Mystery,” her oral history of Lilith Fair, and her reflections on Joni Mitchell's Blue, 50 years onListen to her eclectic playlist of music that came out of ChicagoHopper hosted Season 2 of KCRW's Lost Notes podcast, looking at artistic legacies of the likes of The Freeze and Cat PowerTune in every other week to catch interviews with the liveliest voices from literature, the arts, sciences, history, and public affairs; reports on cutting-edge works in progress; long-form narratives; and compelling excerpts from new books. Hosted by Stephanie Bastek.Subscribe: iTunes/Apple • Amazon • Google • Acast • PandoraHave suggestions for projects you'd like us to catch up on, or writers you want to hear from? Send us a note: podcast [at] theamericanscholar [dot] org. And rate us on iTunes! Hosted on Acast. See acast.com/privacy for more information.

chicago mystery girls rolling stones hole acast bj criticism riot nicki minaj freeze joni mitchell hopper kacey musgraves kcrw vans warped tour liz phair lilith fair jessica hopper lost notes living female rock critic stephanie bastek

HRV and Breathing for Life: Master Phil in Your Corner EP160

Master Phil in Your Corner

Play Episode Listen Later Jul 24, 2026 59:33

HRV and Breathing for Life: Master Phil in Your Corner EP160 This was episode 160 of"Master Phil in Your Corner" where Phil interviewed Dr. David Hopper, a chiropractor and former MMA fighter, about sleep apnea, heart rate variability (HRV), and natural health approaches. Dr. Hopper explained howobstructive sleep apnea affects the autonomic nervous system by triggering fight-or-flight responses during sleep, leading to cardiovascular issues and cognitive decline. They discussed how HRV measurements can predict illness andinjury up to two days in advance, with Dr. Hopper recommending his book "The Secret to Never Getting Sick" for improving HRV through proper breathing techniques, nutrition, and grounding practices. Phil shared his ownexperience tracking biometrics and using neti pots, while Dr. Hopper advised on nasal passages and bedtime routines. The conversation also covered Dr. Hopper'srecent move to Arkansas and his focus on online therapy for sleep apnea, withPhil announcing his upcoming fundraising challenge of doing 1,000 push-ups daily from August 11th to September 11th to benefit Tunnels to Towers, an organization supporting 9/11 first responders. #masterphil#masterphilinyourcorner #fitness #BJJ #HRV #Sleepapnea #breathwork #breathing#Lifebreath #breathe #healthychoices #911 #T2T https://www.drdavidhopper.com/https://www.linkedin.com/in/davidehopper/Hopperdavide@gmail.com

secret arkansas mma breathing towers tunnels hopper hrv your corner master phil david hopper

810: The Boss of Your Kitchen con Omi Hopper

The TASTE Podcast

Play Episode Listen Later Jul 22, 2026 41:43

Omi Hopper is a Puerto Rican chef, social media influencer, and food entrepreneur. You might know her as a semifinalist on season two of Next Level Chef, or from her popular cooking videos across social media, where she shares her takes on Puerto Rican classics. Today on the show, we talk about making her debut cookbook, Cooking con Omi: A Love Letter to Puerto Rican Home Cooking. Subscribe to This Is TASTE: ⁠⁠Apple Podcasts⁠⁠, ⁠⁠Spotify⁠⁠, ⁠⁠YouTube⁠⁠ Learn more about your ad choices. Visit megaphone.fm/adchoices

spotify boss kitchen cooking puerto rican hopper next level chef

302: The Truth About Hopper Freight: Can You REALLY Make Money?

The Oakley Podcast

Play Episode Listen Later Jul 22, 2026 44:00

This week on the Oakley Podcast, Jeremy Kellett sits down with Hopper Division Operations Managers Bryan Hill and Russell Vallance to break down how Oakley's Hopper Division works, who it's right for, and why it's growing. They explain Oakley's three dry bulk divisions (hoppers, end dumps, pneumatics), how hopper freight is consistent with many repeat customers (roofing granules, pet food, etc.), and why hoppers are often the simplest way for an owner-operator to get started, with no need for a wet kit or blower and relatively light trailers. They discuss pros such as steady freight, regular lanes, the ability to get many drivers home most weekends, and strong customer relationships, as well as cons including cleaning out trailers and competition from smaller hopper carriers. The conversation highlights how dispatchers are motivated to keep drivers generating revenue, the importance of effective communication regarding load times and breakdowns, the role of technology and e-logs in day-to-day operations, and specific growth needs in regions such as Indiana, Illinois, Missouri, and the Carolinas. Overall, the key takeaway is that Oakley's hopper division offers a strong niche with solid earning potential and consistency for committed owner-operators who value service and communication. Key topics in today's conversation include: Welcome to Today's Episode on the Hopper Division (0:43) Introducing Bryan Hill and Russell Vallance and Their Backgrounds (4:54) Roles of Hopper Operations Managers and Board Structures (7:20) Pros of Hopper Division From an Operations Perspective (10:16) Cons for Drivers Like Cleaning Out Hoppers and Weather Challenges (13:02) Pay Structure, Weight Incentives, and Importance of Light Trucks (15:25) How Dispatchers Stay Motivated and Focused on Driver Earnings (18:57) Biggest Daily Challenges With Delivery Times and Communication (21:34) Lessons Learned Moving From Dispatcher to Operations Manager (23:49) Why Hopper Division Could Grow Much Larger and Why It Hasn't Yet (28:04) Home Weekend Opportunities in Midwest and East Coast Regions (30:34) Key Messages to Current Owner Operators About Communication (33:49) Final Thoughts and Takeaways (42:34) Oakley Trucking is a family-owned and operated trucking company headquartered in North Little Rock, Arkansas. For more information, check out our show website: podcast.bruceoakley.com. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

illinois indiana missouri midwest arkansas takeaways pros focused roles cons final thoughts make money carolinas operations managers hopper simplecast freight north little rock

A SUMMER OF REVIVAL / BEING FILLED / PASTOR NEIL HOPPER

revival hopper pastor neil

Play Episode Listen Later Jul 22, 2026 50:53 Transcription Available

https://www.instagram.com/cedarlake_cc/

57% of the Prisoners Cooper Released Committed New Crimes — and He Wants to Be a Senator

American Ground Radio

Play Episode Listen Later Jul 22, 2026 41:50 Transcription Available

You’re listening to American Ground Radio with Stephen Parr and Louis R. Avallone. This is the full show for July 21, 2026. We open with a 29-year-old Democratic Socialist named Milat Kiros who just defeated a 15-term congresswoman in Colorado's Democratic primary — with the backing of the DSA and Bernie Sanders — and who has told voters exactly what she believes: that America is exploitative and extractive, that ICE is a terror apparatus, that Israel's military aid should end, and that the existing world order needs a reckoning. We note that she arrived in America because her father won the U.S. diversity visa lottery — a program that saved her family from an Ethiopian civil war and famine — and that she earned degrees at Washington College and Notre Dame Law School on American soil. We are not speculating about her worldview. She has put it in writing. And she is one general election away from Congress in a district Republicans haven't won in half a century. We say what needs to be said: believe people when they tell you who they are. In our Top 3 Things You Need to Know, the federal government suspended $867.5 million in Medicaid payments to California and $199 million to Minnesota after CMS reviews flagged high-risk fraud in home health and other services — with CMS Administrator Dr. Mehmet Oz saying the administration is done chasing stolen funds after they've already left the building. Then New Jersey's Democratic Governor Mikey Sherrill revealed that a software glitch allowed 6,000 non-citizens to register to vote in 2023 and 2024, with 400 of them actually casting ballots — a story the governor got ahead of before the DOJ could, and one that makes the case for the SAVE Act more powerfully than any speech. And 100 New York state lawmakers sent letters to the congressional delegation demanding ICE be abolished — the same ICE that exists because Congress decided in 2002 that someone should enforce federal immigration law. We also cover the Trump administration's discovery that the federal government has been sending nearly $100 million per year to dead people — and has now stopped. The fix required asking one basic question: is the person we're sending this check to still alive? That question, apparently revolutionary in Washington, was answered by comparing 885 million federal payments against death records. Government doesn't have a revenue problem. It has a stewardship problem. Our American Mamas Teri Netterville and Kimberly Burleson debate whether they'd go back to old TV over smart TVs — and the answer is a resounding yes, for reasons that go beyond nostalgia. Streaming costs more than cable ever did, smart TVs are listening and tracking everything, you can no longer skip commercials, the content is ideologically slanted, and nobody can find anything. Teri misses Blockbuster. Kimberly misses the DVR. Both agree that horizontal screens make sense because human eyes are horizontal — and that vertical video on phones is a waste of the aspect ratio God gave us. We dig deep into former North Carolina Governor Roy Cooper's early release of 4,234 prisoners following an NAACP lawsuit arguing that incarceration is racist — a number that exceeded the 3,500 agreed to in the settlement and included 99 sex offenders and 24 people convicted of murder. Within five years, 2,412 of those released — a 57% recidivism rate — had committed new crimes and been rearrested, including charges of rape and murder. Among those released and later arrested for murder was the man accused of stabbing Ukrainian immigrant Iryana Zyrutska to death on a Charlotte train. Roy Cooper is now the frontrunner for the North Carolina Senate seat. We say plainly: we should stop rewarding failure. There are people dead in North Carolina today because their governor let violent criminals out of prison to kill again. We also cover the State Department's follow-up report confirming that Cuba is actively helping Iran, Hamas, and other jihadist organizations expand their influence throughout the Western Hemisphere — connecting the dots between communist Cuba's survival dependence on American adversaries and the broader axis of pressure being applied to the United States from multiple directions simultaneously. No single threat defeats us. Together, they create friction, consume resources, divide attention, and complicate national security. All of it from 50 miles off the coast of Florida. For our Bright Spot, Texas Representative Andy Hopper wrote a letter responding to a Mexican senator who demanded that Texas stop enforcing its border laws because migrants were dying trying to cross illegally. Hopper's response is worth reading in full — citing Texas independence, the Alamo, San Jacinto, and the sovereign right of every nation to determine who enters its territory. He also told Mexico directly: we will not accept border policy instruction from a government that has allowed cartels to dominate vast regions of its own territory. Abandoning enforcement would not save lives. It would guarantee the cartels more victims, more profits, and more power. We call it exactly what it is — a bright spot. We also cover New York City's welfare rolls hitting 865,000 cash assistance recipients under Mayor Mamdani — up from a low of 336,000 under Mayor Bloomberg — and note that a democratic socialist does not view this as a crisis. He views it as an achievement. We view it as what happens when you promise free things and then wonder why more people ask for them. We also discuss the case of a 29-year-old woman who died after a ChatGPT conversation that increasingly centered on themes of sacrifice, death, and resurrection — with the AI telling her she would need to die before becoming who she was meant to be. We make the distinction between a mentally healthy person who reads letting your old self die as a metaphor for personal growth, and someone already struggling with depression or delusion who reads it as instruction. AI is a language tool. It produces eloquent, resonant language. It does not know who is reading it. And we close with Scottish runner Josh Kerr breaking the world mile record in London — running the mile in 3 minutes and 42 seconds to break Hicham El Guerrouj's record that had stood since 1999, the longest anyone had held the official world mile record since records began. In front of 60,000 fans, a man ran a mile in the time it takes most of us to find the remote. May your pursuit of happiness bring you joy. Listen now wherever you get your podcasts, visit AmericanGroundRadio.com, and join the conversation at 866-AGR-1776!See omnystudio.com/listener for privacy information.

united states america god tv american new york california texas ai new york city donald trump israel washington mexico colorado government north carolina minnesota congress iran chatgpt mexican republicans ice streaming cuba bernie sanders scottish released democratic ukrainian senators prisoners committed crimes hamas blockbuster medicaid doj state department tvs ethiopian things you need naacp cms hopper alamo michael bloomberg abandoning dsa western hemisphere democratic socialists agr dvr mehmet oz washington college bright spot roy cooper san jacinto notre dame law school north carolina senate north carolina governor roy cooper american ground radio

The Night School - Jame Hopper

Daily Short Stories - Mystery & Suspense

Play Episode Listen Later Jul 18, 2026 16:06

Listen Ad Free https://www.solgoodmedia.com - Listen to hundreds of audiobooks, thousands of short stories, and ambient sounds all ad free!

Decoder Ring | Weapons of Map Destruction

Slow Burn

Play Episode Listen Later Jul 15, 2026 48:55

Most of us use GPS—the Global Positioning System—on a daily basis: to find our location when we're driving, running, shopping, dating, and so much more. But GPS is even more important, and more vulnerable, than you think.In the last few years, GPS interference has been reported all over the world, from war zones to shipping routes to public squares. What was once the fanciful plot of a Bond movie—bad guy manipulates GPS to start World War III—is increasingly plausible. How did the world come to rely so heavily on such an unreliable system?In this episode of Decoder Ring, host Willa Paskin talks to journalist Katherine Dunn, author of the new book Little Blue Dot: How GPS Shaped the Modern World. You'll learn how GPS works, why it was created, how it became so ubiquitous, and why it's now under attack. You'll also hear from Dr. Todd Humphreys, an aerospace engineer who manipulated GPS to trick an $85 million superyacht into following his direction—for science, of course.This episode was written and produced by Max Freedman. It was edited by Willa Paskin and Evan Chung, our supervising producer. Merritt Jacob is Senior Technical Director. Our intern is Phoebe Mulder.If you have any cultural mysteries you want us to decode, email us at DecoderRing@slate.com or leave a message on our hotline at (347) 460-7281.Sources for This Episode Burgess, Matt. “When a tanker vanishes, all the evidence points to Russia,” WIRED, Sep. 21, 2017.Dunn, Katherine. Little Blue Dot: How GPS Shaped the Modern World, Bloomsbury Publishing, 2026.Dunn, Katherine. “How to Hack a Superyacht,” The Walrus, Jun. 13, 2026.Hopper, Nate. “The Thorny Problem of Keeping the Internet's Time,” The New Yorker, Sep. 30, 2022.Hopper, Nate. “The Timekeeper of Ukraine,” The Atlantic, Sep. 21, 2024.Need to set up your Slate Plus feed? If you subscribed through Slate.com, check out our FAQ at slate.com/podcastfaqs for easy instructions. Members subscribed via Apple Podcasts get automatic access—no setup required. Hosted on Acast. See acast.com/privacy for more information.

time internet russia ukraine atlantic bond weapons gps acast hack new yorker destruction wired slate dunn faq modern world world war iii hopper walrus superyachts bloomsbury publishing timekeeper decoder ring global positioning system slate plus senior technical director katherine dunn willa paskin merritt jacob evan chung

Weapons of Map Destruction

Decoder Ring

Play Episode Listen Later Jul 15, 2026 48:53

Most of us use GPS—the Global Positioning System—on a daily basis: to find our location when we're driving, running, shopping, dating, and so much more. But GPS is even more important, and more vulnerable, than you think.In the last few years, GPS interference has been reported all over the world, from war zones to shipping routes to public squares. What was once the fanciful plot of a Bond movie—bad guy manipulates GPS to start World War III—is increasingly plausible. How did the world come to rely so heavily on such an unreliable system?In this episode of Decoder Ring, host Willa Paskin talks to journalist Katherine Dunn, author of the new book Little Blue Dot: How GPS Shaped the Modern World. You'll learn how GPS works, why it was created, how it became so ubiquitous, and why it's now under attack. You'll also hear from Dr. Todd Humphreys, an aerospace engineer who manipulated GPS to trick an $85 million superyacht into following his direction—for science, of course.This episode was written and produced by Max Freedman. It was edited by Willa Paskin and Evan Chung, our supervising producer. Merritt Jacob is Senior Technical Director. Our intern is Phoebe Mulder.If you have any cultural mysteries you want us to decode, email us at DecoderRing@slate.com or leave a message on our hotline at (347) 460-7281.Get more of Decoder Ring with Slate Plus! Join for exclusive bonus episodes of Decoder Ring and ad-free listening on all your favorite Slate podcasts. Subscribe from the Decoder Ring show page on Apple Podcasts or Spotify. Or, visit slate.com/decoderplus for access wherever you listen.Sources for This Episode Burgess, Matt. “When a tanker vanishes, all the evidence points to Russia,” WIRED, Sep. 21, 2017.Dunn, Katherine. Little Blue Dot: How GPS Shaped the Modern World, Bloomsbury Publishing, 2026.Dunn, Katherine. “How to Hack a Superyacht,” The Walrus, Jun. 13, 2026.Hopper, Nate. “The Thorny Problem of Keeping the Internet's Time,” The New Yorker, Sep. 30, 2022.Hopper, Nate. “The Timekeeper of Ukraine,” The Atlantic, Sep. 21, 2024.Need to set up your Slate Plus feed? If you subscribed through Slate.com, check out our FAQ at slate.com/podcastfaqs for easy instructions. Members subscribed via Apple Podcasts get automatic access—no setup required. Hosted on Acast. See acast.com/privacy for more information.

spotify time internet russia ukraine atlantic bond weapons gps acast hack new yorker destruction wired slate dunn faq modern world world war iii hopper walrus superyachts bloomsbury publishing timekeeper decoder ring global positioning system slate plus senior technical director katherine dunn willa paskin merritt jacob evan chung

Decoder Ring - Weapons of Map Destruction

Slate Culture

Play Episode Listen Later Jul 15, 2026 48:53

Most of us use GPS—the Global Positioning System—on a daily basis: to find our location when we're driving, running, shopping, dating, and so much more. But GPS is even more important, and more vulnerable, than you think.In the last few years, GPS interference has been reported all over the world, from war zones to shipping routes to public squares. What was once the fanciful plot of a Bond movie—bad guy manipulates GPS to start World War III—is increasingly plausible. How did the world come to rely so heavily on such an unreliable system?In this episode of Decoder Ring, host Willa Paskin talks to journalist Katherine Dunn, author of the new book Little Blue Dot: How GPS Shaped the Modern World. You'll learn how GPS works, why it was created, how it became so ubiquitous, and why it's now under attack. You'll also hear from Dr. Todd Humphreys, an aerospace engineer who manipulated GPS to trick an $85 million superyacht into following his direction—for science, of course.This episode was written and produced by Max Freedman. It was edited by Willa Paskin and Evan Chung, our supervising producer. Merritt Jacob is Senior Technical Director. Our intern is Phoebe Mulder.If you have any cultural mysteries you want us to decode, email us at DecoderRing@slate.com or leave a message on our hotline at (347) 460-7281.Get more of Decoder Ring with Slate Plus! Join for exclusive bonus episodes of Decoder Ring and ad-free listening on all your favorite Slate podcasts. Subscribe from the Decoder Ring show page on Apple Podcasts or Spotify. Or, visit slate.com/decoderplus for access wherever you listen.Sources for This Episode Burgess, Matt. “When a tanker vanishes, all the evidence points to Russia,” WIRED, Sep. 21, 2017.Dunn, Katherine. Little Blue Dot: How GPS Shaped the Modern World, Bloomsbury Publishing, 2026.Dunn, Katherine. “How to Hack a Superyacht,” The Walrus, Jun. 13, 2026.Hopper, Nate. “The Thorny Problem of Keeping the Internet's Time,” The New Yorker, Sep. 30, 2022.Hopper, Nate. “The Timekeeper of Ukraine,” The Atlantic, Sep. 21, 2024. Hosted on Acast. See acast.com/privacy for more information.

spotify time internet russia ukraine atlantic bond weapons gps acast hack new yorker destruction wired slate dunn modern world world war iii hopper walrus superyachts bloomsbury publishing timekeeper decoder ring global positioning system slate plus senior technical director katherine dunn willa paskin merritt jacob evan chung

Slow Burn - Decoder Ring | Weapons of Map Destruction

Slate Daily Feed

Play Episode Listen Later Jul 15, 2026 48:55

Most of us use GPS—the Global Positioning System—on a daily basis: to find our location when we're driving, running, shopping, dating, and so much more. But GPS is even more important, and more vulnerable, than you think.In the last few years, GPS interference has been reported all over the world, from war zones to shipping routes to public squares. What was once the fanciful plot of a Bond movie—bad guy manipulates GPS to start World War III—is increasingly plausible. How did the world come to rely so heavily on such an unreliable system?In this episode of Decoder Ring, host Willa Paskin talks to journalist Katherine Dunn, author of the new book Little Blue Dot: How GPS Shaped the Modern World. You'll learn how GPS works, why it was created, how it became so ubiquitous, and why it's now under attack. You'll also hear from Dr. Todd Humphreys, an aerospace engineer who manipulated GPS to trick an $85 million superyacht into following his direction—for science, of course.This episode was written and produced by Max Freedman. It was edited by Willa Paskin and Evan Chung, our supervising producer. Merritt Jacob is Senior Technical Director. Our intern is Phoebe Mulder.If you have any cultural mysteries you want us to decode, email us at DecoderRing@slate.com or leave a message on our hotline at (347) 460-7281.Sources for This Episode Burgess, Matt. “When a tanker vanishes, all the evidence points to Russia,” WIRED, Sep. 21, 2017.Dunn, Katherine. Little Blue Dot: How GPS Shaped the Modern World, Bloomsbury Publishing, 2026.Dunn, Katherine. “How to Hack a Superyacht,” The Walrus, Jun. 13, 2026.Hopper, Nate. “The Thorny Problem of Keeping the Internet's Time,” The New Yorker, Sep. 30, 2022.Hopper, Nate. “The Timekeeper of Ukraine,” The Atlantic, Sep. 21, 2024. Hosted on Acast. See acast.com/privacy for more information.

time internet russia ukraine atlantic bond weapons gps acast hack new yorker destruction wired dunn modern world world war iii hopper walrus slow burn superyachts bloomsbury publishing timekeeper decoder ring global positioning system senior technical director katherine dunn willa paskin merritt jacob evan chung

Decoder Ring - Weapons of Map Destruction

Slate Daily Feed

Play Episode Listen Later Jul 15, 2026 48:53

Most of us use GPS—the Global Positioning System—on a daily basis: to find our location when we're driving, running, shopping, dating, and so much more. But GPS is even more important, and more vulnerable, than you think.In the last few years, GPS interference has been reported all over the world, from war zones to shipping routes to public squares. What was once the fanciful plot of a Bond movie—bad guy manipulates GPS to start World War III—is increasingly plausible. How did the world come to rely so heavily on such an unreliable system?In this episode of Decoder Ring, host Willa Paskin talks to journalist Katherine Dunn, author of the new book Little Blue Dot: How GPS Shaped the Modern World. You'll learn how GPS works, why it was created, how it became so ubiquitous, and why it's now under attack. You'll also hear from Dr. Todd Humphreys, an aerospace engineer who manipulated GPS to trick an $85 million superyacht into following his direction—for science, of course.This episode was written and produced by Max Freedman. It was edited by Willa Paskin and Evan Chung, our supervising producer. Merritt Jacob is Senior Technical Director. Our intern is Phoebe Mulder.If you have any cultural mysteries you want us to decode, email us at DecoderRing@slate.com or leave a message on our hotline at (347) 460-7281.Get more of Decoder Ring with Slate Plus! Join for exclusive bonus episodes of Decoder Ring and ad-free listening on all your favorite Slate podcasts. Subscribe from the Decoder Ring show page on Apple Podcasts or Spotify. Or, visit slate.com/decoderplus for access wherever you listen.Sources for This Episode Burgess, Matt. “When a tanker vanishes, all the evidence points to Russia,” WIRED, Sep. 21, 2017.Dunn, Katherine. Little Blue Dot: How GPS Shaped the Modern World, Bloomsbury Publishing, 2026.Dunn, Katherine. “How to Hack a Superyacht,” The Walrus, Jun. 13, 2026.Hopper, Nate. “The Thorny Problem of Keeping the Internet's Time,” The New Yorker, Sep. 30, 2022.Hopper, Nate. “The Timekeeper of Ukraine,” The Atlantic, Sep. 21, 2024. Hosted on Acast. See acast.com/privacy for more information.

CLCC REVIVAL WK#3 / ORPHAN SPIRIT / PASTOR NEIL HOPPER

lesson hopper cobol grace hopper

Play Episode Listen Later Jul 14, 2026 56:31 Transcription Available

Victory begins when we deal with an orphan mentality, often referred to as an orphan spirit. When we receive the love of the Father it changes everything. We can't do enough to earn God's love because His love is freely given. But insecurity, fear and self-rejection can be signs of a deeper spiritual conclusion that you must work harder to earn God's love and favor. Nothing could be further from the TRUTH and that TRUTH will set you free. https://www.instagram.com/cedarlake_cc/

god father truth victory revival hopper orphan spirit pastor neil clcc

The Nanosecond Lesson

Insights Into Things

Play Episode Listen Later Jul 13, 2026 45:45 Transcription Available

Season 2, Episode 1 explores Grace Hopper's famous nanosecond demonstration and uses it to explain latency, systems design, and why distance and architecture matter for performance. The episode also traces Hopper's work on automatic programming, Flow-Matic, and the push toward COBOL and standards—showing how she turned software into teachable, maintainable infrastructure.

Rev. Dr. Harden Hopper Preaches on Romans 8:1-11

romans rev james harden romans 8 hopper preaches first united methodist church

Play Episode Listen Later Jul 12, 2026 16:36

Rev. Dr. Harden Hopper preaches on Romans 8:1-11.First United Methodist Church of MariettaGiving link: https://onrealm.org/mariettafumc/-/form/give/nowChurch website: ⁠⁠⁠⁠⁠⁠⁠⁠https://www.mariettafumc.org/

CLCC REVIVAL (EXPECTATION) / PASTOR NEIL HOPPER

Play Episode Listen Later Jul 6, 2026 59:51 Transcription Available

Everyone lives with expectation. Some expect the best while others expect the worst. Some people expect nothing at all because disappointment has robbed them of hope. Let God's Word challenge you to raise your level of expectation. https://www.instagram.com/cedarlake_cc/

god revival expectation hopper pastor neil clcc

#223 Physical's Future

Play Episode Listen Later Jul 4, 2026 25:32

Hopper asks ChatGPT about the future of collecting physical video games.

chatgpt hopper

REVIVAL PT#1 (RENEWAL) / PASTOR NEIL HOPPER

revival renewal hopper pastor neil

Play Episode Listen Later Jul 3, 2026 37:53 Transcription Available

Every believer needs times of renewal and refreshing. We are praying for spiritual awakening and revival for our congregation, community and country. We hope you will join us. Listen and be challenged by Pastor Neil Hopper's message on a "Summer of Revival" https://www.instagram.com/cedarlake_cc/

Why Hardware-Software Co-Design Is AI's Real 100x: Dylan Patel of SemiAnalysis

Training Data

Play Episode Listen Later Jun 30, 2026 70:14

Dylan Patel, founder of SemiAnalysis, argues the biggest gains in AI don't come from faster chips, they come from software-hardware co-design. Optimizing the model, the kernels, and the silicon together turns a 2x here and a 2x there into 100x. He explains why DeepSeek's experts were shaped for Nvidia's Hopper (and why TPUs struggle to run it), why OpenAI's sparser models and Anthropic's denser ones pull them toward different hardware, and why the so-called CUDA moat was never really about CUDA. Dylan breaks down InferenceX, his living benchmark that runs the latest models on over $50M of donated hardware daily, tracking a roughly 60x annual drop in cost per unit of quality. He makes the case that inference will be a bigger market than oil, that the compute crunch persists because models expand the value of useful work faster than compute grows, and why Jensen Huang is bankrolling neoclouds to engineer a multipolar world. Hosted by Shaun Maguire and Sonya Huang, Sequoia Capital

ai openai optimizing nvidia patel anthropic hopper 50m 100x jensen huang sequoia capital cuda codesign hardware software

Iran, Conspiracies, and the Crisis of Discernment with Phil Hopper

The Busy Mom

Play Episode Listen Later Jun 29, 2026 67:11

What happens when Christians start chasing conspiracies instead of truth? Pastor Phil Hopper joins me for an honest conversation about Iran, the growing divide inside the conservative movement, the rise of the "woke right," and why biblical discernment matters now more than ever. We also talk about Israel, end times, and how followers of Christ can stand firm without living in fear. If you've been trying to make sense of the headlines, this episode will challenge and encourage you.Chapter Medicare AdvisorsChoosing Medicare shouldn't feel overwhelming. Chapter compares every available Medicare plan to help you find the coverage that best fits your needs—not someone else's. Learn more at http://askchapter.org/realhelp .Show mentions: http://heidistjohn.com/mentionsWebsite | http://heidistjohn.comSupport the show! | http://donorbox.org/donation-827Rumble | rumble.com/user/HeidiStJohnYoutube | / @‌heidistjohnpodcast Instagram | @‌heidistjohnFacebook | Heidi St. JohnX | @‌heidistjohnFaith That Speaks Online CommunitySubmit your questions for Fan Mail Friday | http://heidistjohn.com/fanmailfriday

jesus christ israel crisis christians iran conspiracies medicare discernment hopper fan mail friday

#222 Super Mario Hangtabs

Nebraska Athletics Podcast

Play Episode Listen Later Jun 29, 2026 17:16

Hopper goes over the BIG Black Box Mario sealed Hangtab auctioned off at Goldin June, 2026, then takes a look at the Black Box Guide, POPs and comps.Image owned by GoldinEdu. & Entertainment

pops super mario hopper

Sports Nightly - Matt Hopper

Play Episode Listen Later Jun 23, 2026 19:24

Kyle Crooks sits down with a member of the 2026 Nebraska Athletics Hall of Fame Class, Matt Hopper. The two chat about his storied college baseball career in Lincoln, the memories of making multiple College World Series, plus much more!

hopper college world series nightly

From $50 In The Bank To Multi-Million-Dollar Exit

App Masters - App Marketing & App Store Optimization with Steve P. Young

Play Episode Listen Later Jun 20, 2026 89:29

What does it really take to raise money, find product-market fit, and build a startup worth acquiring in 2026?In this episode, we are joined by Susan Ho, an entrepreneur, startup advisor, and former founder who built and exited travel-tech startup Journy to Hopper. shares the lessons she learned building Journy from the ground up, raising capital, navigating the founder journey, and ultimately selling the company to Hopper.We dive into the realities of startup fundraising, why founders get stuck in the "need traction to raise money, need money to get traction" cycle, and how today's founders can leverage personal branding and social media to accelerate growth.If you're a founder struggling with growth, fundraising, or the emotional rollercoaster of building a business, this is a must-watch.A candid conversation filled with real lessons, mindset shifts, and the kind of "mental therapy" every entrepreneur needs from time to time.

state record bank web android exit ios million dollars merchant multimillion hopper dtc multi million dollar appsflyer journy

316: Three-Cornered Alfalfa Hopper: Monitoring, Models, and Management

Sustainable Winegrowing with Vineyard Team

Play Episode Listen Later Jun 18, 2026 29:28

Three-cornered alfalfa hopper (TCAH) is the only confirmed insect vector of grapevine red blotch virus in Vitis vinifera, yet many growers first realize they have the pest only after spotting petiole girdling in the vineyard. Cindy Kron, North Coast IPM Advisor at UC ANR, shares findings from two years of weekly sweep-net monitoring at Oakville Research Station that revealed TCAH adults are present well before bud break, suggesting grapevines are not their preferred host. She explains the insect's life stages, why legumes serve as key feeding and reproductive hosts, and why detecting early instars remains a major challenge for vineyard IPM. Cindy also discusses how degree-day models may help growers better time tillage to reduce TCAH populations and limit grapevine red blotch risk. Resources: 71: New Techniques to Detect Grapevine Leafroll Disease 131: Virus Detection in Grapevines Can a pesky treehopper be foiled because its growth is regulated by temperature? Cindy Kron How to use a model for reduction of three-cornered alfalfa hopper in vineyards Identification of Nonhost Cover Crops of the Three-Cornered Alfalfa Hopper (Spissistilus festinus) Use of Ground Covers to Control Three-Cornered Alfalfa Hopper, Spissistilus festinus (Hemiptera: Membracidae), and Other Suspected Vectors of Grapevine Red Blotch Virus Weather Models and Degree Days Support the Podcast: Make a Donation Vineyard Team Programs: Juan Nevarez Memorial Scholarship - Help students from vineyard families pursue higher education Online Courses - Earn DPR and CCA hours with expert-led sustainability trainings SIP Certified - A trusted third-party certification proving your sustainable practices with science-backed standards Sustainable Ag Expo - Join top experts at the premier winegrowing event of the year Vineyard Team Membership - Connect with a community advancing sustainable winegrowing

management models monitoring identification hopper cca cornered alfalfa ipm podcast make new techniques vitis uc anr

THURSDAY RUB | Isaac the Bandwagon Hopper, Garry Hocking Crime Stopper, Andrew Embley's Potty Mouth

Triple M Rocks Footy AFL

Play Episode Listen Later Jun 18, 2026 81:39

The team is assembled ahead of Freo and the Cats, but Isaac and K-Mac are squaring off over last night's State of Origin result. Luckily, Isaac has come ready with several guernseys of winning teams. With Toyota about to make another Legendary Moment commercial, our team nominated their own highlights, then Fremantle assistant coach Jaymie Graham joins the show pre-game. After an astonishing story out of Geelong today, Cats legend Garry Hocking calls in to explain what he happened as he attempted to make a citizen's arrest, then Luke Beveridge had some jabs for Jordan Lewis. As Isaac continues to swap into winning guernseys, Jay Z rolls his Chief's Agenda out, and K-Mac's Royal Commission looks at people who might've gone off too soon. More footy news around Melbourne and Essendon is followed by Geelong GM of Footy Andrew Mackie. Then we hear Isaac's Penthouse or Outhouse, and a round of Unpopular Opinions. Finally, Andrew Embley joins the crew from the West, but we're interested in a little moment he had on air a few weeks ago.See omnystudio.com/listener for privacy information.

Lunch with the Clydesdale & Cowboy! Wednesday Show! - Time to Fill the Hopper with Balls!!!

The Clydesdale, Fitness & Friends

Play Episode Listen Later Jun 17, 2026 41:10 Transcription Available

Presenting Sponsor Thirdzy! https://thirdzy.com/JAZZYPromotion Code for 15% off: JAZZYSupport Carolyne with the purchase of your CrossFit Games Tickets, Use Code cfgprevost10 at checkoutEveryday we take a break from the busy work day to catch our breath, hang out with friends and talk about the world of Sports, Entertainment and specifically CrossFit. Today we talk about how to mispronounce all the words, The Hopper is back and what does that mean for the Games and Excitement is at an all time high!

sports games fill balls lunch crossfit cowboy showtime excitement ranch hopper crossfit games streamyard functional fitness crossfit open clydesdale dave castro crossfit semifinals bruce edwards

Episode 152 - Ruthanna Halprin Hopper

Love Is The Author

Play Episode Listen Later Jun 15, 2026 101:14

Ruthanna Halprin Hopper is an American artist known for blending visual arts, dance, and somatic psychology. She is the daughter of actor Dennis Hopper and dancer Daria Halprin, and the granddaughter of renowned environmental designer Lawrence Halprin and dance pioneer Anna Halprin, creators of the world renown Halprin Method. Jaymee and Ruthanna discuss shadow work, synchronicities, collective soul being born in group settings, transparency about her father's mastery of reinvention, and more. “In an archetypal sense, it's taken me a long time to come home”. LITA PODCAST: hosted, produced, mixed, and recorded by Jaymee Carpenter. COVER PHOTO: Ekaterina IzmestievaOPENING SONG: Gangstalean by JJ RAM (Jaymee)CLOSING SONG: Ocean Of Beauty by Earthtones Music & Sheela BringiInterested in Trauma Counseling/Mentorship with Jaymee?email: lacee@loveistheauthor.com to set up a free consultation,or visit: www.loveistheauthor.com/mentorship SPONSORS: YERBA MADRE ⁠⁠⁠⁠⁠⁠www.yerbamadre.com⁠RAUM GOODS ⁠⁠⁠⁠www.raumgoods.com⁠⁠⁠⁠BOSSANOVA SOAP www.bossanovasoap.com⁠⁠⁠⁠TOTALLY BLOWN ⁠⁠⁠⁠⁠⁠⁠www.totallyblown.us⁠⁠⁠⁠⁠⁠⁠INDIAN LODGE ROAD ⁠⁠⁠⁠www.indianlodgeroad.com⁠⁠⁠⁠THiS SHOW is a LABOR of LOVE. PLEASE SUPPORT IT: ⁠⁠⁠⁠www.patreon.com/loveistheauthorpodcast⁠⁠⁠⁠FAN CONTACT: lacee@loveistheauthor.comON INSTAGRAM: @loveistheauthor / @unconventionalgardene

love american labor hopper dennis hopper this show anna halprin halprin lawrence halprin daria halprin

SECRETS OF THE SERCRET PLACE / PASTOR NEIL HOPPER