Podcasts about mdps

The Storm Skiing Journal and Podcast is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Welcome to the Storm's short-form, news-focused podcast. Don't worry, I still write long-form newsletters. Paid subscribers can leave a comment in the article below, or by joining The Storm's chat (also below). I'll respond to some comments in the next episode, which is scheduled for Wednesday, Sept. 16.Transcript updateFollowing up on yesterday's feedback, there are now two transcripts available for each podcast:* Substack's transcript: This transcript is auto-generated when I upload the podcast file to Substack, which is the platform that hosts stormskiing.com. This one is cool because you can click on any block of text and the audio will jump to that point in the conversation. Click the “transcript” button above to access. * Zoom's transcript: Zoom automatically generates this transcript as I record. It should be more or less identical to Substack's transcript, but with names attached to each quote block. The downside of this transcript is that you can't teleport to that point in the pod by clicking on the text block, as you can with Substack's. Another downside: the timestamps are off, since I typically start recording prior to the start of the podcast. I'll work on cleaning this up.So until the robots get smarter, we're stuck with two different transcripts. Both are auto-generated, and I don't proofread either one, as that would be the fastest way to make sure my quick-turnaround, up-to-date podcast turns into a slow-turnaround, out-of-date one. My hope is that Substack will soon hire or invent or weld from raw iron a robot that's smart enough to know who's speaking when. I'll just keep following the robots around and using their hyper-intelligence until they decide to kill me.The Zoom transcript (click “transcript” above for the Substack transcript, which will zoom to any point in the video when you click on the associated text block; timestamps below DO NOT MATCH THE VIDEO)Stuart Winchester: Welcome to The Storm! I'm your host, Stuart Winchester. It is Tuesday, September 15th, 2026, and I am very excited today to bring you a conversation with the General Manager of Monarch Mountain, Colorado, Chris Haggerty. Now, I've never featured Monarch either on the long-form pod or on the short-form pod.00:09:57.000 --> 00:10:11.000Stuart Winchester: There's never really a reason behind that, other than this is just kind of when it happened. If you run an area, a ski area anywhere in America, I do want to talk to you. I've just never had the opportunity to feature Monarch before, so we will talk about Monarch in just a little bit here.00:10:11.000 --> 00:10:27.000Stuart Winchester: Part of the reason why we're talking about Monarch is because they were one of two Western Mountains to this year join Snow Operating's Snow Pass. Now, the Snow Pass, if you missed the news, it works a lot like the Indy Pass or the Mountain Collective Pass, where you get two days each.00:10:27.000 --> 00:10:37.000Stuart Winchester: at a set number of ski areas. In this case, it's planned to be 14 ski areas for the 2026-27 ski season. And…00:10:37.000 --> 00:10:45.000Stuart Winchester: That product went on sale last Tuesday. I had Snow Operating CEO Joe Heschen on the podcast to talk through it with us.00:10:45.000 --> 00:11:01.000Stuart Winchester: Couple days later, I got an email from Snow Operating saying, Hey, guess what? Our first round of Snow Operating and Snow Triple Play, which are two different products, and I'll explain the difference in a moment, sold out. So we're taking it off sale. We're putting it back on sale next Monday, September 21st.00:11:01.000 --> 00:11:13.000Stuart Winchester: at 10 a.m. Eastern. It will be $349.99 for Snowpass, and $184.99 for Snow Triple Play. Still a pretty good deal. Those products at full price.00:11:13.000 --> 00:11:24.000Stuart Winchester: the prices that Snow Partners gave me, and they anticipate selling both products through Christmas Eve, December 24th, just like they did last year with the first season of the Triple Play.00:11:24.000 --> 00:11:40.000Stuart Winchester: They, the final prices they anticipate will be 3 99 99 for snow pass and 1 99 99 for snow triple play. So you get a nice little discount, particularly with the snow pass, $50 off if you buy it in the next round. So let's talk a little bit about the snow pass because.00:11:40.000 --> 00:11:43.000Stuart Winchester: When this product debuted.00:11:43.000 --> 00:11:53.000Stuart Winchester: I… I've been saying for years, there's 4 national ski passes, right? There's obviously Epic and Icon, and those have some season pass unlimited tiers at certain mountains.00:11:53.000 --> 00:12:08.000Stuart Winchester: And a limited number of days at other partner mountains. And then you have Indian Mountain Collective, which essentially worked the same way, 2 days at each ski area that is on their roster. Uh, Mountain Collective is more expensive, tends to be your bigger, more high-end resorts, Snowbird, Jackson Hole.00:12:08.000 --> 00:12:23.000Stuart Winchester: Uh, Sun Peaks up in Canada. And Indy tended to be the lesser-known ski areas, but still some really dynamite stuff there, and a lot more ski areas. Mountain Collective always has around 30, and Indy is more like 250, not counting their cross-country.00:12:23.000 --> 00:12:36.000Stuart Winchester: ski areas. So, I always thought there was probably room for a fifth national pass, but I wasn't sure, and as time went on, and more and more mountains joined multi-mountain passes.00:12:36.000 --> 00:12:37.000Stuart Winchester: I…00:12:37.000 --> 00:12:41.000Stuart Winchester: I thought, okay, maybe there's not room for a fifth national pass. Maybe everyone who's…00:12:41.000 --> 00:12:44.000Stuart Winchester: who wants to join a pass has, because I know Indy, uh.00:12:44.000 --> 00:12:58.000Stuart Winchester: Now, I'm not saying they don't have standards, but they'll let just about anybody on the pass just to help them get that exposure and that national marketing. But Snow Operating, the outfit that owns Mountain Creek in New Jersey and the indoor big snow ski area in New Jersey, they launched.00:12:58.000 --> 00:13:15.000Stuart Winchester: their Snow Pass, and it's a pretty good roster. Here, I'm gonna share my screen for those of you watching on video, and if you're not watching on video, don't worry, I'm gonna talk through it. So, here's the Snow Triple Play East roster, and here's the Snow Pass roster. Now, the Snow Pass has 14 ski areas on it.00:13:15.000 --> 00:13:18.000Stuart Winchester: And you get 2 days at each ski area.00:13:18.000 --> 00:13:22.000Stuart Winchester: The Snow Triple Play has…00:13:22.000 --> 00:13:27.000Stuart Winchester: 23 total ski areas on it, and you get 3 days total.00:13:27.000 --> 00:13:43.000Stuart Winchester: across all 23 ski areas. That's why it's so much less expensive. The most days you can use at any ski area is 2, so you can't buy this as a 3-pack Force A Gore Mountain, alright? So, that's the first important.00:13:43.000 --> 00:13:59.000Stuart Winchester: distinction is that Snow Triple Play is a much more limited basket of days. You also don't want to assume that they have the same roster. Just because something's on Snow Pass doesn't mean it's on Snow Triple Play, and vice versa. Snow Triple Play right now has a pretty…00:13:59.000 --> 00:14:15.000Stuart Winchester: Big roster that I wish was on the, uh, Snow Pass, because it has Jiminy Peak in, uh, Massachusetts, and also Cranmore in New Hampshire, which is jointly owned. It has Platykill up in the New York Catskills.00:14:15.000 --> 00:14:30.000Stuart Winchester: It has all these summit ski areas around Quebec. It has Bromont, one of the great eastern townships ski areas. Uh, so that's Snow Triple Play. Snow Triple Play does not have anything in the west. It's all east. It's all east coast, either eastern Canada or eastern New York.00:14:30.000 --> 00:14:45.000Stuart Winchester: Uh, Snow Pass has… it's a pretty interesting roster, and it's a great roster for a very particular type of skier. Now, it does have two ski areas in the West, uh, Snow King, which has been controversial, I'm gonna set that aside, and Monarch, which we'll talk about in a moment.00:14:45.000 --> 00:15:00.000Stuart Winchester: The core of the Snow Pass, and the reason why it makes sense, is because of the south to north line of ski areas that it offers. So if you buy the Snow Pass, again, goes back on sale for $349.99 next Monday, September 21st.00:15:00.000 --> 00:15:05.000Stuart Winchester: You would get two days each at Bel Air, which is in New York's Catskills.00:15:05.000 --> 00:15:07.000Stuart Winchester: Gore.00:15:07.000 --> 00:15:16.000Stuart Winchester: higher up in the Adirondacks, and Whiteface, which are all owned and run and operated by the state, all fantastic ski areas. So it's 2 days at each of those, that alone would be worth $350.00:15:16.000 --> 00:15:33.000Stuart Winchester: Then you add in the fact that you get two days at Mountain Creek, which is right there in New Jersey, uh, two days at the indoor snow dome, big snow American dream. Uh, if you have kids, two days at Butternut and two days at Mount Southington in Connecticut. Uh, Butternut is in Massachusetts. They're both day drive distance from New York City.00:15:33.000 --> 00:15:42.000Stuart Winchester: Let them wander around, let them explore. They're both, uh, especially Butternut. It's very up-to-date, very modern lifts, very, very well-run ski area. So…00:15:42.000 --> 00:15:45.000Stuart Winchester: You have a really great little, you know.00:15:46.000 --> 00:16:03.000Stuart Winchester: 5, 6, 7 ski areas that really make sense. Now, Ski Martok doesn't really make sense. It's all the way up in the middle of nowhere in the, in the Atlantic provinces in, in Canada. Uh, and, and Pleasant Mountain makes less sense. It's, it's up in Maine and, uh, harder to get to from, from the folks who would likely buy this as.00:16:03.000 --> 00:16:13.000Stuart Winchester: But what you have to ask yourself is, okay, the snow pass, that's a pretty good deal, right? It's $349.99 for 2 days each of 14 mountains. Then you look at Indy.00:16:13.000 --> 00:16:16.000Stuart Winchester: And Indy Pass is 254.00:16:16.000 --> 00:16:34.000Stuart Winchester: uh, alpine ski areas, and it just went off sale for $429 for the base pass. So, it would be 2 days each at all of these, and the… the plus pass, the ND plus pass, was $469. It had no blackouts. Now, the other thing to know about Snow Triple Play.00:16:34.000 --> 00:16:50.000Stuart Winchester: and snow passes, there are a lot of blackouts, okay? So, a lot of these are holidays, you probably expect that. Uh, you know, here I see Big Snow, Oak Mountain, Plattekill, a lot of holiday blackouts, Butternut holiday blackouts, Gore, uh, Bel Air. Bel Air, though, you want to stop.00:16:50.000 --> 00:17:06.000Stuart Winchester: Bel Air is blacked out every Saturday and Sunday. That's a big deal, because when do you probably want to go to Bel Air if you're someone skiing around with your little kids? Probably on a Saturday or Sunday. So, you want to be really careful and really look at these, uh, these blackout dates before you pull the trigger.00:17:06.000 --> 00:17:09.000Stuart Winchester: on the snow pass. So…00:17:09.000 --> 00:17:14.000Stuart Winchester: So if you have a snow pass, I I'm calling it.00:17:14.000 --> 00:17:18.000Stuart Winchester: A national pass, kind of, uh…00:17:18.000 --> 00:17:24.000Stuart Winchester: out of anticipation, right? It's more of a, I think they'll get there, sort of designation, because right now.00:17:24.000 --> 00:17:40.000Stuart Winchester: it's not really a national pass, right? You have one small skier in Wyoming, one small skier in Colorado. Is someone who lives in New York's metro really going to travel to Colorado just to ski that because they have a snow pass, right? Because it's not cheap to get up to Colorado.00:17:40.000 --> 00:17:56.000Stuart Winchester: And there's a lot of much bigger mountains out there. So… so I think that Snow Pass, much like Indy did when it started in 2019, will grow into a national… a true national pass. Uh, for now, just based on the reputation of the operators, uh, the fact that I think that they will be able to grow this thing, the fact that I…00:17:56.000 --> 00:18:11.000Stuart Winchester: know their relationships with folks all around the country, and how much folks like working with them, I am assuming that this pass will grow up, and if it does, Monarch will be the OG Western partner. So, let's talk more about Monarch now, and I'm gonna bring.00:18:11.000 --> 00:18:13.000Stuart Winchester: Chris Haggerty in.00:18:14.000 --> 00:18:17.000Stuart Winchester: And we will, uh… Chris, how you doing?00:18:17.000 --> 00:18:20.000Chris Haggerty: I'm doing well today, Stuart. How are you?00:18:19.000 --> 00:18:37.000Stuart Winchester: I'm doing great. All right, you passed the sound check, so I'm going to give you an intro here. Joining us now is the general manager of Monarch Mountain, Colorado. Monarch has six chairlifts serving 1,017 acres of terrain, including the 377-acre no-name expansion that opened last winter.00:18:37.000 --> 00:18:44.000Stuart Winchester: Monarch's 10,727-foot base elevation is the third highest in American skiing.00:18:44.000 --> 00:18:59.000Stuart Winchester: Prior to taking the top job at Monarch in 2024, he spent 21 years as general manager of Mountain Creek, New Jersey. Chris Haggerty is my guest. Chris, welcome to The Storm. Awesome to connect with you. I've really been wanting to spotlight Monarch for a long time now, and I'm so hyped we get to.00:19:00.000 --> 00:19:10.000Chris Haggerty: Yeah. Thanks for having me, and I'm excited to be here and chat Monarch. It's a special place, so excited to share some discussion about it.00:19:10.000 --> 00:19:23.000Stuart Winchester: So, the New Jersey to Colorado pipeline is maybe not necessarily a well-trod one, although there are a surprising amount of IntraWest vets working in the ski industry because of its relationship with IntraWest, but talk to us about…00:19:24.000 --> 00:19:40.000Stuart Winchester: Getting to Monarch, and you were taking over for Randy Stroud, super experienced guy, he was retiring, uh, he'd been there 23 years, a decade as GM, and he stayed on, I believe, for, for a year. So, talk to us about going from New Jersey, where they have a lot of snowmaking, because they have to.00:19:39.000 --> 00:19:41.000Chris Haggerty: Mmhm.00:19:40.000 --> 00:19:47.000Stuart Winchester: to Monarch, where they have no snowmaking, and just what that adjustment period was like, and how you settled into the role.00:19:48.000 --> 00:20:07.000Chris Haggerty: Yeah, just a quick story is that, you know, as you mentioned, I was at Mountain Creek for a long time, and, you know, really where I found my love for the ski industry, and after college, I had taken a trip out to Colorado and got my first taste of, like.00:20:07.000 --> 00:20:12.000Chris Haggerty: the Rockies, right? So it's always been something in the back of my mind. Um…00:20:12.000 --> 00:20:28.000Chris Haggerty: And I ended up stumbling upon Monarch. I hadn't been to Monarch before, and I was like, oh, this place looks pretty cool. It definitely screamed the ski independent. I definitely wanted to.00:20:28.000 --> 00:20:50.000Chris Haggerty: you know, if I was going to relocate, wanted to find an area where the resort and the local community both had that kind of just feeling of community and stuff. So a monarch was it. And I was blessed to step in after Randy Stroud, and we shared a season together. So I got to learn a lot of institutional knowledge from him.00:20:50.000 --> 00:20:58.000Chris Haggerty: Um, but also bring my viewpoints coming from New Jersey in an area where, you know, we're not…00:20:58.000 --> 00:21:15.000Chris Haggerty: maybe blessed with natural snow. Um, so last year was, like, a challenging winter, and, um, I was ready for it, because I was like, man, 120 inches, like, if you gave me that in New Jersey, we'd be celebrating for years, right? So, um, you know, just kind of a unique perspective, but…00:21:15.000 --> 00:21:21.000Chris Haggerty: um, definitely happy to be here, and just pumped to keep the Monarch kind of spirit going forward.00:21:21.000 --> 00:21:37.000Stuart Winchester: So you've been there for a couple years, one full year as GM, and one of the… there's been tons of stuff coming out of Monarch, it's amazing for a small ski area, which is why I wanted to get you on. So you recently joined the Snow Pass, one of two Western partners to join the Snow Pass, which gives pass holders.00:21:32.000 --> 00:21:34.000Chris Haggerty: Thank you.00:21:37.000 --> 00:21:54.000Stuart Winchester: Two days at Monarch. I don't think there's blackouts on that, but correct me if I'm wrong now. Now, Multimon Pass has been around for years and almost everybody in Colorado is on either Epic or Icon if they're large and if they're small, a lot of them have joined Indy. So the option has been there for years.00:21:54.000 --> 00:22:00.000Stuart Winchester: Why was it finally time for Monarch to join a multi-mountain pass, and why the snow pass?00:22:02.000 --> 00:22:15.000Chris Haggerty: Yeah, two parts there. I think on the, I mentioned it before, we're ski independent as our tagline, right? Part of independence is like the definition is freedom.00:22:10.000 --> 00:22:11.000Stuart Winchester: Mmhm.00:22:15.000 --> 00:22:19.000Chris Haggerty: Right? Like, you make your own choices. You do what you wanna do.00:22:19.000 --> 00:22:31.000Chris Haggerty: we've been approached in the past for IndyPass, and, you know, it wasn't right for us because it kind of went against our spirit of being independent. Like, we want to be able to work with who we…00:22:31.000 --> 00:22:43.000Chris Haggerty: who we choose at any given moment, um, and that just isn't something that, um, is available through the Unipass, so… so we've passed on that. Um, we were even approached about it this year. Um…00:22:43.000 --> 00:23:00.000Chris Haggerty: As far as Snowpass goes, they respect that. They're all about, hey, you want to work with XYZ Pass, go for it. And obviously, I've spent some time working with the Snow Partners group.00:23:00.000 --> 00:23:05.000Chris Haggerty: You know, from my perspective, that group is all about, um…00:23:05.000 --> 00:23:17.000Chris Haggerty: you know, rising tides lifts all boats, right? And, um, you know, for me personally, they've opened doors for me, right? And helped me throughout growing in my career, so, um…00:23:39.000 --> 00:23:40.000Chris Haggerty: We're down.00:23:40.000 --> 00:23:52.000Stuart Winchester: I want to share, because you've sort of created your own… I'm sharing my screen now, if you can see that, and hopefully the folks watching can as well, and if you're listening, I'll talk through it. So, Monarch has…00:23:45.000 --> 00:23:46.000Chris Haggerty: Thank you.00:23:52.000 --> 00:24:07.000Stuart Winchester: not joined a pass, necessarily, but you've kind of built your own pass, and what I have on the screen right now is a list of Monarch's reciprocals, and it's a pretty amazing pass all in and of itself, so if you buy a Monarch season pass, which I believe now is, what, $6.59, less if you renew.00:24:07.000 --> 00:24:24.000Stuart Winchester: It comes with a lot of reciprocal days. Some of the big ones that people probably are really attracted to, 3 days at A-Basin, 3 days at Copper. Now there's blackouts on that, but still, it's freaking Copper Mountain. Uh, 3 days at Purgatory. And then, uh, 3 days each at all of the Mountain Capital Partners-owned skiers, which Purgatory.00:24:07.000 --> 00:24:08.000Chris Haggerty: Mm-hmm.00:24:14.000 --> 00:24:16.000Chris Haggerty: I'll do that.00:24:24.000 --> 00:24:25.000Chris Haggerty: Yes.00:24:24.000 --> 00:24:39.000Stuart Winchester: Also, Arizona Snowball. And there's a bunch more. I don't want to list them all, given our time constraints here today. But talk about this reciprocal program, because essentially, you're saying you're going to give pass holders of all these mountains.00:24:33.000 --> 00:24:34.000Chris Haggerty: Yes.00:24:37.000 --> 00:24:38.000Chris Haggerty: Sure.00:24:39.000 --> 00:24:46.000Stuart Winchester: free visits to Monarch in exchange for your pass holders getting free visits there, so…00:24:41.000 --> 00:24:42.000Chris Haggerty: Oh.00:24:43.000 --> 00:24:44.000Chris Haggerty: Yes.00:24:46.000 --> 00:24:52.000Stuart Winchester: whatever you did or didn't like about IndyPass, it does pay per visit, so what was attractive about.00:24:47.000 --> 00:24:49.000Chris Haggerty: Just a second.00:24:52.000 --> 00:24:57.000Stuart Winchester: this, and why did it make sense to do these free visit exchanges with other ski areas?00:24:57.000 --> 00:25:15.000Chris Haggerty: We look at it as other skiers are not our competition, right? There's 350 million people roughly in the United States. Less than 5% actually ski or snowboard, right? So there's a lot of people out there that just need to be introduced into the sport.00:25:15.000 --> 00:25:17.000Chris Haggerty: Um…00:25:17.000 --> 00:25:23.000Chris Haggerty: rather than trying to fight for visits between resorts. So… so one of the things is just…00:25:23.000 --> 00:25:39.000Chris Haggerty: How do we get more people skiing and snowboarding, which is obviously a huge conversation you could talk about forever. But with this one, it's just like, hey, these are other resorts that are interested in also giving their pass holders as well as ours.00:25:55.000 --> 00:26:10.000Chris Haggerty: allow our pass holders to try each out while still knowing that, like most people that have a copper pass, they're not migrating to Monarch because they came here for two days.00:26:10.000 --> 00:26:28.000Chris Haggerty: they're bought a Copper Pass for a reason, they love that, and they're like, yeah, if I could get two days over at Monarch, cool. Same thing with our Monarch guests, like, they… they… they love Monarch, they want most of their days to be here, and if we could break it up for them, um, you know, a couple times a year, or they could go to A-Basin and Copper, then.00:26:16.000 --> 00:26:17.000Stuart Winchester: Yeah. Okay.00:26:28.000 --> 00:26:37.000Chris Haggerty: That adds a lot of value to them and gives them that flexibility to move around and go where it works for them.00:26:37.000 --> 00:26:52.000Stuart Winchester: You know, Monarch is small by Colorado standards, but it always surprised me to see that it actually does have a pseudo national following. And I get emails from people all over the country that say they buy the Monarch Pass because it gives them all those reciprocals and they come out and they like to support the small ski area.00:26:52.000 --> 00:26:59.000Stuart Winchester: I want to get your take on something here, Chris. I'm going to share my screen again, and I want to share a chart that I made.00:26:59.000 --> 00:27:14.000Stuart Winchester: Uh, so, so this is a chart of Monarch's skier visits since 1997, up to 2024 to 25, right? And I've compiled these from all kinds of different places, and I have a little bit of a gap there. But as you, as you see right here, they've been going steadily up, right?00:27:14.000 --> 00:27:29.000Stuart Winchester: So, so you were averaging below $200,000 for most of the, of the 2010s, you know, much lower some years, and now you're pretty steadily adding, averaging over $200,000. Now, last year, I don't have your numbers, I'm, I'm sure that they were, uh, pretty bad, but.00:27:29.000 --> 00:27:33.000Stuart Winchester: You know, in relation to what your past numbers were.00:27:33.000 --> 00:27:35.000Stuart Winchester: So here's my question.00:27:35.000 --> 00:27:48.000Stuart Winchester: Monarch was supposed to be one of the ski areas that got squeezed out by Epic, squeezed out by Icon. That was one of the big worries. Uh, independent ski areas can't compete. We're seeing the opposite, because I've put some things on this timeline.00:27:48.000 --> 00:28:04.000Stuart Winchester: So, for example, 2007, IntraWest buys Steamboat. Uh, 2008, the Epic Pass launches. 2009, Powder buys Copper. Uh, the Mountain Collective launched in 2012. Uh, Mountain Capital Partners bought Purgatory in 2015. Powder bought Eldora the following year.00:28:04.000 --> 00:28:21.000Stuart Winchester: Uh In 2017, Alterra combined Steamboat and Winter Park in 2018, they launched the icon pass. Uh And then we had the 20% epic price cut in 2022. So all these things you think, ok, Vale and Alterra putting their mountains on sale. These are big world famous mountains, Steamboat Vale Mountain.00:28:21.000 --> 00:28:25.000Stuart Winchester: Monarch must be getting crushed. No, it's not. Why?00:28:27.000 --> 00:28:45.000Chris Haggerty: I think it's a couple of reasons. Number one is we've just stayed true to who we are. And as the population has grown in the front range, like since 2010, Denver and Colorado Springs have grown by about 20%.00:28:45.000 --> 00:29:00.000Chris Haggerty: Right? So there's just more people, um, that are finding Monarch, and they're, they're like, wow, we like this, this is what we want. You know, you also throw in the fact that, um, I-70, for those in Colorado, are pretty familiar with, like, the traffic that is.00:28:49.000 --> 00:28:50.000Stuart Winchester: Mmhm.00:29:00.000 --> 00:29:17.000Chris Haggerty: created there on the weekends and it's not for everybody. So thankfully the drive to Monarch currently is a pretty attractive one, a lot less traffic. And I give credit to our ownership group, our current ownership group. They bought the resort in.00:29:17.000 --> 00:29:33.000Chris Haggerty: 2002. And part of it was, you know, a little bit of a passion project. Right? Like, they had come here and just really liked the vibe at Monarch and and wanted to protect that. And what they've been able to do is, number one, protect it.00:29:33.000 --> 00:29:53.000Chris Haggerty: Um, by not selling, you know, to, to one of the larger players, um, to just investing into our, our core, um, like deferred maintenance projects and those types of things, and then being really committed to the no name project, which we talked about and, and making sure timing was right on that and.00:29:53.000 --> 00:30:11.000Chris Haggerty: Um, that we were a healthy company before we jumped into that, so we didn't need a lifeline from, you know, someone else. So, um, there's kind of a lot of pieces that go into it. Um, and it's also worth noting that, like, our pass holders, like, and I think this goes to the growth, like, once people find us.00:30:11.000 --> 00:30:25.000Chris Haggerty: Um, our pass holder, uh, renewal rate is about 85% every year. Um, and the industry standard, depending on where you, you know, who you ask, it's somewhere between, like, 55% and 62%.00:30:25.000 --> 00:30:42.000Chris Haggerty: Um, and, and most of the 15% that we lose as we've dug into it is because of, uh, we have a lot of season pass holders from Fort Carson, um, military base outside of Colorado Springs, and they've just moved out of the area, right? So, um, I think our growth is partly due to, like.00:30:34.000 --> 00:30:35.000Stuart Winchester: Mmhm.00:30:42.000 --> 00:30:49.000Chris Haggerty: Once people come here and they discover it and they love it and they just keep coming back and we're just building upon that.00:30:49.000 --> 00:31:06.000Stuart Winchester: So you have owners that are willing to invest and that's really important. And Randy Stroud, I interviewed him, not on the podcast, but just for an article a couple of years ago. And he told me, and I'm not sure if this ended up being true, but their intention was to pay for the entire no name expansion in cash, including a brand new SkyTrak.00:31:06.000 --> 00:31:22.000Stuart Winchester: triple chair, or Doppelmayr, I can't remember what it was offhand, but regardless, it spoke to a very stable operation. Now, meanwhile, you still have some pretty old lifts, some pretty old infrastructure on the front side of the mountain, and recently Monarch's.00:31:22.000 --> 00:31:36.000Stuart Winchester: Master Development Plan was accepted by the U.S. Forest Service, which, for the listeners, just means that they said, okay, we'll let you consider doing these things on a project-by-project basis in the future. I have a lot of questions about the Master Plan, but first.00:31:36.000 --> 00:31:52.000Stuart Winchester: I want to focus on one that you called out, that Monarch Mountain called out in its first blog post announcing acceptance of the Master Plan, and it was about snowmaking. And Dan Torsell, the longtime GM of Ski Cooper, told me a long time ago when he came on this pod that.00:31:52.000 --> 00:32:04.000Stuart Winchester: that Ski Cooper and Monarch were the only big lift-surf ski areas in Colorado that don't have snowmaking. I think Silverton is too, but that sometimes gets lost in the shuffle. So, uh, you have it in your master development plan.00:32:04.000 --> 00:32:19.000Stuart Winchester: to a sort of rudimentary snowmaking system that would allow you to open early. But, this is a quote from your blog, quote, Before anyone panics when they read about snowmaking, let me assure you that Monarch is committed to remaining 100% all natural.00:32:19.000 --> 00:32:37.000Stuart Winchester: Snowmaking has been a part of our previous MDPs, Master Development Plan, and we have chosen not to implement it as we truly believe that natural snow provides the best skiing experience. This continues to be the case as snowmaking is in the plan to cover all bases, but we will not implement anything unless Mother Nature gives us no other choice.00:32:37.000 --> 00:32:40.000Stuart Winchester: Care to expand upon that at all?00:32:41.000 --> 00:32:53.000Chris Haggerty: Definitely. And yeah, and we are committed to that. And I actually, last summer, I spoke in front of some of the locals and pass holders, and when I let them know that we were committed to.00:32:41.000 --> 00:32:43.000Stuart Winchester: Or elaborate.00:32:53.000 --> 00:33:09.000Chris Haggerty: 100% all natural. That was the loudest cheer we got all day. So, like, that's what our, that's what our guests want. That's what we want. Um, as far that, you know, snowmaking has come a long way, and, and it is a necessity for a bunch of resorts.00:33:09.000 --> 00:33:19.000Chris Haggerty: Um, we're not at that point, um, and we're gonna push as far as we can, um, leaning into, like, snow farming and everything else before we ever, um.00:33:19.000 --> 00:33:35.000Chris Haggerty: have snow… have snowmaking. But, you know, in the Master Development Plan, it's in there because, like, part of putting things into that document is because it's a conceptual document. It's, like, it's designed to be a 15-year document. If it's not in there.00:33:35.000 --> 00:33:51.000Chris Haggerty: The Forest Service really doesn't even consider something. So if 15 years from now we're like, we need to put snowmaking in and it's not in there, we'd have some challenges. I will say our base elevation being basically 10,800 feet.00:33:51.000 --> 00:34:02.000Chris Haggerty: has helped us out, right? Like, as weather patterns has changed, and temperatures have rised, and winters are, you know, different every year, um…00:34:02.000 --> 00:34:13.000Chris Haggerty: we've been pretty lucky because of that base elevation, where we're… we are seeing snow, where other resorts are seeing some of that other precipitation falling, right? So, um…00:34:13.000 --> 00:34:30.000Chris Haggerty: you know, and then I mentioned the snow farming, like, we're… we're gonna do everything we can, like, on that side of things, right? And that's all the way from, like, fences strategically placed throughout the mountain, whether they're permanent or temporary, to catch the snow when the winds are coming from whatever direction.00:34:30.000 --> 00:34:47.000Chris Haggerty: And we do early season, we take the snow in the parking lots and we plow it all up and get it into one of the corners and then our groomers take it from there and start to build our base in the base area. So we have some different techniques to try to insulate us as much as possible.00:34:47.000 --> 00:34:53.000Chris Haggerty: And we're going to keep going that way as long as Mother Nature allows us to.00:34:53.000 --> 00:35:09.000Stuart Winchester: So snowmaking, it sounds like, is Monarch's emergency parachute. Meanwhile, there are a lot of big chairlift upgrades outlined on the master development plan. So Monarch has had substantially the same footprint for decades until last season's snowmaking upgrade.00:35:09.000 --> 00:35:24.000Stuart Winchester: And like I said before, more and more skier visits, and you were still running, with the exception of the 1999 Pioneer lift, the quad that was put in, with a lot of older double chairs. So Breezeway is a 1968 hall, Garfield is a 1969 hall, Panorama is a 1980 hall.00:35:24.000 --> 00:35:40.000Stuart Winchester: Tumbling as a 1981 hall. Those are all doubles. Now they've all gotten new drive terminals. So I understand they're, they're more modern in their actual functioning than, than in their look. However, the capacity is probably not a whole lot more than maybe it would have been. The master development plan proposes.00:35:40.000 --> 00:35:43.000Stuart Winchester: Replacing all of these with.00:35:43.000 --> 00:35:54.000Stuart Winchester: is advanced of a lift as a high-speed quad. Now, I realize you wanna… maybe the same reason as somebody… you say high-speed quad because you can always take it down, easier to go down than up, but talk us through…00:35:54.000 --> 00:36:05.000Stuart Winchester: your lift upgrade plan, and whether you think that high speed would be good for Monarch, or if you think that maybe fixed grip is the way you want to stay.00:36:06.000 --> 00:36:09.000Chris Haggerty: Yeah, fixed script is, you know.00:36:09.000 --> 00:36:24.000Chris Haggerty: Well, let me back up. A lot of that is in there because it's got to be in there if we choose to do that. And with no name opening, there's a lot of unknowns as far as, like, how is that going to change skier traffic around the mountain, right?00:36:14.000 --> 00:36:15.000Stuart Winchester: Mmhm.00:36:24.000 --> 00:36:41.000Chris Haggerty: One of the things we worked with the SE group, so they're, you know, a company that helps ski areas, right, with all sorts of development projects, big and small. And one of the things that they do is they do this density analysis, right? And it's a complicated, you know.00:36:41.000 --> 00:36:57.000Chris Haggerty: Analysis that takes on a lot of different data, but basically it's like they look at your uphill capacity based on the speed of the lift, the size of the lift, the length of the trip up, the time, and then what terrain it accesses from that lift.00:36:57.000 --> 00:37:14.000Chris Haggerty: And, like, an optimal score where, like, you have, like, perfect harmony from uphill and downhill is, like, a hundred percent. If you're below that, you're actually putting fewer people up the hill, that the downhill terrain can handle.00:37:14.000 --> 00:37:34.000Chris Haggerty: Monarch, in our current state, sits at a 56%. So basically, like, when you ski the runs at Monarch, even on our busiest day, because of the capacity that we have at our lifts, like, you have, like, semi-private runs, right? If we were to do all of our lift upgrades in the 2025 MDP.00:37:34.000 --> 00:37:49.000Chris Haggerty: we would actually get to an 84% density analysis. So even still, you wouldn't feel overcrowded on the runs. So knowing that, we're like, wow, well, we don't need to right away push for any kind of high speed lifts.00:37:49.000 --> 00:37:58.000Chris Haggerty: Um, and in fact, again, things that our guests like about them is twofold. Number one, it's like…00:37:58.000 --> 00:38:11.000Chris Haggerty: that slow metering from a fixed grip lift just adds, like, this extra time, right? Like, part of why we ski and snowboard is because we like to just disconnect, right, from reality for a little while and, like, be out in nature and, like.00:38:11.000 --> 00:38:21.000Chris Haggerty: you know, our guests like taking a step back and slowing things down here, and it feels like things are just moving a little slower. And second, one thing that we feel is like a…00:38:21.000 --> 00:38:36.000Chris Haggerty: a strong differentiator for us is, um, having a fixed grip lift. Our lift operators have to be at, like, the load point, and, um, you know, bumping the chairs and helping our guests on, and.00:38:36.000 --> 00:38:40.000Chris Haggerty: It's this, like, human interaction piece that, like.00:38:40.000 --> 00:38:56.000Chris Haggerty: You don't get everywhere else, especially with some of the newer lifts where, you know, it's everything's so automated, um, which has its value, but for us, again, with who Monarch is and taking that step back in time a little bit like that, that face-to-face interaction.00:38:56.000 --> 00:39:14.000Chris Haggerty: like, so many of our reviews are constantly calling out our lift operators, like, oh, yeah, I remember so-and-so, like, he's awesome, he helped me on, or he just said hello, or he remembered what kind of skis I'm wearing, or told me about this secret powder stash, which… another benefit of a slower uphill capacity is, like.00:39:14.000 --> 00:39:23.000Chris Haggerty: Monarch, even though it's smaller in size, doesn't get tracked out. Like, you get powder turns at the end of the day on a powder day.00:39:23.000 --> 00:39:34.000Stuart Winchester: So the, I'm gonna share my screen to, to talk about this next bit. So those are the lift upgrades that you're proposing. There's also two, and I'm, I, so for those watching.00:39:34.000 --> 00:39:49.000Stuart Winchester: On YouTube or stormskiing.com, I have the master development plan, conceptual plan up. There's also two infill lifts. Okay, so there would be this planned NNB lift, which is a redundant lift in the No Name Basin expansion. This is the new lift that opened last year. And then there's this…00:39:49.000 --> 00:40:04.000Stuart Winchester: Plan, divide, express. So, again, it's two more lifts on the same footprint. And I didn't see in the master plan, maybe I missed it, that you do have all this cat skiing terrain over here, so…00:40:04.000 --> 00:40:10.000Stuart Winchester: So, talk us through the logic of putting in more lists to serve the same amount of terrain, and if you've considered.00:40:10.000 --> 00:40:18.000Stuart Winchester: moving lift surf skiing into the cat skiing, as Targi did with its peaked mountain expansion a couple of years ago.00:40:19.000 --> 00:40:37.000Chris Haggerty: Yeah, so no name is actually was a part of our cat skiing terrain. So we did take a little bit of that. That redundant lift out there is just because we'll see what happens, how many people just access the terrain and no name. And if we ever need a second lift there, then.00:40:23.000 --> 00:40:24.000Stuart Winchester: Hmm. Okay.00:40:37.000 --> 00:40:55.000Chris Haggerty: um, we'll consider it. As far as further into cat skiing terrain, um, you know, our current MDP outlines our next 15 years, so we're not looking to push any further into that. Um, we also have, um, the Mirkwood terrain, which is a little bit of a hike-to terrain.00:40:55.000 --> 00:41:04.000Chris Haggerty: Um, between, like, the front side and no name, and we get… that's… that's the area we most commonly get asked about a lift, and uh…00:41:04.000 --> 00:41:12.000Chris Haggerty: I will tell you from my, uh, initial conversations with owners, it's… it's not a, um…00:41:12.000 --> 00:41:27.000Chris Haggerty: the answer isn't no, we're not putting a lift there, it's a expletive no, we're not putting a lift there. Um, we take pride in having, like, being a small mountain, but still offering, like, all these other things, right? Which one being hike to terrain, um, which is pretty cool for…00:41:27.000 --> 00:41:31.000Chris Haggerty: a small mountain like this. And then that, um…00:41:31.000 --> 00:41:41.000Chris Haggerty: the Divide Express, right, that would get us all the way from the base area to the summit on the front side. Um, that… that is one that we've, you know, had some…00:41:41.000 --> 00:41:55.000Chris Haggerty: serious discussions about, and, um, you know, it probably… number one, it would just, you know, allow guests to not have to take two lifts to get to the… to the summit on the front side. Um, but also, if we…00:41:55.000 --> 00:42:06.000Chris Haggerty: explore, um, summer operations in the future. You know, having a high-speed, you know, detachable lift gives you more flexibility to do that.00:42:06.000 --> 00:42:13.000Chris Haggerty: So that kind of, you know, would add into that if we get to that point.00:42:12.000 --> 00:42:24.000Stuart Winchester: So let me ask a stupid question, because that's what I'm here for. So Garfield Lyft, as I said, was built in the 60s. Pioneer Lyft came along in 1999. So I would have thought, okay, they cross over each other.00:42:24.000 --> 00:42:41.000Stuart Winchester: Because they, they retroactively built Pioneer and didn't realize it. Uh, I would have thought, hey, the master plan is an opportunity to straighten those out so they don't have to cross over each other. Maybe I'm overthinking it and it's not that complicated, but, but talk about this interesting crisscross pattern you have here and, and why that's where Garfield and Pioneer are slated.00:42:24.000 --> 00:42:25.000Chris Haggerty: Mm-hmm.00:42:41.000 --> 00:42:45.000Stuart Winchester: to continue to exist, even if you were to upgrade their lifts.00:42:47.000 --> 00:43:05.000Chris Haggerty: Yeah, I mean, I don't know that I have the best answer. We do have a proposal to shift the base of the bottom terminal of Pioneer uphill a little bit, just because the terrain is a little more desirable there. And obviously, if we needed to add the Divide Express, we need to open that area up.00:43:05.000 --> 00:43:06.000Chris Haggerty: um…00:43:06.000 --> 00:43:21.000Chris Haggerty: but other than that, right, we just kind of… we like the terrain that they serve right now, we don't really have a better spot, so the lifts crisscrossing wasn't like a, oh man, we need to change this. It wasn't something that we were focused on solving, um, in a way.00:43:21.000 --> 00:43:24.000Chris Haggerty: I think it's kind of a little cool, I guess.00:43:24.000 --> 00:43:40.000Stuart Winchester: I do like criss-crossing lifts, so I'm glad we're doing this one a video, because every time I bring this up, I get tomatoes thrown at me. Uh, Monarch skier visits have gone up and up, as we said. However, as they've gone up and up, there's still nowhere to stay on the mountain. There never has been. Uh.00:43:40.000 --> 00:43:51.000Stuart Winchester: a philosophical choice? Is that just a practical matter of, hey, the Forest Service owns the land at the base? Is it both? Talk us through why there's nowhere to stay at Monarch as a drive ski area only.00:43:52.000 --> 00:43:55.000Chris Haggerty: Yeah, there's a couple of reasons.00:43:56.000 --> 00:44:02.000Chris Haggerty: And, you know, one of them, again, I'll say our pass holders and our guests, like…00:44:02.000 --> 00:44:18.000Chris Haggerty: are pretty vocal about not wanting that, right? Like that's what separates Monarch. We're not that resort that has the big base village. And again, that's not to knock those resorts. I think that that's, you know, I've gone to those resorts as a guest. I think it's awesome.00:44:18.000 --> 00:44:38.000Chris Haggerty: I think another side of it for us is we're really focused on our community, right? We're a small mountain ski area and there's a small mountain community 15 miles from where we are that has a really cool, the town of Salida has a really cool downtown, a main street.00:44:38.000 --> 00:44:52.000Chris Haggerty: Um, with a ton of restaurants, a ton of mom-and-pop-owned shops, right? And, um, you know, we're our largest, uh, driver for winter visits, and, and, you know, um…00:44:52.000 --> 00:45:01.000Chris Haggerty: in the wintertime, and we want to make sure we embrace that community. If we moved a village up onto the mountain, like.00:45:01.000 --> 00:45:16.000Chris Haggerty: Salida is going to suffer. All those guests now are going to stay in the lodging up at Monarch versus in town. And we want to be good community partners and be like, no, here, you guys should be the ones to have lodging. And we live.00:45:16.000 --> 00:45:35.000Chris Haggerty: all of us live in the community as well, and we want to see those restaurants, you know, stay active all year round, the mom-and-pop shops, like, you know, thrive. So part of it is just, like, wanting to be a part of something bigger, right, in that larger community, rather than.00:45:35.000 --> 00:45:38.000Chris Haggerty: moving everything up to… to Monarch.00:45:38.000 --> 00:45:53.000Stuart Winchester: So if Snowpass skiers do decide to come out and visit Monarch, they'll stay in Salida, most likely, and drive up that day. So leave us with this, Chris, because Joe Heschen, the CEO of Snow Partners.00:45:53.000 --> 00:45:58.000Stuart Winchester: Said we get so caught up on stats, but stats aren't aren't always the thing and and.00:45:58.000 --> 00:46:04.000Stuart Winchester: when you look statistically at Monarch, it's 1,100 vertical feet.00:46:04.000 --> 00:46:21.000Stuart Winchester: Lift serve drop. That's 25th in Colorado. Uh, 1017 acres that ranks 17th in Colorado is bigger than Aspen Mountain. I'll point, I'll point out, uh, it's 308 inch average annual snowfall is good, but it's, it's good for ninth best. It has the third highest base, as I said, but the 12th highest summ.00:46:21.000 --> 00:46:28.000Stuart Winchester: It has those older lifts in a state that has 120 high-speed lifts, and there's nowhere to stay on the mountain, so…00:46:28.000 --> 00:46:34.000Stuart Winchester: I'm just stating those facts, right? So what's your pitch to someone to come.00:46:34.000 --> 00:46:37.000Stuart Winchester: use that snow pass, and try out Monarch.00:46:39.000 --> 00:46:44.000Chris Haggerty: Yeah, the big one is to start with, we're 100% natural, right?00:46:44.000 --> 00:46:59.000Chris Haggerty: you want natural snow, it is a definitely a different feel, even as good as snowmaking is these days, like, come check it out. And this, and the second part is, um, the people that are at Monarch are a unique group, right? Like, we're, we're.00:46:59.000 --> 00:47:18.000Chris Haggerty: past the characters, whether you work here or you're a guest here. And, um, we're looking out for each other, right? Like, it's not, um, it's not judgmental, like, no matter what you show up wearing or what your ability is, like, we just want everyone here to have a fun day on the mountain.00:47:18.000 --> 00:47:37.000Chris Haggerty: whether you're the extreme person out in the cat skiing terrain, hiking Mirkwood, the Never Ever, or the SnowTuber, which I love SnowTubers, you know, I think that's who we all are at our core. Um, you know, it's just… it's just a cool vibe that you gotta just come check it out, and our track record's saying that.00:47:29.000 --> 00:47:31.000Stuart Winchester: Yeah.00:47:37.000 --> 00:47:40.000Chris Haggerty: People love it once they find it. So give it a shot.00:47:40.000 --> 00:47:51.000Stuart Winchester: I'm always telling my readers that these old lifts, these hauls, these riblets, they're not gonna be around for a whole lot longer, right? There's still hundreds in operation, but they don't last forever.00:47:51.000 --> 00:48:08.000Stuart Winchester: And the best thing you can do to memorialize them, you can't really preserve them forever because they're functional machines, right? But go ride them. And right now, Monarch, along with Lost Trail, is probably the best place in the West to go and ride all Old Hall Liftley. And I understand, again, they've been updated.00:48:08.000 --> 00:48:23.000Stuart Winchester: But the aesthetic is similar. So, alright, Chris, look, I gotta take my own advice and get out there and ride those hauls with you this winter. It was really great to talk to you. I really appreciate you coming on. I can't wait to catch up and hopefully make some turns together this winter.00:48:23.000 --> 00:48:27.000Chris Haggerty: Yeah, thanks a lot for the time and look forward to seeing you out here at Monarch.00:48:26.000 --> 00:48:29.000Stuart Winchester: All right. Talk to you soon. Thanks so much, Chris.00:48:28.000 --> 00:48:30.000Chris Haggerty: Thank you. Bye.00:48:31.000 --> 00:48:47.000Stuart Winchester: Alright, that was Chris Haggerty, the General Manager of Monarch Mountain. That was really fun, I'm really glad that he came on. Uh, Randy Stroud, just to be clear, the former GM, it's not that he ever didn't want to come on the podcast, it's just that.00:48:47.000 --> 00:49:00.000Stuart Winchester: I covered Monarch in a different way for the story about the expansion. So he was very good to the storm as well. I do want to tell you that today's podcast is brought to you by Profile Search International.00:49:00.000 --> 00:49:08.000Stuart Winchester: If you want to find a great leader for your mountain team, I want to introduce you to the pros at Profile Search International.00:49:08.000 --> 00:49:18.000Stuart Winchester: They are the ski industry talent acquisition experts, and they are the only executive search and recruitment firm in the entire world that is 100% focused on the ski industry.00:49:18.000 --> 00:49:31.000Stuart Winchester: ProfileSearch has used their intimate understanding of skiing and related industries, and of available candidates worldwide, to place hundreds of transformational leaders at the best and most progressive ski areas over the past 30 years.00:49:31.000 --> 00:49:44.000Stuart Winchester: Profile Search has offices in the US and Canada. They find and negotiate with the right leaders for your team, which sometimes you don't even know that you need in this era of online everything and big data. You can reach Profile Search.00:49:44.000 --> 00:49:59.000Stuart Winchester: Directly at ProfileSearch.com, where you will find their email or phone number. You can also send me a note. I will be happy to connect you directly with this expert team. My email address is skiing at substack.com. Alright, we are going to wrap up today.00:49:59.000 --> 00:50:05.000Stuart Winchester: With some reader reaction to yesterday's podcast on…00:50:06.000 --> 00:50:22.000Stuart Winchester: with Midwest skiers, Matt Zabranski, and there wasn't a lot of feedback. I think, number one, because the Midwest doesn't get as much feedback in general, but also because I published this one pretty late, I think I put it out at 8 or 9 at night, and so folks may just have not had a chance.00:50:24.000 --> 00:50:27.000Stuart Winchester: really process it yet. So…00:50:28.000 --> 00:50:30.000Stuart Winchester: Nell says…00:50:30.000 --> 00:50:41.000Stuart Winchester: Midwest Skiers, meaning MidwestSkiers.com, Matthew Zbrensky's site, is a regional treasure. The ride-throughs are so thorough, and what Nell means by that is.00:50:41.000 --> 00:50:52.000Stuart Winchester: Uh, Midwest Skiers puts up on their YouTube channel a, hey, look, here's the trail map at, say, Trollhagen, Wisconsin. Here's all 32 trails, whatever it is.00:50:52.000 --> 00:51:03.000Stuart Winchester: and then it rides the reach for the GoPro, and tells you about them, and gives you a really specific breakdown, uh, with some great graphs. So the production quality is really, really good. So, I second that now. Really good.00:51:03.000 --> 00:51:15.000Stuart Winchester: breakdowns of the ski areas from Midwest Skiers. Eric Morris says, great Midwest content, it'd be great to talk to the Dean of Southern Skiing, Randy Johnson, who wrote Southern Snow, the new guide to winter sports.00:51:15.000 --> 00:51:31.000Stuart Winchester: Couldn't agree more, Eric. If you know him, I would love an introduction. I am not acquainted with Mr. Randy Johnson, but I would love to meet him, as the South has certainly been an undercovered region from my point of view, and not on purpose, just because there's…00:51:31.000 --> 00:51:44.000Stuart Winchester: a little different culturally than the rest of the United States from a ski point of view, and I have not sunk as far of roots in there as I would like to, especially since I'm not too far from those ski areas. There was also…00:51:44.000 --> 00:51:47.000Stuart Winchester: We're not getting as much action on the chat, but here's one.00:51:47.000 --> 00:51:57.000Stuart Winchester: question on the chat, and this is not to do with the podcast from yesterday, but I want to talk about it a little. What will it take for Vail Resorts to turn around? That's from Saurabh Jain.00:51:57.000 --> 00:52:06.000Stuart Winchester: And commenter Slappy said, redefine what success looks like and suspend the dividend and stock buyback. Of course, those last two may be non-negotiable.00:52:06.000 --> 00:52:24.000Stuart Winchester: I'm not going to get out of my lane and talk too much about dividends and stock buybacks, though I do like stocks and follow that world nominally. It's certainly nothing that I'm an expert to speak on. What Vail needs to do to turn around, I think, is very simple, and I've said this many, many times.00:52:24.000 --> 00:52:25.000Stuart Winchester: And I've written about it.00:52:25.000 --> 00:52:37.000Stuart Winchester: many times, and we've talked about it on the podcast many times, is they need to treat every single ski resort like it's Vail Mountain, and every single guest like they're the President of the United States. They…00:52:37.000 --> 00:52:40.000Stuart Winchester: They need to make sure that.00:52:41.000 --> 00:52:48.000Stuart Winchester: the experience of a lifetime ethos is transferred to any Skiera, whether it's Beaver Creek.00:52:48.000 --> 00:53:03.000Stuart Winchester: or it's Mad River, Ohio. And that has not always seemed to be the case since Vail Resorts purchased Peak Resorts in 2019 and sort of doubled the size of its owned resort portfolio within a year's span. Now.00:53:03.000 --> 00:53:08.000Stuart Winchester: Certainly, Vail has recently recommitted to.00:53:08.000 --> 00:53:11.000Stuart Winchester: To that sort of quality, and we'll see how that plays out.00:53:11.000 --> 00:53:13.000Stuart Winchester: I think…00:53:13.000 --> 00:53:19.000Stuart Winchester: maybe a little more shuffling with the Epic Pass, uh, maybe a higher-priced.00:53:19.000 --> 00:53:22.000Stuart Winchester: Full pass to to.00:53:22.000 --> 00:53:37.000Stuart Winchester: help traffic some of the more popular ski areas, and maybe some more options. They do a really good job, actually, with the Epic Regionals, with the Tahoe Local, the Northeast Local, and the Summit County Local. So, I think Vail's done a nice job with that. And I think…00:53:38.000 --> 00:53:45.000Stuart Winchester: From my experience, Vail executives, and I've hosted many of them on this podcast. They're very receptive to criticism.00:53:45.000 --> 00:53:47.000Stuart Winchester: Uh, and they are…00:53:48.000 --> 00:53:53.000Stuart Winchester: Willing to listen and change and evolve. I'm not sure.00:53:53.000 --> 00:53:57.000Stuart Winchester: If that always comes through.00:53:57.000 --> 00:54:04.000Stuart Winchester: Outside of direct lines of questioning, if that makes sense. So, so the, the owning of mistakes and saying, okay, you know what?00:54:04.000 --> 00:54:19.000Stuart Winchester: we messed up because Paley Peaks was only open 25 days a year in 2021, or whatever it was. There was a year when a lot of its Midwest ski areas were extremely restricted, or the COVID restrictions and everyone else, but the local ski area operators there.00:54:19.000 --> 00:54:24.000Stuart Winchester: we're out competing them. Uh, you know, when Vail first bought Wilmot, Wisconsin.00:54:24.000 --> 00:54:27.000Stuart Winchester: In 2016, there…00:54:27.000 --> 00:54:38.000Stuart Winchester: their… or maybe that was 2012. They bought Afton Alps, Wilmot, and Mount Brighton in some order, two of them in 2012 and one in 2016. Anyway, there was a great article I read that said.00:54:38.000 --> 00:54:49.000Stuart Winchester: what Wilmot skiers can expect, and Wilmot's a small ski area outside of, well, it's broad, it has a lot of lifts, outside of Wisconsin and Chicago, uh, Milwaukee and Chicago.00:54:49.000 --> 00:54:56.000Stuart Winchester: And what they said was, we're gonna bring Western standards to the Midwest, right? And that was very out front. Dale spokespeople were saying that.00:54:56.000 --> 00:55:11.000Stuart Winchester: You never hear them say that anymore, and it's not clear that that is the standard, and I had a very nice time at Wilmot when I went two years ago, and super friendly staff, and all the lifts were running, even though clearly they didn't need to be, because it was a Monday, and hardly anyone was there, so I'm certain they were.00:55:11.000 --> 00:55:19.000Stuart Winchester: losing money on that day, and yet they still did their best to provide an experience. I'm just not…00:55:19.000 --> 00:55:21.000Stuart Winchester: It's just not always…00:55:21.000 --> 00:55:31.000Stuart Winchester: super clear throughout the portfolio, particularly in the Lower Midwest, Indiana, Missouri, Ohio, uh, in some of the Pennsylvania ski areas, Mid-Atlantic.00:55:31.000 --> 00:55:36.000Stuart Winchester: that that this is their operating philosophy. So I think that.00:55:36.000 --> 00:55:38.000Stuart Winchester: I think Vail's better than most…00:55:38.000 --> 00:55:49.000Stuart Winchester: than you would think, scrolling through social media. Uh, in general, I've been to pretty much every Vail resort in North America, other than, I think, the two in Missouri and Paoli Peaks.00:55:49.000 --> 00:56:07.000Stuart Winchester: And I generally have a really good time and a positive experience, and I find the staff friendly, and that's a super important thing to me. I love that Vail puts bars on all the lifts. I love that they're focused on safety in general, that they help police speed. As someone who skis with my kids, that's important to me.00:56:07.000 --> 00:56:13.000Stuart Winchester: So I generally have a pretty good experience. I think…00:56:14.000 --> 00:56:29.000Stuart Winchester: there's a little too much gum in the machine, maybe, when it comes to communicating these things and owning and correcting mistakes. So, anyway, with that, I will leave Vel alone for today, and you're probably wondering why I haven't yet covered the new Slate.00:56:29.000 --> 00:56:44.000Stuart Winchester: of proposed board members by Oasis Capital for Vail Resorts. I don't want to get too into the weeds on stock stuff, but I do have some thoughts on that. I want to wait till I have a little bit more time to break it down.00:56:45.000 --> 00:56:50.000Stuart Winchester: So I thank you for coming today. I hope you enjoyed that podcast episode.00:56:50.000 --> 00:56:58.000Stuart Winchester: If you are not already subscribed to the Storm Skiing newsletter, please click over to stormskiing.com and subscribe.00:56:58.000 --> 00:57:14.000Stuart Winchester: That will come right to your inbox. It is not only podcasts, I do a ton of writing. And if you want to upgrade to a paid subscription to the Storm Skiing Journal and Podcast, that will give you everything below the paywall, which is a lot of really deep analysis.00:57:14.000 --> 00:57:29.000Stuart Winchester: that I do on the lift service ski industry, particularly in North America, and especially in the United States. Also, if you're a paid subscriber to The Storm at stormskiing.com, that is the only way to interact with the podcast, because I'm only reading comments from the.00:57:29.000 --> 00:57:30.000Stuart Winchester: Previous.00:57:31.000 --> 00:57:47.000Stuart Winchester: Day's podcast and only paid subscribers can comment. Only paid subscribers can join the chat, which is the second way that they can interact with the podcast. If you want to rep the storm, we have a swag store. You can buy shirts, hats, et cetera, at stormskiing.myshopify.com.00:57:47.000 --> 00:57:48.000Stuart Winchester: Okay.00:57:48.000 --> 00:57:52.000Stuart Winchester: Please follow me on YouTube. YouTube is…00:57:52.000 --> 00:57:55.000Stuart Winchester: Where we will increasingly.00:57:55.000 --> 00:58:02.000Stuart Winchester: It'll be increasingly important to us, and right now, we don't have that many subscribers there. Uh, everything is going there, but it's just not…00:58:02.000 --> 00:58:12.000Stuart Winchester: made it into the cultural fabric of that site just yet, so please follow me at Storm Ski Journal on YouTube. Also, Instagram, which I post on more during the winter season.00:58:12.000 --> 00:58:27.000Stuart Winchester: I will be back tomorrow, or I'm scheduled to be back tomorrow, with, and this is one a lot of people have asked me for, a conversation with the general manager of what I think was perhaps the most improved ski area in the country last year, Camelback.00:58:27.000 --> 00:58:38.000Stuart Winchester: Pennsylvania. If you're rolling your eyes, if you live out west, I encourage you to tune in anyway, because the story of Camelback is a really good story of how do we do things wrong?Stuff referenced in the pod:Monarch's amazing season pass reciprocal network:A full breakdown of Monarch's U.S. Forest Service masterplan:Conversation with Snow Partners' CEO Joe Hession: Get full access to The Storm Skiing Journal and Podcast at www.stormskiing.com/subscribe

USC Gerontology Professor Dr. Changhan David Lee breaks down the discovery of MOTS-c, a novel mitochondrial-derived peptide that functions as an exercise mimetic, cellular stress responder, and immunomodulator.Are mitochondria more than just cellular powerhouses? In this episode of Modern Healthspan, Dr David Lee, Associate Professor at the USC Leonard Davis School of Gerontology and co-discoverer of MOTS-c, joins us for a deep dive into the emerging world of mitochondrial-derived peptides (MDPs). Dr. Lee explains how these small microproteins, encoded within the mitochondrial 12S rRNA locus, act as critical retrograde signaling molecules. We explore how MOTS-c translocates to the nucleus during metabolic stress, coordinates with AMPK and hypothalamic POMC neurons, and functions as an interferon-responsive host defense peptide. Additionally, Dr. Lee shares findings from preclinical models examining sarcopenia, muscle homeostasis, and age-related physical capacity, while providing an update on the current regulatory state, WADA anti-doping classification, and human translational pipeline of MDP research.

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: A Simple Toy Coherence Theorem, published by johnswentworth on August 2, 2024 on The AI Alignment Forum. This post presents a simple toy coherence theorem, and then uses it to address various common confusions about coherence arguments. Setting Deterministic MDP. That means at each time t there's a state S[t][1], the agent/policy takes an action A[t] (which can depend on both time t and current state S[t]), and then the next state S[t+1] is fully determined by S[t] and A[t]. The current state and current action are sufficient to tell us the next state. We will think about values over the state at some final time T. Note that often in MDPs there is an incremental reward each timestep in addition to a final reward at the end; in our setting there is zero incremental reward at each timestep. One key point about this setting: if the value over final state is uniform, i.e. same value for all final states, then the MDP is trivial. In that case, all policies are optimal, it does not matter at all what the final state is or what any state along the way is, everything is equally valuable. Theorem There exist policies which cannot be optimal for any values over final state except for the trivial case of uniform values. Furthermore, such policies are exactly those which display inconsistent revealed preferences transitively between all final states. Proof As a specific example: consider an MDP in which every state is reachable at every timestep, and a policy which always stays in the same state over time. From each state S every other state is reachable, yet the policy chooses S, so in order for the policy to be optimal S must be a highest-value final state. Since each state must be a highest-value state, the policy cannot be optimal for any values over final state except for the trivial case of uniform values. That establishes the existence part of the theorem, and you can probably get the whole idea by thinking about how to generalize that example. The rest of the proof extends the idea of that example to inconsistent revealed preferences in general. Bulk of Proof (click to expand) Assume the policy is optimal for some particular values over final state. We can then start from those values over final state and compute the best value achievable starting from each state at each earlier time. That's just dynamic programming: V[S,t]=max S' reachable in next timestep from S V[S',t+1] where V[S,T] are the values over final states. A policy is optimal for final values V[S,T] if-and-only-if at each timestep t1 it chooses a next state with highest reachable V[S,t]. Now, suppose that at timestep t there are two different states either of which can reach either state A or state B in the next timestep. From one of those states the policy chooses A; from the other the policy chooses B. This is an inconsistent revealed preference between A and B at time t: sometimes the policy has a revealed preference for A over B, sometimes for B over A. In order for a policy with an inconsistent revealed preference between A and B at time t to be optimal, the values must satisfy V[A,t]=V[B,t] Why? Well, a policy is optimal for final values V[S,T] if-and-only if at each timestep t1 it chooses a next state with highest reachable V[S,t]. So, if an optimal policy sometimes chooses A over B at timestep t when both are reachable, then we must have V[A,t]V[B,t]. And if an optimal policy sometimes chooses B over A at timestep t when both are reachable, then we must have V[A,t]V[B,t]. If both of those occur, i.e. the policy has an inconsistent revealed preference between A and B at time t, then V[A,t]=V[B,t]. Now, we can propagate that equality to a revealed preference on final states. We know that the final state which the policy in fact reaches starting from A at time t must have the highest reachable value, a...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: A Simple Toy Coherence Theorem, published by johnswentworth on August 2, 2024 on LessWrong. This post presents a simple toy coherence theorem, and then uses it to address various common confusions about coherence arguments. Setting Deterministic MDP. That means at each time t there's a state S[t][1], the agent/policy takes an action A[t] (which can depend on both time t and current state S[t]), and then the next state S[t+1] is fully determined by S[t] and A[t]. The current state and current action are sufficient to tell us the next state. We will think about values over the state at some final time T. Note that often in MDPs there is an incremental reward each timestep in addition to a final reward at the end; in our setting there is zero incremental reward at each timestep. One key point about this setting: if the value over final state is uniform, i.e. same value for all final states, then the MDP is trivial. In that case, all policies are optimal, it does not matter at all what the final state is or what any state along the way is, everything is equally valuable. Theorem There exist policies which cannot be optimal for any values over final state except for the trivial case of uniform values. Furthermore, such policies are exactly those which display inconsistent revealed preferences transitively between all final states. Proof As a specific example: consider an MDP in which every state is reachable at every timestep, and a policy which always stays in the same state over time. From each state S every other state is reachable, yet the policy chooses S, so in order for the policy to be optimal S must be a highest-value final state. Since each state must be a highest-value state, the policy cannot be optimal for any values over final state except for the trivial case of uniform values. That establishes the existence part of the theorem, and you can probably get the whole idea by thinking about how to generalize that example. The rest of the proof extends the idea of that example to inconsistent revealed preferences in general. Bulk of Proof (click to expand) Assume the policy is optimal for some particular values over final state. We can then start from those values over final state and compute the best value achievable starting from each state at each earlier time. That's just dynamic programming: V[S,t]=max S' reachable in next timestep from S V[S',t+1] where V[S,T] are the values over final states. A policy is optimal for final values V[S,T] if-and-only-if at each timestep t1 it chooses a next state with highest reachable V[S,t]. Now, suppose that at timestep t there are two different states either of which can reach either state A or state B in the next timestep. From one of those states the policy chooses A; from the other the policy chooses B. This is an inconsistent revealed preference between A and B at time t: sometimes the policy has a revealed preference for A over B, sometimes for B over A. In order for a policy with an inconsistent revealed preference between A and B at time t to be optimal, the values must satisfy V[A,t]=V[B,t] Why? Well, a policy is optimal for final values V[S,T] if-and-only if at each timestep t1 it chooses a next state with highest reachable V[S,t]. So, if an optimal policy sometimes chooses A over B at timestep t when both are reachable, then we must have V[A,t]V[B,t]. And if an optimal policy sometimes chooses B over A at timestep t when both are reachable, then we must have V[A,t]V[B,t]. If both of those occur, i.e. the policy has an inconsistent revealed preference between A and B at time t, then V[A,t]=V[B,t]. Now, we can propagate that equality to a revealed preference on final states. We know that the final state which the policy in fact reaches starting from A at time t must have the highest reachable value, and that value...

Link to original articleWelcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: A Simple Toy Coherence Theorem, published by johnswentworth on August 2, 2024 on LessWrong. This post presents a simple toy coherence theorem, and then uses it to address various common confusions about coherence arguments. Setting Deterministic MDP. That means at each time t there's a state S[t][1], the agent/policy takes an action A[t] (which can depend on both time t and current state S[t]), and then the next state S[t+1] is fully determined by S[t] and A[t]. The current state and current action are sufficient to tell us the next state. We will think about values over the state at some final time T. Note that often in MDPs there is an incremental reward each timestep in addition to a final reward at the end; in our setting there is zero incremental reward at each timestep. One key point about this setting: if the value over final state is uniform, i.e. same value for all final states, then the MDP is trivial. In that case, all policies are optimal, it does not matter at all what the final state is or what any state along the way is, everything is equally valuable. Theorem There exist policies which cannot be optimal for any values over final state except for the trivial case of uniform values. Furthermore, such policies are exactly those which display inconsistent revealed preferences transitively between all final states. Proof As a specific example: consider an MDP in which every state is reachable at every timestep, and a policy which always stays in the same state over time. From each state S every other state is reachable, yet the policy chooses S, so in order for the policy to be optimal S must be a highest-value final state. Since each state must be a highest-value state, the policy cannot be optimal for any values over final state except for the trivial case of uniform values. That establishes the existence part of the theorem, and you can probably get the whole idea by thinking about how to generalize that example. The rest of the proof extends the idea of that example to inconsistent revealed preferences in general. Bulk of Proof (click to expand) Assume the policy is optimal for some particular values over final state. We can then start from those values over final state and compute the best value achievable starting from each state at each earlier time. That's just dynamic programming: V[S,t]=max S' reachable in next timestep from S V[S',t+1] where V[S,T] are the values over final states. A policy is optimal for final values V[S,T] if-and-only-if at each timestep t1 it chooses a next state with highest reachable V[S,t]. Now, suppose that at timestep t there are two different states either of which can reach either state A or state B in the next timestep. From one of those states the policy chooses A; from the other the policy chooses B. This is an inconsistent revealed preference between A and B at time t: sometimes the policy has a revealed preference for A over B, sometimes for B over A. In order for a policy with an inconsistent revealed preference between A and B at time t to be optimal, the values must satisfy V[A,t]=V[B,t] Why? Well, a policy is optimal for final values V[S,T] if-and-only if at each timestep t1 it chooses a next state with highest reachable V[S,t]. So, if an optimal policy sometimes chooses A over B at timestep t when both are reachable, then we must have V[A,t]V[B,t]. And if an optimal policy sometimes chooses B over A at timestep t when both are reachable, then we must have V[A,t]V[B,t]. If both of those occur, i.e. the policy has an inconsistent revealed preference between A and B at time t, then V[A,t]=V[B,t]. Now, we can propagate that equality to a revealed preference on final states. We know that the final state which the policy in fact reaches starting from A at time t must have the highest reachable value, and that value...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Linear infra-Bayesian Bandits, published by Vanessa Kosoy on May 10, 2024 on The AI Alignment Forum. Linked is my MSc thesis, where I do regret analysis for an infra-Bayesian[1] generalization of stochastic linear bandits. The main significance that I see in this work is: Expanding our understanding of infra-Bayesian regret bounds, and solidifying our confidence that infra-Bayesianism is a viable approach. Previously, the most interesting IB regret analysis we had was Tian et al which deals (essentially) with episodic infra-MDPs. My work here doesn't supersede Tian et al because it only talks about bandits (i.e. stateless infra-Bayesian laws), but it complements it because it deals with a parameteric hypothesis space (i.e. fits into the general theme in learning-theory that generalization bounds should scale with the dimension of the hypothesis class). Discovering some surprising features of infra-Bayesian learning that have no analogues in classical theory. In particular, it turns out that affine credal sets (i.e. such that are closed w.r.t. arbitrary affine combinations of distributions and not just convex combinations) have better learning-theoretic properties, and the regret bound depends on additional parameters that don't appear in classical theory (the "generalized sine" S and the "generalized condition number" R). Credal sets defined using conditional probabilities (related to Armstrong's "model splinters") turn out to be well-behaved in terms of these parameters. In addition to the open questions in the "summary" section, there is also a natural open question of extending these results to non-crisp infradistributions[2]. (I didn't mention it in the thesis because it requires too much additional context to motivate.) 1. ^ I use the word "imprecise" rather than "infra-Bayesian" in the title, because the proposed algorithms achieves a regret bound which is worst-case over the hypothesis class, so it's not "Bayesian" in any non-trivial sense. 2. ^ In particular, I suspect that there's a flavor of homogeneous ultradistributions for which the parameter S becomes unnecessary. Specifically, an affine ultradistribution can be thought of as the result of "take an affine subspace of the affine space of signed distributions, intersect it with the space of actual (positive) distributions, then take downwards closure into contributions to make it into a homogeneous ultradistribution". But we can also consider the alternative "take an affine subspace of the affine space of signed distributions, take downwards closure into signed contributions and then intersect it with the space of actual (positive) contributions". The order matters! Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Measuring Coherence of Policies in Toy Environments, published by dx26 on March 18, 2024 on LessWrong. This post was produced as part of the Astra Fellowship under the Winter 2024 Cohort, mentored by Richard Ngo. Thanks to Martín Soto, Jeremy Gillien, Daniel Kokotajlo, and Lukas Berglund for feedback. Summary Discussions around the likelihood and threat models of AI existential risk (x-risk) often hinge on some informal concept of a "coherent", goal-directed AGI in the future maximizing some utility function unaligned with human values. Whether and how coherence may develop in future AI systems, especially in the era of LLMs, has been a subject of considerable debate. In this post, we provide a preliminary mathematical definition of the coherence of a policy as how likely it is to have been sampled via uniform reward sampling (URS), or uniformly sampling a reward function and then sampling from the set of policies optimal for that reward function, versus uniform policy sampling (UPS). We provide extensions of the model for sub-optimality and for "simple" reward functions via uniform sparsity sampling (USS). We then build a classifier for the coherence of policies in small deterministic MDPs, and find that properties of the MDP and policy, like the number of self-loops that the policy takes, are predictive of coherence when used as features for the classifier. Moreover, coherent policies tend to preserve optionality, navigate toward high-reward areas of the MDP, and have other "agentic" properties. We hope that our metric can be iterated upon to achieve better definitions of coherence and a better understanding of what properties dangerous AIs will have. Introduction Much of the current discussion about AI x-risk centers around "agentic", goal-directed AIs having misaligned goals. For instance, one of the most dangerous possibilities being discussed is of mesa-optimizers developing within superhuman models, leading to scheming behavior and deceptive alignment. A significant proportion of current alignment work focuses on detecting, analyzing (e.g. via analogous case studies of model organisms), and possibly preventing deception. Some researchers in the field believe that intelligence and capabilities are inherently tied with "coherence", and thus any sufficiently capable AI will approximately be a coherent utility function maximizer. In their paper "Risks From Learned Optimization" formally introducing mesa-optimization and deceptive alignment, Evan Hubinger et al. discuss the plausibility of mesa-optimization occurring in RL-trained models. They analyze the possibility of a base optimizer, such as a hill-climbing local optimization algorithm like stochastic gradient descent, producing a mesa-optimizer model that internally does search (e.g. Monte Carlo tree search) in pursuit of a mesa-objective (in the real world, or in the "world-model" of the agent), which may or may not be aligned with human interests. This is in contrast to a model containing many complex heuristics that is not well-defined internally as a consequentialist mesa-optimizer; one extreme example is a tabular model/lookup table that matches observations to actions, which clearly does not do any internal search or have any consequentialist cognition. They speculate that mesa-optimizers may be selected for because they generalize better than other models, and/or may be more compressible information-theoretic wise, and may thus be selected for because of inductive biases in the training process. Other researchers believe that scheming and other mesa-optimizing behavior is implausible with the most common current ML architectures, and that the inductive bias argument and other arguments for getting misaligned mesa-optimizers by default (like the counting argument, which suggests that there are many more ...

Link to original articleWelcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Measuring Coherence of Policies in Toy Environments, published by dx26 on March 18, 2024 on LessWrong. This post was produced as part of the Astra Fellowship under the Winter 2024 Cohort, mentored by Richard Ngo. Thanks to Martín Soto, Jeremy Gillien, Daniel Kokotajlo, and Lukas Berglund for feedback. Summary Discussions around the likelihood and threat models of AI existential risk (x-risk) often hinge on some informal concept of a "coherent", goal-directed AGI in the future maximizing some utility function unaligned with human values. Whether and how coherence may develop in future AI systems, especially in the era of LLMs, has been a subject of considerable debate. In this post, we provide a preliminary mathematical definition of the coherence of a policy as how likely it is to have been sampled via uniform reward sampling (URS), or uniformly sampling a reward function and then sampling from the set of policies optimal for that reward function, versus uniform policy sampling (UPS). We provide extensions of the model for sub-optimality and for "simple" reward functions via uniform sparsity sampling (USS). We then build a classifier for the coherence of policies in small deterministic MDPs, and find that properties of the MDP and policy, like the number of self-loops that the policy takes, are predictive of coherence when used as features for the classifier. Moreover, coherent policies tend to preserve optionality, navigate toward high-reward areas of the MDP, and have other "agentic" properties. We hope that our metric can be iterated upon to achieve better definitions of coherence and a better understanding of what properties dangerous AIs will have. Introduction Much of the current discussion about AI x-risk centers around "agentic", goal-directed AIs having misaligned goals. For instance, one of the most dangerous possibilities being discussed is of mesa-optimizers developing within superhuman models, leading to scheming behavior and deceptive alignment. A significant proportion of current alignment work focuses on detecting, analyzing (e.g. via analogous case studies of model organisms), and possibly preventing deception. Some researchers in the field believe that intelligence and capabilities are inherently tied with "coherence", and thus any sufficiently capable AI will approximately be a coherent utility function maximizer. In their paper "Risks From Learned Optimization" formally introducing mesa-optimization and deceptive alignment, Evan Hubinger et al. discuss the plausibility of mesa-optimization occurring in RL-trained models. They analyze the possibility of a base optimizer, such as a hill-climbing local optimization algorithm like stochastic gradient descent, producing a mesa-optimizer model that internally does search (e.g. Monte Carlo tree search) in pursuit of a mesa-objective (in the real world, or in the "world-model" of the agent), which may or may not be aligned with human interests. This is in contrast to a model containing many complex heuristics that is not well-defined internally as a consequentialist mesa-optimizer; one extreme example is a tabular model/lookup table that matches observations to actions, which clearly does not do any internal search or have any consequentialist cognition. They speculate that mesa-optimizers may be selected for because they generalize better than other models, and/or may be more compressible information-theoretic wise, and may thus be selected for because of inductive biases in the training process. Other researchers believe that scheming and other mesa-optimizing behavior is implausible with the most common current ML architectures, and that the inductive bias argument and other arguments for getting misaligned mesa-optimizers by default (like the counting argument, which suggests that there are many more ...

We name our MVPs and MDPs (most disappointing players) for offense, defense, and special teams. We also discuss the new ST hire as well as the DC search. Enjoy!Advertising Inquiries: https://redcircle.com/brandsPrivacy & Opt-Out: https://redcircle.com/privacy

We are joined by Koen Holtman, an independent AI researcher focusing on AI safety. Koen is the Founder of Holtman Systems Research, a research company based in the Netherlands. Koen started the conversation with his take on an AI apocalypse in the coming years. He discussed the obedience problem with AI models and the safe form of obedience. Koen explained the concept of Markov Decision Process (MDP) and how it is used to build machine learning models. Koen spoke about the problem of AGIs not being able to allow changing their utility function after the model is deployed. He shared another alternative approach to solving the problem. He shared how to engineer AGI systems now and in the future safely. He also spoke about how to implement safety layers on AI models. Koen discussed the ultimate goal of a safe AI system and how to check that an AI system is indeed safe. He discussed the intersection between large language Models (LLMs) and MDPs. He shared the key ingredients to scale the current AI implementations.

In the latest episode of Product Team Success, host Ross Webb talks with Apple exec Jonathan Boice about AI and product team success. Here are three key takeaways from this insightful episode: 1. AI can amplify human behavior and experiences, but it can also disregard empathy and consideration for human impact if not used properly. 2. Businesses should aim for MDPs (minimal, delightful products) rather than MVPs (minimal viable products) to create a core product that is already good before adding extra features. 3. Balancing empathy and data-driven decision making is crucial for success in product management, and AI technology can help achieve this balance. Listen now for more insights on AI and product team success from this thought-provoking episode Jonathan Boice is the founder of Digital Visionary, a consulting agency that assists businesses in navigating the world of artificial intelligence (AI) without sacrificing their human touch. Boice recognizes the growing apprehension and excitement surrounding AI and aims to help companies use it to create impactful products and improve people's lives. He prides himself on finding the middle ground between the autonomous business model and the traditional approach, ensuring that AI is used to supplement and not replace human workers. With a focus on the product space, Boice is dedicated to helping companies incorporate AI in a mindful and responsible way.

Patreon: https://www.patreon.com/mlst Discord: https://discord.gg/ESrGqhf5CB Twitter: https://twitter.com/MLStreetTalk In this exclusive interview, Dr. Tim Scarfe sits down with Minqi Jiang, a leading PhD student at University College London and Meta AI, as they delve into the fascinating world of deep reinforcement learning (RL) and its impact on technology, startups, and research. Discover how Minqi made the crucial decision to pursue a PhD in this exciting field, and learn from his valuable startup experiences and lessons. Minqi shares his insights into balancing serendipity and planning in life and research, and explains the role of objectives and Goodhart's Law in decision-making. Get ready to explore the depths of robustness in RL, two-player zero-sum games, and the differences between RL and supervised learning. As they discuss the role of environment in intelligence, emergence, and abstraction, prepare to be blown away by the possibilities of open-endedness and the intelligence explosion. Learn how language models generate their own training data, the limitations of RL, and the future of software 2.0 with interpretability concerns. From robotics and open-ended learning applications to learning potential metrics and MDPs, this interview is a goldmine of information for anyone interested in AI, RL, and the cutting edge of technology. Don't miss out on this incredible opportunity to learn from a rising star in the AI world! TOC Tech & Startup Background [00:00:00] Pursuing PhD in Deep RL [00:03:59] Startup Lessons [00:11:33] Serendipity vs Planning [00:12:30] Objectives & Decision Making [00:19:19] Minimax Regret & Uncertainty [00:22:57] Robustness in RL & Zero-Sum Games [00:26:14] RL vs Supervised Learning [00:34:04] Exploration & Intelligence [00:41:27] Environment, Emergence, Abstraction [00:46:31] Open-endedness & Intelligence Explosion [00:54:28] Language Models & Training Data [01:04:59] RLHF & Language Models [01:16:37] Creativity in Language Models [01:27:25] Limitations of RL [01:40:58] Software 2.0 & Interpretability [01:45:11] Language Models & Code Reliability [01:48:23] Robust Prioritized Level Replay [01:51:42] Open-ended Learning [01:55:57] Auto-curriculum & Deep RL [02:08:48] Robotics & Open-ended Learning [02:31:05] Learning Potential & MDPs [02:36:20] Universal Function Space [02:42:02] Goal-Directed Learning & Auto-Curricula [02:42:48] Advice & Closing Thoughts [02:44:47] References: - Why Greatness Cannot Be Planned: The Myth of the Objective by Kenneth O. Stanley and Joel Lehman https://www.springer.com/gp/book/9783319155234 - Rethinking Exploration: General Intelligence Requires Rethinking Exploration https://arxiv.org/abs/2106.06860 - The Case for Strong Emergence (Sabine Hossenfelder) https://arxiv.org/abs/2102.07740 - The Game of Life (Conway) https://www.conwaylife.com/ - Toolformer: Teaching Language Models to Generate APIs (Meta AI) https://arxiv.org/abs/2302.04761 - OpenAI's POET: Paired Open-Ended Trailblazer https://arxiv.org/abs/1901.01753 - Schmidhuber's Artificial Curiosity https://people.idsia.ch/~juergen/interest.html - Gödel Machines https://people.idsia.ch/~juergen/goedelmachine.html - PowerPlay https://arxiv.org/abs/1112.5309 - Robust Prioritized Level Replay: https://openreview.net/forum?id=NfZ6g2OmXEk - Unsupervised Environment Design: https://arxiv.org/abs/2012.02096 - Excel: Evolving Curriculum Learning for Deep Reinforcement Learning https://arxiv.org/abs/1901.05431 - Go-Explore: A New Approach for Hard-Exploration Problems https://arxiv.org/abs/1901.10995 - Learning with AMIGo: Adversarially Motivated Intrinsic Goals https://www.researchgate.net/publication/342377312_Learning_with_AMIGo_Adversarially_Motivated_Intrinsic_Goals PRML https://www.microsoft.com/en-us/research/uploads/prod/2006/01/Bishop-Pattern-Recognition-and-Machine-Learning-2006.pdf Sutton and Barto https://web.stanford.edu/class/psych209/Readings/SuttonBartoIPRLBook2ndEd.pdf

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: POWERplay: An open-source toolchain to study AI power-seeking, published by Edouard Harris on October 24, 2022 on The AI Alignment Forum. We're open-sourcing POWERplay, a research toolchain you can use to study power-seeking behavior in reinforcement learning agents. POWERplay was developed by Gladstone AI for internal research. POWERplay's main use is to estimate the instrumental value that a reinforcement learning agent can get from a state in an MDP. Its implementation is based on a definition of instrumental value (or "POWER") first proposed by Alex Turner et al. We've extended this definition to cover certain tractable multi-agent RL settings, and built an implementation behind a simple Python API. We've used POWERplay previously to obtain some suggestive early results in single-agent and multi-agent power-seeking. But we think there may be more low-hanging fruit to be found in this area. Beyond our own ideas about what to do next, we've also received some interesting conceptual questions in connection with this work. A major reason we're open-sourcing POWERplay is to lower the cost of converting these conceptual questions into real experiments with concrete outcomes, that can support or falsify our intuitions about instrumental convergence. Ramp-up We've designed POWERplay to make it as easy as possible for you to get started with it. Follow the installation and quickstart instructions to get moving quickly. Use the replication API to trivially reproduce any figure from any post in our instrumental convergence sequence. Design single-agent and multi-agent MDPs and policies, launch experiments on your local machine, and visualize results with clear figures and animations. POWERplay comes with "batteries included", meaning all the code samples in the documentation should just work out-of-the-box if it's been installed successfully. It also comes with pre-run examples of experimental results, so you can understand what "normal" output is supposed to look like. While this does make the repo weigh in at about 500 MB, it's worth the benefits of letting you immediately start playing around with visualizations on preexisting data. If we've done our job right, a smart and curious grad student (with a bit of Python experience) should be able to start reproducing our previous experiments within an hour, and to have some new — and hopefully interesting! — results within a week. We're looking forward to seeing what people do with this. If you have any questions or comments about POWERplay, feel free to reach out to Edouard at edouard@gladstone.ai. Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.

17 - MDPs & Value/Policy Iteration

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Open Problems in Negative Side Effect Minimization, published by Fabian Schimpf on May 6, 2022 on The AI Alignment Forum. Acknowledgments We want to thank Stuart Armstrong, Remmelt Ellen, David Lindner, Michal Pokorny, Achyuta Rajaram, Adam Shimi, and Alex Turner for helpful discussions and valuable feedback on earlier drafts of this post. Fabian Schimpf and Lukas Fluri are part of this year's edition of the AI Safety Camp. Our gratitude goes to the camp organizers: Remmelt Ellen, Sai Joseph, Adam Shimi, and Kristi Uustalu. TLDR; Negative side effects are one class of threats that misaligned AGIs pose to humanity. Many different approaches have been proposed to mitigate or prevent AI systems from having negative side effects. In this post, we present three requirements that a side-effect minimization method (SEM) should fulfill to be applied in the real world and argue that current methods do not yet satisfy these requirements. We also propose future work that could help to solve these requirements. Introduction Avoiding negative side-effects of agents acting in environments has been a core problem in AI safety since the field started to be formalized. Therefore, as part of our AI safety camp project, we took a closer look at state-of-the-art approaches like AUP and Relative Reachability. After months of discussions, we realized that we were confused about how these (and similar methods) could be used to solve problems we care about outside the scope of the typical grid-world environments. We formalized these discussions into distinct desiderata that we believe are currently not sufficiently addressed and, in part, maybe even overlooked. This post attempts to summarize these points and provide structured arguments to support our critique. Of course, we expect to be partially wrong about this, as we updated our beliefs even while writing up this post. We welcome any feedback or additional input to this post. The sections after the summary table and anticipated questions contain our reasoning for the selected open problems and do not need to be read in order. Background The following paragraphs make heavy use of the following terms and side-effect minimization methods (SEMs). For a more detailed explanation we refer to the provided links MDP: A Markov Decision Process is a 5-tuple ⟨S,A,T,R,γ⟩ In the setting of side-effect minimization, the goal generally is to maximize the cumulative reward without causing (negative) side-effects. RR: In its simplest form Stepwise Relative Reachability is an SEM, acting in MDPs, which tries to avoid side-effects by replacing the old reward function R with the compositionr(st,at,st+1)=R(st,at,st+1)−λdRR(st+1,s′t+1) where dRR(st+1,s′t+1)=1|S|∑s∈Smax(R(s′st+1;s)−R(st+1;s),0) is a deviation measure punishing the agent if the average “reachability” of all states of the MDP has been decreased by taking action at compared to taking a baseline action anop (like doing nothing). The idea is that side-effects reduce the reachability of certain states (i.e. breaking a vase makes all states that require an intact vase unreachable) and punishing such a decrease in reachability hence also punishes the agent for side-effects. AUP: Attainable Utility Preservation (see also here and here) is an SEM, acting in MDPs, which tries to avoid side-effects by replacing the old reward function R with the composition r(st,at,st+1)=R(st,at,st+1)−λdAUP(st,at,st+1) where dAUP(st,at,st+1)=1N∑Ri=1|QRi(st,at,st+1−QRi(st,anop,s′t+1)| is a normalized deviation measure punishing the agent if its ability to maximize any of its provided auxiliary reward functions Ri∈R changes by taking action at compared to taking a baseline action anop (like doing nothing). The idea is that the true (side-effect free) reward function (which is very hard to specify) is correlated with many ...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Seeking Power is Often Convergently Instrumental in MDPs , published by TurnTrout, elriggs on the LessWrong. Crossposted from the AI Alignment Forum. May contain more technical jargon than usual. This is a linkpost for In 2008, Steve Omohundro's foundational paper The Basic AI Drives conjectured that superintelligent goal-directed AIs might be incentivized to gain significant amounts of power in order to better achieve their goals. Omohundro's conjecture bears out in toy models, and the supporting philosophical arguments are intuitive. In 2019, the conjecture was even debated by well-known AI researchers. Power-seeking behavior has been heuristically understood as an anticipated risk, but not as a formal phenomenon with a well-understood cause. The goal of this post (and the accompanying paper, Optimal Policies Tend to Seek Power) is to change that. Motivation It's 2008, the ancient wild west of AI alignment. A few people have started thinking about questions like “if we gave an AI a utility function over world states, and it actually maximized that utility... what would it do?" In particular, you might notice that wildly different utility functions seem to encourage similar strategies. Resist shutdown? Gain computational resources? Prevent modification of utility function? Paperclip utility ✔️ ✔️ ✔️ Blue webcam pixel utility ✔️ ✔️ ✔️ People-look-happy utility ✔️ ✔️ ✔️ These strategies are unrelated to terminal preferences: the above utility functions do not award utility to e.g. resource gain in and of itself. Instead, these strategies are instrumental: they help the agent optimize its terminal utility. In particular, a wide range of utility functions incentivize these instrumental strategies. These strategies seem to be convergently instrumental. But why? I'm going to informally explain a formal theory which makes significant progress in answering this question. I don't want this post to be Optimal Policies Tend to Seek Power with cuter illustrations, so please refer to the paper for the math. You can read the two concurrently. We can formalize questions like “do ‘most' utility maximizers resist shutdown?” as “Given some prior beliefs about the agent's utility function, knowledge of the environment, and the fact that the agent acts optimally, with what probability do we expect it to be optimal to avoid shutdown?” The table's convergently instrumental strategies are about maintaining, gaining, and exercising power over the future, in some sense. Therefore, this post will help answer: What does it mean for an agent to “seek power”? In what situations should we expect seeking power to be more probable under optimality, than not seeking power? This post won't tell you when you should seek power for your own goals; this post illustrates a regularity in optimal action across different goals one might pursue. Formalizing Convergent Instrumental Goals suggests that the vast majority of utility functions incentivize the agent to exert a lot of control over the future, assuming that these utility functions depend on “resources.” This is a big assumption: what are “resources”, and why must the AI's utility function depend on them? We drop this assumption, assuming only unstructured reward functions over a finite Markov decision process (MDP), and show from first principles how power-seeking can often be optimal. Formalizing the Environment My theorems apply to finite MDPs; for the unfamiliar, I'll illustrate with Pac-Man. Full observability: You can see everything that's going on; this information is packaged in the state s. In Pac-Man, the state is the game screen. Markov transition function: the next state depends only on the choice of action a and the current state s. It doesn't matter how we got into a situation. Discounted reward: future rewards get geometrically discoun...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Seeking Power is Often Convergently Instrumental in MDPs, published by Paul Christiano on the AI Alignment Forum. (Thanks to Ajeya Cotra, Nick Beckstead, and Jared Kaplan for helpful comments on a draft of this post.) I really don't want my AI to strategically deceive me and resist my attempts to correct its behavior. Let's call an AI that does so egregiously misaligned (for the purpose of this post). Most possible ML techniques for avoiding egregious misalignment depend on detailed facts about the space of possible models: what kind of thing do neural networks learn? how do they generalize? how do they change as we scale them up? But I feel like we should be possible to avoid egregious misalignment regardless of how the empirical facts shake out--it should be possible to get a model we build to do at least roughly what we want. So I'm interested in trying to solve the problem in the worst case, i.e. to develop competitive ML algorithms for which we can't tell any plausible story about how they lead to egregious misalignment. This is a much higher bar for an algorithm to meet, so it may just be an impossible task. But if it's possible, there are several ways in which it could actually be easier: We can potentially iterate much faster, since it's often easier to think of a single story about how an algorithm can fail than it is to characterize its behavior in practice. We can spend a lot of our time working with simple or extreme toy cases that are easier to reason about, since our algorithm is supposed to work even in these cases. We can find algorithms that have a good chance of working in the future even if we don't know what AI will look like or how quickly it will advance, since we've been thinking about a very wide range of possible failure cases. I'd guess there's a 25–50% chance that we can find an alignment strategy that looks like it works, in the sense that we can't come up with a plausible story about how it leads to egregious misalignment. That's a high enough probability that I'm very excited to gamble on it. Moreover, if it fails I think we're likely to identify some possible “hard cases” for alignment — simple situations where egregious misalignment feels inevitable. What this looks like (3 examples) My research basically involves alternating between “think of a plausible alignment algorithm” and “think of a plausible story about how it fails.” Example 1: human feedback In an unaligned benchmark I describe a simple AI training algorithm: Our AI observes the world through a bunch of cameras and outputs motor actions. We train a generative model that predicts these camera observations given the motor actions. We ask humans to evaluate possible futures by looking at the predicted videos output by the model. We then train a model to predict these human evaluations. At test time the AI searches for plans that lead to trajectories that look good to humans. In the same post, I describe a plausible story about how this algorithm leads to egregious misalignment: Our generative model understands reality better than human evaluators. There are plans that acquire influence in ways that are obvious to the generative model but completely incomprehensible and invisible to humans. It's possible to use that influence to “hack” the cameras, in the sense of creating a fiction that looks convincing to a human looking at predicted videos. The fiction can look much better than the actual possible futures. So our planning process finds an action that covertly gathers resources and uses them to create a fiction. I don't know if or when this kind of reward hacking would happen — I think it's pretty likely eventually, but it's far from certain and it might take a long time. But from my perspective this failure mode is at least plausible — I don't see any contradictions between ...

Recorded by Robert Miles: http://robertskmiles.com More information about the newsletter here: https://rohinshah.com/alignment-newsletter/ YouTube Channel: https://www.youtube.com/channel/UCfGGFXwKpr-TJ5HfxEFaFCg   HIGHLIGHTS Collaborating with Humans without Human Data (DJ Strouse et al) (summarized by Rohin): We've previously seen that if you want to collaborate with humans in the video game Overcooked, it helps to train a deep RL agent against a human model (AN #70), so that the agent “expects” to be playing against humans (rather than e.g. copies of itself, as in self-play). We might call this a “human-aware” model. However, since a human-aware model must be trained against a model that imitates human gameplay, we need to collect human gameplay data for training. Could we instead train an agent that is robust enough to play with lots of different agents, including humans as a special case? This paper shows that this can be done with Fictitious Co-Play (FCP), in which we train our final agent against a population of self-play agents and their past checkpoints taken throughout training. Such agents get significantly higher rewards when collaborating with humans in Overcooked (relative to the human-aware approach in the previously linked paper). In their ablations, the authors find that it is particularly important to include past checkpoints in the population against which you train. They also test whether it helps to have the self-play agents have a variety or architectures, and find that it mostly does not make a difference (as long as you are using past checkpoints as well). Read more: Related paper: Maximum Entropy Population Based Training for Zero-Shot Human-AI Coordination Rohin's opinion: You could imagine two different philosophies on how to build AI systems -- the first option is to train them on the actual task of interest (for Overcooked, training agents to play against humans or human models), while the second option is to train a more robust agent on some more general task, that hopefully includes the actual task within it (the approach in this paper). Besides Overcooked, another example would be supervised learning on some natural language task (the first philosophy), as compared to pretraining on the Internet GPT-style and then prompting the model to solve your task of interest (the second philosophy). In some sense the quest for a single unified AGI system is itself a bet on the second philosophy -- first you build your AGI that can do all tasks, and then you point it at the specific task you want to do now. Historically, I think AI has focused primarily on the first philosophy, but recent years have shown the power of the second philosophy. However, I don't think the question is settled yet: one issue with the second philosophy is that it is often difficult to fully “aim” your system at the true task of interest, and as a result it doesn't perform as well as it “could have”. In Overcooked, the FCP agents will not learn specific quirks of human gameplay that could be exploited to improve efficiency (which the human-aware agent could do, at least in theory). In natural language, even if you prompt GPT-3 appropriately, there's still some chance it ends up rambling about something else entirely, or neglects to mention some information that it “knows” but that a human on the Internet would not have said. (See also this post (AN #141).) I should note that you can also have a hybrid approach, where you start by training a large model with the second philosophy, and then you finetune it on your task of interest as in the first philosophy, gaining the benefits of both. I'm generally interested in which approach will build more useful agents, as this seems quite relevant to forecasting the future of AI (which in turn affects lots of things including AI alignment plans).   TECHNICAL AI ALIGNMENT LEARNING HUMAN INTENT Inverse Decision Modeling: Learning Interpretable Representations of Behavior (Daniel Jarrett, Alihan Hüyük et al) (summarized by Rohin): There's lots of work on learning preferences from demonstrations, which varies in how much structure they assume on the demonstrator: for example, we might consider them to be Boltzmann rational (AN #12) or risk sensitive, or we could try to learn their biases (AN #59). This paper proposes a framework to encompass all of these choices: the core idea is to model the demonstrator as choosing actions according to a planner; some parameters of this planner are fixed in advance to provide an assumption on the structure of the planner, while others are learned from data. This also allows them to separate beliefs, decision-making, and rewards, so that different structures can be imposed on each of them individually. The paper provides a mathematical treatment of both the forward problem (how to compute actions in the planner given the reward, think of algorithms like value iteration) and the backward problem (how to compute the reward given demonstrations, the typical inverse reinforcement learning setting). They demonstrate the framework on a medical dataset, where they introduce a planner with parameters for flexibility of decision-making, optimism of beliefs, and adaptivity of beliefs. In this case they specify the desired reward function and then run backward inference to conclude that, with respect to this reward function, clinicians appear to be significantly less optimistic when diagnosing dementia in female and elderly patients. Rohin's opinion: One thing to note about this paper is that it is an incredible work of scholarship; it fluently cites research across a variety of disciplines including AI safety, and provides a useful organizing framework for many such papers. If you need to do a literature review on inverse reinforcement learning, this paper is a good place to start. Human irrationality: both bad and good for reward inference (Lawrence Chan et al) (summarized by Rohin): Last summary, we saw a framework for inverse reinforcement learning with suboptimal demonstrators. This paper instead investigates the qualitative effects of performing inverse reinforcement learning with a suboptimal demonstrator. The authors modify different parts of the Bellman equation in order to create a suite of possible suboptimal demonstrators to study. They run experiments with exact inference on random MDPs and FrozenLake, and with approximate inference on a simple autonomous driving environment, and conclude: 1. Irrationalities can be helpful for reward inference, that is, if you infer a reward from demonstrations by an irrational demonstrator (where you know the irrationality), you often learn more about the reward than if you inferred a reward from optimal demonstrations (where you know they are optimal). Conceptually, this happens because optimal demonstrations only tell you about what the best behavior is, whereas most kinds of irrationality can also tell you about preferences between suboptimal behaviors. 2. If you fail to model irrationality, your performance can be very bad, that is, if you infer a reward from demonstrations by an irrational demonstrator, but you assume that the demonstrator was Boltzmann rational, you can perform quite badly. Rohin's opinion: One way this paper differs from my intuitions is that it finds that assuming Boltzmann rationality performs very poorly if the demonstrator is in fact systematically suboptimal. I would have instead guessed that Boltzmann rationality would do okay -- not as well as in the case where there is no misspecification, but only a little worse than that. (That's what I found in my paper (AN #59), and it makes intuitive sense to me.) Some hypotheses for what's going on, which the lead author agrees are at least part of the story: 1. When assuming Boltzmann rationality, you infer a distribution over reward functions that is “close” to the correct one in terms of incentivizing the right behavior, but differs in rewards assigned to suboptimal behavior. In this case, you might get a very bad log loss (the metric used in this paper), but still have a reasonable policy that is decent at acquiring true reward (the metric used in my paper). 2. The environments we're using may differ in some important way (for example, in the environment in my paper, it is primarily important to identify the goal, which might be much easier to do than inferring the right behavior or reward in the autonomous driving environment used in this paper). FORECASTING Forecasting progress in language models (Matthew Barnett) (summarized by Sudhanshu): This post aims to forecast when a "human-level language model" may be created. To build up to this, the author swiftly covers basic concepts from information theory and natural language processing such as entropy, N-gram models, modern LMs, and perplexity. Data for perplexity achieved from recent state-of-the-art models is collected and used to estimate - by linear regression - when we can expect to see future models score below certain entropy levels, approaching the hypothesised entropy for the English Language. These predictions range across the next 15 years, depending which dataset, method, and entropy level is being solved for; there's an attached python notebook with these details for curious readers to further investigate. Preemptly disjunctive, the author concludes "either current trends will break down soon, or human-level language models will likely arrive in the next decade or two." Sudhanshu's opinion: This quick read provides a natural, accessible analysis stemming from recent results, while staying self-aware (and informing readers) of potential improvements. The comments section too includes some interesting debates, e.g. about the Goodhart-ability of the Perplexity metric. I personally felt these estimates were broadly in line with my own intuitions. I would go so far as to say that with the confluence of improved generation capabilities across text, speech/audio, video, as well as multimodal consistency and integration, virtually any kind of content we see ~10 years from now will be algorithmically generated and indistinguishable from the work of human professionals. Rohin's opinion: I would generally adopt forecasts produced by this sort of method as my own, perhaps making them a bit longer as I expect the quickly growing compute trend to slow down. Note however that this is a forecast for human-level language models, not transformative AI; I would expect these to be quite different and would predict that transformative AI comes significantly later. MISCELLANEOUS (ALIGNMENT) Rohin Shah on the State of AGI Safety Research in 2021 (Lucas Perry and Rohin Shah) (summarized by Rohin): As in previous years (AN #54), on this FLI podcast I talk about the state of the field. Relative to previous years, this podcast is a bit more introductory, and focuses a bit more on what I find interesting rather than what the field as a whole would consider interesting. Read more: Transcript   NEAR-TERM CONCERNS RECOMMENDER SYSTEMS User Tampering in Reinforcement Learning Recommender Systems (Charles Evans et al) (summarized by Zach): Large-scale recommender systems have emerged as a way to filter through large pools of content to identify and recommend content to users. However, these advances have led to social and ethical concerns over the use of recommender systems in applications. This paper focuses on the potential for social manipulability and polarization from the use of RL-based recommender systems. In particular, they present evidence that such recommender systems have an instrumental goal to engage in user tampering by polarizing users early on in an attempt to make later predictions easier. To formalize the problem the authors introduce a causal model. Essentially, they note that predicting user preferences requires an exogenous variable, a non-observable variable, that models click-through rates. They then introduce a notion of instrumental goal that models the general behavior of RL-based algorithms over a set of potential tasks. The authors argue that such algorithms will have an instrumental goal to influence the exogenous/preference variables whenever user opinions are malleable. This ultimately introduces a risk for preference manipulation. The author's hypothesis is tested using a simple media recommendation problem. They model the exogenous variable as either leftist, centrist, or right-wing. User preferences are malleable in the sense that a user shown content from an opposing side will polarize their initial preferences. In experiments, the authors show that a standard Q-learning algorithm will learn to tamper with user preferences which increases polarization in both leftist and right-wing populations. Moreover, even though the agent makes use of tampering it fails to outperform a crude baseline policy that avoids tampering. Zach's opinion: This article is interesting because it formalizes and experimentally demonstrates an intuitive concern many have regarding recommender systems. I also found the formalization of instrumental goals to be of independent interest. The most surprising result was that the agents who exploit tampering are not particularly more effective than policies that avoid tampering. This suggests that the instrumental incentive is not really pointing at what is actually optimal which I found to be an illuminating distinction.   NEWS OpenAI hiring Software Engineer, Alignment (summarized by Rohin): Exactly what it sounds like: OpenAI is hiring a software engineer to work with the Alignment team. BERI hiring ML Software Engineer (Sawyer Bernath) (summarized by Rohin): BERI is hiring a remote ML Engineer as part of their collaboration with the Autonomous Learning Lab at UMass Amherst. The goal is to create a software library that enables easy deployment of the ALL's Seldonian algorithm framework for safe and aligned AI. AI Safety Needs Great Engineers (Andy Jones) (summarized by Rohin): If the previous two roles weren't enough to convince you, this post explicitly argues that a lot of AI safety work is bottlenecked on good engineers, and encourages people to apply to such roles. AI Safety Camp Virtual 2022 (summarized by Rohin): Applications are open for this remote research program, where people from various disciplines come together to research an open problem under the mentorship of an established AI-alignment researcher. Deadline to apply is December 1st. Political Economy of Reinforcement Learning schedule (summarized by Rohin): The date for the PERLS workshop (AN #159) at NeurIPS has been set for December 14, and the schedule and speaker list are now available on the website. FEEDBACK I'm always happy to hear feedback; you can send it to me, Rohin Shah by replying to this email. PODCAST An audio podcast version of the Alignment Newsletter is available. This podcast is an audio version of the newsletter, recorded by Robert Miles (http://robertskmiles.com). Subscribe here:

Vikas Srivastava, Chief Revenue Officer, IntegralIntegral recently surveyed 94 heads of FX trading and senior FX managers, reviewing the global effects of the Covid-19 pandemic on the FX business. Among the survey's findings, 28% of respondents said they will operate their FX technology completely in the cloud within the next 5 years, compared to just 2% at present. In distribution channels, 46% confirmed that multi-dealer platforms (MDPs) will see the biggest rise in trading in the next 12 months. Robin Amlôt of IBS Intelligence speaks to Vikas Srivastava, Chief Revenue Officer, Integral, about the tech trends in FX trading.

If you’re thinking about buying a Kia Seltos - here’s everything you need to know. You can help support the podcast, securely via PayPal: As usual, Kia conscripted its on-call dynamics wizard to do his mad, Jedi voodoo and turn the conventional vomit-spec South Korean suspension into what is actually an outstanding platform to drive on our preposterously crap ‘Strayan roads. The drive program on the launch was on mainly these B and C roads around Noosa, and I’d have to say the body control and steering feedback is excellent. So, big tick there. There was probably 90 minutes of freeway driving as well - it’s quiet and composes at 110. Interestingly enough - this vehicle has a next-generation motor driven power steering assistance system. That means an electrical servo motor provides the steering assistance. It detects input from you, and a computer tells it how much to help. That’s when you’re turning in. But when you’re on the way out of a bend, MDPS typically defaults to ‘off’ and the self-centring steering effect you feel (If any) is just mechanical control feedback. But in this system, the motor also provides self-centring feedback assistance. It’s really excellent. Here’s the range. You get S, Sport, Sport+ and GT-Line in order of increasing appeal and price. 2.0-litre CVT only on S and Sport. 1.6 Turbo only on GT-Line. But you can have either engine in Sport+. So the fuel economy powertrain is available in the first three variants. The performance powertrain on the top two. They overlap at Sport+. Here’s how you tell the four variants apart like an automotive ninja. (This is gunna help at the dealership when they jam one under your snout for a test drive - if you know this, you cannot be bullshat to about which one you’re driving. And before you say it in the comments: ‘bullshat’ is the past participle of the verb ‘to bullshit’.) The poverty S model rolls on steel wheels. That’s dead easy to spot. If you’re looking at a Seltos with alloy wheels and a folding key, it’s a Sport. If it’s got 17-inch alloys and a pushbutton start it’s Sport+ and if it’s got 18-inch alloys (with a bright red highlight around the hub) and a head-up display, it’s a GT-Line. There’s more safety gear on Sport+ and GT-Line, but you can get that on S and Sport for $1000 as an option. So, I’m not going to bore you with the spec sheet - but the salient observations arising from the spec sheet are: S is a real poverty pack. Anything that can be removed to cut costs basically has been, and this is done primarily to appease the great cheapskates of the automotive universe: Fleet managers. It’s a big step - $3500 - to go from S to Sport, but it’s well worth it for a private owner. You get alloys, a full-sized spare, the big centre infotainment screen, SUNA live traffic and 10 years of free mapcare updates (and, I’m assured, there are no strings attached to that - you just get the updates when they’re available). Sport+ is the pick of the range - because you get adaptive cruise and the better safety gear standard. Plus front parking sensors, nicer interior, proximity key. And it’s $5500 cheaper than GT-Line, which is loaded with all the nice toys, certainly, but do you really need all that stuff? Probably not. I’d strongly suggest you buy the 1.6 turbo if sporty engaging driving matters to you. The CVT that goes with the 2.0-litre is a little bit frustrating for enthusiastic driving. It displays this noticeable re-engagement lag, getting on the gas when you clip an apex and want to start feeding the power on smoothly.

My guest today is Kanaad. He is a friend of mine from Berkeley, and he is a Cavs and RL enthusiast. We talk about a fascinating paper from the 2018 Sloan Sports Analytics Conference. The main idea is that by modeling each possession as a Markov Decision Process, it becomes easy to build a simulator that can help us answer questions like "How much more efficient would a team be if they took 20% fewer midrange shots early in the shot clock?" The paper: http://www.lukebornn.com/papers/sandholtz_ssac_2018.pdf

In a previous episode, we discussed Markov Decision Processes or MDPs, a framework for decision making and planning. This episode explores the generalization Partially Observable MDPs (POMDPs) which are an incredibly general framework that describes most every agent based system.

Lecture by Professor Andrew Ng for Machine Learning (CS 229) in the Stanford Computer Science department. Professor Ng discusses the topic of reinforcement learning, focusing particularly on continuous state MDPs and discretization.

Lecture by Professor Andrew Ng for Machine Learning (CS 229) in the Stanford Computer Science department. Professor Ng discusses the topic of reinforcement learning, focusing particularly on MDPs, value functions, and policy and value iteration.