POPULARITY
#732 Show Notes:https://wetflyswing.com/732 Presented By: Grand Teton Fly Fishing, Four Wheel Campers, Pescador on the Fly In this episode, Chad Johnson, a legendary guide on the White River, shares his best tips for targeting big browns with streamers. You'll hear a classic Dave Whitlock story, learn when the big fish bite, and find out why Chad still loves the area—even after the famous Dally's Fly Shop moved on. Plus, there's a big secret dropping today about a brand-new podcast. Don't miss it! Show Notes: https://wetflyswing.com/732
Fundraiser at Small Acres Cyder this weekend: See omnystudio.com/listener for privacy information.
Send us a textDally Ward is the President at Bonner Carrington. Bonner Carrington is a vertically integrated, privately owned and operated commercial real estate development, construction and property management company.I'm Moshe Crane connect with me on LinkedIn. My day job is the VP of Branding and Strategic Initiatives at Sage Ventures. Check out my newsletter Zag.Sage Ventures is a commercial real estate firm based in Baltimore, MD. The company buys and operates multifamily rental properties. The company also builds and develops homes that we sell.
California mom Sherri Dally's routine Monday Target run took a shocking turn when a woman flashing a badge approached her in the Ventura County store's parking lot. Wooga Junes Journey : https://wooga-junes-journey.onelink.me/M4rK/de6f3d47?c=jj_us_mobile_proseeders_uspodcast_killerinstinct_jan25 Get exclusive Killer Instinct content on my patreon : https://www.patreon.com/killerinstinct If you like the show, telling a friend about it would be helpful! You can text, email, Tweet, or send this link to a friend: http://bit.ly/KillerInstinctPod Follow Savannah on IG: @savannahbrymer Follow Savannah on Twitter: @savannahbrymer Learn more about your ad choices. Visit podcastchoices.com/adchoices
There is so much conversation today about the need to have great company culture, which is fantastic. However, the work still needs to get done. Dally shares how to bridge that gap between people and performance. If you're exploring ways to create a welcoming and engaging community, events can be powerful tools for fostering connections. But not all events work the same way. Micah and Bella have specific types of events they've found most successful in drawing people from the fringes into deeper relationships.
www.OCC.edu - Ozark Christian College - Joplin, Missouri - 417.626.1234
Lefty's barge pulls into Shell Island and the crew splits up. Drip stays on the Float with Lefty and his family, while Mr. 4 and Dally take a trip inland to the elegant resort called the Conch. But what's behind the Skreegull attacks that have the villagers so fearful, and why don't the Vushtin stay on the island at night? Instead of our regular promo, we'd love it if listeners reach out to some LA institutions helping those who've lost everything in the fires. Here are the links:World Central Kitchen: https://wck.org/LA Dream Center: https://www.dreamcenter.org/LA Regional Foodbank: https://secure.lafoodbank.org/
Inspired by NVIDIA's announcements at CES, we are looking back at one of our favorite episodes. The explosion of generative AI-powered technologies has forever changed the tech landscape. But the path to the current AI revolution was paved by many breakthroughs. Bill Dally had a hand in quite a few of them. In this week's episode of Generative Now, we are revisiting a conversation Lightspeed Partner and host Michael Mignano had with Bill Dally, NVIDIA's Chief Scientist and Senior Vice President for Research, whose nearly four-decade long career spans from experiments with neural networks in the 1980s to leading NVIDIA's groundbreaking research. He discusses the work at NVIDIA in advancing GPU technology, as well as the shift towards generative AI. Michael and Bill also explore the challenges of keeping up with rapid advancements in AI, the future implications for the industry, and advice for aspiring engineers. Bill Dally joined NVIDIA in January 2009 as chief scientist, after spending 12 years at Stanford University, where he was chairman of the computer science department. He has published over 250 papers, holds over 120 issued patents, and is an author of four textbooks. Bill received a bachelor's degree in Electrical Engineering from Virginia Tech, a master's in Electrical Engineering from Stanford University and a Ph.D. in Computer Science from CalTech. Episode Chapters (00:00) Introduction (00:22) Bill Dally's Journey from Neural Networks to NVIDIA (01:39) The Evolution of AI and Computing: A Personal Account (05:48) The AI Revolution: Expectations vs. Reality (08:58) Inside NVIDIA: The Role of Chief Scientist and the Power of Research (11:30) Exploring the Frontiers of Generative AI and Research (15:24) AI's Role in the Future of Autonomous Vehicles (21:54) The Impact of AI on Chip Design and Efficiency (26:56) Building NVIDIA's Elite Research Team (37:01) Anticipating the Future: Advice for the Next Generation (40:59) Closing Thoughts Stay in touch: www.lsvp.com X: https://twitter.com/lightspeedvp LinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/ Instagram: https://www.instagram.com/lightspeedventurepartners/ Subscribe on your favorite podcast app: generativenow.co The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. For more details please see lsvp.com/legal.
durée : 00:20:14 - Galette des rois de la Maison Dally à Uzos - 3ème au Concours départemental de la meilleure galette feuilletée - Unique représentant béarnais sur le podium du concours, la boulangerie met en avant une tradition et une qualité qui séduisent les amateurs de galettes des rois.
Well, folks, the Chums have flown away, and our cartographers are stuck on ground duty. Don't feel too broken up, though! While the vertical may be abandoned for now, we'll trace and retrace Middle America alongside Merle Rideout and begin zooming in on Dally, a true heroine-to-be. Alchemy, photography, lovelornity, and more, await our kind listeners within this episode of Mapping the Zone.Next episode: Pages 81-96 (ending on “Veikko was pouring vodka. 'Happy Fourth of July, Webb.'”)If you like what we're doing and want to support the show, please consider making a donation on Ko-Fi. Funds we receive will be used to upgrade equipment, pay hosting fees, and help make the show better.https://ko-fi.com/mappingthezoneIf you enjoyed our discussion, please check out the following media that relates to these chapters:Films/TV: Godland (2022; dir. Hlynur Pálmason)Books/Authors: The Public Burning by Robert CooverAs always, thanks so much for listening!Email: mappingthezonepod@gmail.comBluesky: https://bsky.app/profile/mappingthezone.bsky.socialTwitter: https://twitter.com/pynchonpodInstagram:https://www.instagram.com/mappingthezonepodcast/
Episode 108 – End of Year Update Going through state exam questions to help real estate students pass their state exam. Update on law school finals Tips on how to study Suggestions for real world issues in Secured Transactions, Trusts and Estates, and Income Tax List of recent grads: Latoya, Sandy, Denise, Rachel, Shelby, Michael, Erika, Brittany, James, John, Femi, Dally, Rosalind, Kayla, Lori, Brianna, Reese, Brian, Melissa, Matt, Charles, Jessica, Alyssa, Blessing, Grace, Meranda, Devin, Jeff, Gabriela, Ethan, Eddie, Cole, Heidi, Mark, Leidary, Veronica, Courtney, Kaylen, Zoe, Ruth, Evelyn, Jaquelyn, Dana, Caleb, Connor, Breondra, Kylena, Shelley, Mohammad, Jessica, Delaney, Lacey, Shana, April, Johana, Jaxson, Janie, Lancia, Iris, Tania, Peyton, Sheryl, Kyle, Shawn, Caitlin, Della, Renn, Sem, Kevin, Brooke, Andrea, Danna, Wendy, Joseph, Karigan, Ashia, Jasmine, Martin, Sheldon, Whitney, Denise, Jeff, Justin, Sali, Lynn (happy picture on FB page), and Daniel. Listener messages from Sandy in New Jersey, Rachael in Washington State, and Denise in Ohio. Go to www.ahareep.com and sign up for the program for only $35, use discount code: legend15 to save 15% off. Go to www.indianarealestateinstitute.com for Indiana real estate classes. A-Ha LINKS Email info@ahareep.com Web www.ahareep.com Facebook https://www.facebook.com/AHA.REEP YouTube https://www.youtube.com/channel/UCrxAjI5Li4Ll3Epwcyc0i6A
It's our last and final day at the Ag Gear Booth, live from the Las Vegas Convention Center during the 2024 NFR!In this episode, we reflect on an incredible week—sharing the highs, the lows, and even our loss in the fantasy tattoo league. We also welcome our final guest, the one and only Dally Boy!Dally Boy has become a beloved figure in the western community, known for his viral social media content showcasing the day-to-day life of a working cowboy, along with his iconic classic cowboy style. (Who doesn't love a good Dally Boy fit check?)Based out of a ranch in Tucson, Arizona, Dally shares stories about working through extreme heat and harsh conditions, how he ended up in this role, and the responsibility we all have in inspiring and guiding those who dream of becoming cowboys and day workers.Follow Dally Boy
Hit internet sensations are here! It's Spanky, the Pony That Hates and his good friend Dally, to ride him. Will they do anything else for 90 minutes? Of course not!
Retiring Capitola Police Chief Andy Dally returns to the show
Its Grand Final week! We have a look at both the Melbourne Storm and Penrith Panthers, we talk about the historic implications this game has, and we tell you who we think will win and why. We chat about the Dally M winners and losers, a little about Super League, and go off in a thousand different tangents just the way you like it! Twitter AndrewRLP and LeagueFreak Patreon Rugby League Project Patreon and the League Freak Patreon! Websites The Rugby League Project League Freak NRL and Rugby League News Rugby League Podcasting Network NRL Rumours NRL Breaking News Podcast Links Site: FergoandTheFreak.com Twitter: Fergo And The Freak on Twitter Instagram: Fergo and The Freak on Instagram Youtube: Fergo and The Freak on Youtube Youve found the best 2024 NRL Podcast! The Official NRL website, For the latest NRL News or the 2025 NRL Draw just click the links! Also if you're looking to Buy 2025 NRL Tickets you know where to go!
A joint mix with artist Charline Dally broadcast on Radio AlHara, Sept. 24, 2024. Track listing below, artist name(s) and then track name. Original artwork by Charline. 01. Jad Atoui, The Last Bloom 02. Yu Su, Touch-Me-Not 03. VAMPI + Nadah El Shazly, Myst 04. Arca, Fossil 05. Kamilya Jubran + Werner Hasler, Gharbeebah 06. Vivian Li + Thomas Augustin, Éclipse 07. Büşra Kayıkçı, Qarib 08. Jordan Christoff, Waves 09. DahL, Plesiosaur 10. koeosaeme, III 11. Relative Resonance, Feat. Kris Davis, Chris Speed, Chris Tordini & Devin Gray - Notester
Merriam-Webster's Word of the Day for September 17, 2024 is: dally DAL-ee verb The word dally has a number of meanings. To dally can be to physically linger or dawdle, or to waste time. Dally may also mean "to act playfully," especially in a romantic sense, or "to deal with something lightly or in a way that is not serious." // Three members of the hiking group were dallying and didn't arrive at the overlook until others were already starting to head back to the trailhead. // Alton has been dallying with the idea of starting a bakery. See the entry > Examples: "Just as businesses that dallied too long before moving into the era of computing lost ground and eventually faded away, companies that delay in adopting the technologies of the future will find it impossible to keep up with those that take the necessary steps quickly." — Pritom Das, Entrepreneur, 21 May 2021 Did you know? English speakers have been futzing around with dally since the late Middle Ages. They first started using it to mean "to chat," which was also the meaning of dalier, the Anglo-French word whence it came, but this sense fell into disuse. Next, they applied it to acting playfully with someone especially in amorous and flirtatious ways (the noun dalliance, meaning "an act of dallying," is to this day often used for situations where people get all smoochy and whatnot). The idea of more figurative flirtatiousness soon led to a sense of dally meaning "to deal with lightly or in a way that is not serious." Finally, by the mid-16th century, perhaps because fuddy-duddies saw all of this fun and frivolity as a waste of time, dally gained the additional meanings of "to waste time" and "to dawdle."
We didn't think celeb breakups could get bigger than Maya and Stormzy, or Bennifer, but now Molly Mae and Tommy Fury have just proven that this very much is the year of the celeb breakup (and it's killing us!!). We try to understand why we're all taking it so personally, and assess the very icky rumours surrounding Tommy's recent trip to Macedonia as well as the friction between Molly's boss bitch success and the Fury's traditional values. We're also devo to report that our faves Barry Keoghan and Sabrina Carpenter may have joined the heartbreak train. Plus, Ellie gives us the verdict on Zoe Kravitz's buzzy directorial debut Blink Twice, a sinister horror film seemingly inspired by Jeffrey Epstein's ‘paedo island'. And of course we had to debrief on Dolly Aderton's interview with the Sunday Times (anyone else have a Dolly Google Alert..?). Enjoy hunnys! DM us your thoughts on Instagram @straightuppod, or email us at hello@straightuppodcast.co.uk and as ever please leave us a review on Apple Podcasts and a rating on Spotify, lysm! Thanks so much to our amazing partner Whitebox Cocktails, who make the tastiest canned cocktails we have ever tried, from Maragritas and Negronis, frozen Martinis to Old Fashioneds. GET 20% OFF with our code SU20 at https://whiteboxcocktails.com/ Reccs Short N'Sweet by Sabrina Carpenter Chicken Shop Date with Sabrina Carpenter, YouTube At Home with the Furys, Netflix Dolly Alderton on fashion: There's more fun to be had in heels, The Times (and have a listen to our big Dolly Alderton deepdive from Nov 2023!) Blink Twice (in cinemas) Blink Twice is guilty of a vile horror trope, Slate Promising Young Woman, Amazon Prime Video The Lost City, Netflix Inside Out 2, Disney + Big Boys, Channel 4 The Conversation We Should Be Having Around It Ends With Us, The Cut We are so excited to have finally announced our live event with The Trouble Club on 26th September at The Dally in Islington. Get your ticket for £15 with the code STRAIGHTUP50 here: https://www.thetroubleclub.com/events/straight-up-live-podcast-recording
We're talking books this week, hunnies, as what better time of year than to get stuck into some summer reads than holiday szn? From the hottest books of 2024 (think: Butter, Blue Sisters and Brotherless Night) to the all-time faves we're still thinking about years on, these are the titles we suggest adding to your bucket-list immediately. Thank you so much for your reccs so far, we've added the ones we didn't get a chance to talk about today below and will keep on ploughing through, please do send more! We are so excited to have finally announced our live event with The Trouble Club on 26th September at The Dally in Islington. Get your ticket for £15 with the code STRAIGHTUP50 here: https://www.thetroubleclub.com/events/straight-up-live-podcast-recording DM us your thoughts on Instagram @straightuppod, or email us at hello@straightuppodcast.co.uk and as ever please leave us a review on Apple Podcasts and a rating on Spotify, lysm! Thanks so much to our amazing partners: Whitebox Cocktails, who make the tastiest canned cocktails we have ever tried, from Maragritas and Negronis, frozen Martinis to Old Fashioneds. GET 20% OFF with our code SU20 at https://whiteboxcocktails.com/ Incite Nutrition, a UK based supplement company making our favourite Vitamin D spray and biotin (plus marine collagen and lots more). Incredibly reasonably priced and amazing quality. GET 15% OFF via Amazon with the code STRAIGHTUP at https://amzn.to/4eY4odu Books / culture discussed: Butter by Asako Yuzuki Slutty Chef column, Vogue Blue Sisters by Coco Mellors Banal Nightmare by Halle Butler Brotherless Night by V. V. Ganeshananthan A Little Dust on the Eyes by Minoli Salgado Anil's Ghost by Michael Ondaatje The Island of the Missing Trees by Elif Shafak 10 Minutes 38 Seconds in This Strange World by Elif Shafak Shuggie Bain by Douglas Stuart Young Mungo by Douglas Stuart Sociopath: A Memoir by Patric Gagnee Kill Your Friends by John Niven The Psychopath Test by Jon Ronson Reach For the Stars by Michael Cragg Educated by Tara Westover How Lucky Blue and Nara Aziza Smith Made Viral Internet Fame From Scratch, GQ The School for Good Mothers by Jessamine Chan What Tweens Get from Sephora and What They Get from Us, Jia Tolentino for the New Yorker Patricia Lockwood No one is Talking About This The Unwilding by Marina Kemp You Are Here by David Nicholls Us by David Nicholls The Husbands by Holly Gramazio Sandwich by Catherine Newman Margot's Got Money Troubles by Rufi Thorpe Caledonian Road by Andrew O'Hagen Doppelganger by Naomi Klein Experience by Kate Young Mrs S by K Patrick
Jim Haynes talks to Michael McLaren about business-woman and fashion model, June Dally-Watkins.See omnystudio.com/listener for privacy information.
Vortex Optics - Industry Leading Scopes, Rangefinders, and Binoculars Moultrie Mobile - The ONLY Cellular Trail Cameras with A.I. Integration Canis Technical Hunting Apparel - Discount Code: 'OZARK' for 15% off Umarex Airguns - Discount Code: 'OZARKAIR' for 12% off Pack Rat Outdoor Center - Gear for the Adventurous Outdoorsman We've got something brewing that has NEVER been discussed in a podcast format...and we're taking time to build it out the best we can. SO, today is a re-release of one of our favorite episodes with Steve Dally about Hopper Fishing for BIG Brown Trout on the White River. Enjoy! SUBSCRIBER ONLY PODCAST --> The Check Station The Ozark podcast sits down with men and women from the Ozarks who have a passion for the outdoors. Our aim is to listen, learn, and pass along their knowledge and experiences to help you become a better outdoorsman. Our hosts are Kyle Veit (@kyleveit_) and Kyle Plunkett (@kyle_plunkett) AND our producer is Daniel Matthews Theme music: 'American Millionaire' by JD Clayton Follow us on Instagram: @theozarkpodcast PLEASE reach out to us with any recommendations or inquiries: theozarkpodcast@gmail.com
The explosion of generative AI-powered technologies has forever changed the tech landscape. But the path to the current AI revolution was paved by many breakthroughs. Bill Dally had a hand in quite a few of them. In this week's episode of Generative Now, Lightspeed Partner and host Michael Mignano speaks with Bill Dally, NVIDIA's Chief Scientist and Senior Vice President for Research, whose nearly four-decade long career spans from experiments with neural networks in the 1980s to leading NVIDIA's groundbreaking research. He discusses the work at NVIDIA in advancing GPU technology, as well as the shift towards generative AI. Michael and Bill also explore the challenges of keeping up with rapid advancements in AI, the future implications for the industry, and advice for aspiring engineers. Bill Dally joined NVIDIA in January 2009 as chief scientist, after spending 12 years at Stanford University, where he was chairman of the computer science department. He has published over 250 papers, holds over 120 issued patents, and is an author of four textbooks. Bill received a bachelor's degree in Electrical Engineering from Virginia Tech, a master's in Electrical Engineering from Stanford University and a Ph.D. in Computer Science from CalTech. Episode Chapters(00:00) Introduction(00:22) Bill Dally's Journey from Neural Networks to NVIDIA(01:39) The Evolution of AI and Computing: A Personal Account(05:48) The AI Revolution: Expectations vs. Reality(07:58) Inside NVIDIA: The Role of Chief Scientist and the Power of Research(10:29) Exploring the Frontiers of Generative AI and Research(14:23) AI's Role in the Future of Autonomous Vehicles(20:53) The Impact of AI on Chip Design and Efficiency(25:55) Building NVIDIA's Elite Research Team(36:00) Anticipating the Future: Advice for the Next Generation(39:58) Closing Thoughts Stay in touch:www.lsvp.comX: https://twitter.com/lightspeedvpLinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/Instagram: https://www.instagram.com/lightspeedventurepartners/Subscribe on your favorite podcast app: generativenow.coEmail: generativenow@lsvp.com The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. For more details please see lsvp.com/legal.
A man who's spent more than three decades in prison for the rape and murder of 13-year-old girl Karla Cardno has been told he has a lot of work to do to prepare for life outside prison, Jimmy Ellingham was at the hearing.
This weeks episode, I have my good friend from Oaks, Dallas Morison, also known as Marlboro Slim and 420 Dally. Currently wrestler and producer, seat back and enjoy a great conversation!
Greggy Jay gets obsessed with made up abbreviations with Rizzy, Mezza and Cee! Plus, Jamie Laing plays What's My Age Again and listener Sarah is set a Springwatch challenge.
Küat Racks - High Performing Vehicle Racks Canis Technical Hunting Apparel - Discount Code: 'OZARK' for 15% off Umarex Airguns - Discount Code: 'OZARKAIR' for 12% off Cache River Leather - Discount Code: 'OZARK10' for 10% off We met up with White River super guide and friend, Steve Dally to discuss the highly anticipated periodical Cicada Hatch that last appeared 13 years ago in 2011. We discuss Steve's research on where he hypothesizes the bugs will be, what features to look for, and what type of setup he'll be using over the next 6 weeks. Enjoy! Episode mentioned: Ep. 38 - Hopper Fishing SUBSCRIBER ONLY PODCAST --> The Check Station The Ozark podcast sits down with men and women from the Ozarks who have a passion for the outdoors. Our aim is to listen, learn, and pass along their knowledge and experiences to help you become a better outdoorsman. Our hosts are Kyle Veit (@kyleveit_) and Kyle Plunkett (@kyle_plunkett) AND our producer is Daniel Matthews Theme music: 'American Millionaire' by JD Clayton Follow us on Instagram: @theozarkpodcast PLEASE reach out to us with any recommendations or inquiries: theozarkpodcast@gmail.com Learn more about your ad choices. Visit megaphone.fm/adchoices
Message: Elijah Dally, Teaching PastorPassage: Mark 16:1-8Series: The One
Women Veterans ROCK On The Hill - The Podcast! - Its New People, New Policies & Prose with Dynamic Military Women On The Move! In today's episode of "The Power of Posterity" you will meet Cadet Navil Perez, Lt. Colonel and Cadet Dally Perez, Chief Master Sgt. - two young and amazing JROTC Cadet Delegates who describe the life changing experiences in The Women Veterans STEM Leadership Program for Cadet Girls. We present highlights of our Nationwide National Enduring Partners. Contact Us at www.WomenVetsRock.org to learn more about The Work & The Vision of The 2030 Commission For Women Veterans & Military Families. Subscribe Today and Stay Connected As A Member of "The Women Veterans ROCK Podcast Posse!" ABOUT THE HOST Deborah Harmon-Pugh is a recognized authority on Women's Leadership in America. She has dedicated the past two decades to assisting women advance into positions of influence by leveraging their expertise and leadership strengths. She is the creator of proven and powerful Leadership Development Programs that guide women to becoming leaders in Civic Leadership, Business Leadership, and Nonprofit Leadership. Professor Deborah Harmon-Pugh is the National Campaign Chair of Women Veterans ROCK; The Women Veterans Civic Leadership Institute; and The Women Veterans Public Policy Delegation To Capitol Hill. She teaches in the Graduate School of Studies at Chestnut Hill College. Professor Deborah Harmon-Pugh is a retired Military Spouse of 27 years. ABOUT OUR SPONSOR Comcast NBCUniversal - We thank Comcast NBCUniversal for their support of Women Veterans, Military Families, and America's entire Military Community. For more information on how Comcast NBCUniversal is supporting the military community, visit the link below. www.corporate.comcast.com/values/military VISIT US & SUBSCRIBE TODAY Our Website Is: WomenVetsRock.org FOLLOW US ON SOCIAL MEDIA Facebook: @WomenVeteransRock Twitter: @WomenVetsRock LinkedIn: @WomenVeteransRock Instagram: @WomenVetsRock YouTube: @WomenVetsRock
Can Drip, Dally, Mr. Four, and their Riftlings deal with a flock of ferocious Skreegulls, save the mistwhales, Ugloo and Belloo, and the barge's passengers? And just what IS Madrigal hiding in those crates?Check out the Patreon and support us at patreon.com/storiesrpgPlay along at home! Download the Gigacity Guardians game sheet here: storiesrpg.gumroad.com/l/gigacity1StoriesRPG is the play-at-home, listen-and-learn game that helps people of all ages tell amazing tales together. Each season, we explore a new world, create characters, and go on epic adventures - and provide play-to-learn, coloring book, choose-your-own-adventure games to help listeners of all ages fall in love with writing!Created by Michael Low of LuckofLegends.com and Daniel Hinds of StoriesPodcast.com, StoriesRPG is a thrilling way to start bringing a love of writing, fantasy, fiction, and literacy into your home.
Katie from Alcohol-Free in Dally is here to give us the must-have tips and tricks for booze-free vacations. From weekend getaways to wild bachelorette parties, Katie gives you tips to prepare yourself for the best booze-free travel. In this episode, you will learn how to have fun without FOMO, what to drink for every occasion, and the #1 mistake to avoid when traveling sober.
„Denk nicht an morgen und arbeite mit dem, was du heute hast.“ Nach diesem Motto improvisieren die Klangkünstlerinnen Rhéa Dally und Freya Edmondes mit Gegenständen aus dem wirklichen Leben. Von USER SYNDROME (Rhéa Dally und Freya Edmondes) www.deutschlandfunkkultur.de, Klangkunst
This week Travis Knights and special guest Allie Stewart are joined by the co-founder of The Jazz Tap Ensemble, Lynn Dally. The gems that Lynn drops in this episode shine so bright. Visit https://jazztapensemblelegacy.org/ to purchase archives of her decades of work. Join our Patreon here: https://www.patreon.com/travisknights
Today we introduce a new series: Riftlings! In a world recovering from an ancient apocalypse, can humans learn to coexist with the magical creatures known as Riftlings? Our crew - Mr. Four and his Floatapus Mr. Eight, Dilly and Dally, a goblin and her Crocadillo sister, and Drip the gnome with his hat full of Zindarian tailor bees - will find out! With Scriv the Bard, Dan from Almost Bedtime Theatre, and Classical Glaiza, step into a new world with us.Check out the Patreon and support us at patreon.com/storiesrpgPlay along at home! Download the Gigacity Guardians game sheet here: storiesrpg.gumroad.com/l/gigacity1StoriesRPG is the play-at-home, listen-and-learn game that helps people of all ages tell amazing tales together. Each season, we explore a new world, create characters, and go on epic adventures - and provide play-to-learn, coloring book, choose-your-own-adventure games to help listeners of all ages fall in love with writing!Created by Michael Low of LuckofLegends.com and Daniel Hinds of StoriesPodcast.com, StoriesRPG is a thrilling way to start bringing a love of writing, fantasy, fiction, and literacy into your home.
We are running an end of year survey for our listeners! Please let us know any feedback you have, what episodes resonated with you, and guest requests for 2024! Survey link here!Before language models became all the rage in November 2022, image generation was the hottest space in AI (it was the subject of our first piece on Latent Space!) In our interview with Sharif Shameem from Lexica we talked through the launch of StableDiffusion and the early days of that space. At the time, the toolkit was still pretty rudimentary: Lexica made it easy to search images, you had the AUTOMATIC1111 Web UI to generate locally, some HuggingFace spaces that offered inference, and eventually DALL-E 2 through OpenAI's platform, but not much beyond basic text-to-image workflows.Today's guest, Suhail Doshi, is trying to solve this with Playground AI, an image editor reimagined with AI in mind. Some of the differences compared to traditional text-to-image workflows:* Real-time preview rendering using consistency: as you change your prompt, you can see changes in real-time before doing a final rendering of it.* Style filtering: rather than having to prompt exactly how you'd like an image to look, you can pick from a whole range of filters both from Playground's model as well as Stable Diffusion (like RealVis, Starlight XL, etc). We talk about this at 25:46 in the podcast.* Expand prompt: similar to DALL-E3, Playground will do some prompt tuning for you to get better results in generation. Unlike DALL-E3, you can turn this off at any time if you are a prompting wizard* Image editing: after generation, you have tools like a magic eraser, inpainting pencil, etc. This makes it easier to do a full workflow in Playground rather than switching to another tool like Photoshop.Outside of the product, they have also trained a new model from scratch, Playground v2, which is fully open source and open weights and allows for commercial usage. They benchmarked the model against SDXL across 1,000 prompts and found that humans preferred the Playground generation 70% of the time. They had similar results on PartiPrompts:They also created a new benchmark, MJHQ-30K, for “aesthetic quality”:We introduce a new benchmark, MJHQ-30K, for automatic evaluation of a model's aesthetic quality. The benchmark computes FID on a high-quality dataset to gauge aesthetic quality.We curate the high-quality dataset from Midjourney with 10 common categories, each category with 3K samples. Following common practice, we use aesthetic score and CLIP score to ensure high image quality and high image-text alignment. Furthermore, we take extra care to make the data diverse within each category.Suhail was pretty open with saying that Midjourney is currently the best product for imagine generation out there, and that's why they used it as the base for this benchmark. I think it's worth comparing yourself to maybe the best thing and try to find like a really fair way of doing that. So I think more people should try to do that. I definitely don't think you should be kind of comparing yourself on like some Google model or some old SD, Stable Diffusion model and be like, look, we beat Stable Diffusion 1.5. I think users ultimately want care, how close are you getting to the thing that people mostly agree with? [00:23:47]We also talked a lot about Suhail's founder journey from starting Mixpanel in 2009, then going through YC again with Mighty, and eventually sunsetting that to pivot into Playground. Enjoy!Show Notes* Suhail's Twitter* “Starting my road to learn AI”* Bill Gates book trip* Playground* Playground v2 Announcement* $40M raise announcement* “Running infra dev ops for 24 A100s”* Mixpanel* Mighty* “I decided to stop working on Mighty”* Fast.ai* CivitTimestamps* [00:00:00] Intros* [00:02:59] Being early in ML at Mixpanel* [00:04:16] Pivoting from Mighty to Playground and focusing on generative AI* [00:07:54] How DALL-E 2 inspired Mighty* [00:09:19] Reimagining the graphics editor with AI* [00:17:34] Training the Playground V2 model from scratch to advance generative graphics* [00:21:11] Techniques used to improve Playground V2 like data filtering and model tuning* [00:25:21] Releasing the MJHQ30K benchmark to evaluate generative models* [00:30:35] The limitations of current models for detailed image editing tasks* [00:34:06] Using post-generation user feedback to create better benchmarks* [00:38:28] Concerns over potential misuse of powerful generative models* [00:41:54] Rethinking the graphics editor user experience in the AI era* [00:45:44] Integrating consistency models into Playground using preview rendering* [00:47:23] Interacting with the Stable Diffusion LoRAs community* [00:51:35] Running DevOps on A100s* [00:53:12] Startup ideas?TranscriptAlessio: Hey everyone, welcome to the Latent Space podcast. This is Alessio, partner and CTO-in-Residence at Decibel Partners, and I'm joined by my co-host Swyx, founder of Smol AI. [00:00:15]Swyx: Hey, and today in the studio we have Suhail Doshi, welcome. [00:00:18]Suhail: Yeah, thanks. Thanks for having me. [00:00:20]Swyx: So among many things, you're a CEO and co-founder of Mixpanel, and I think about three years ago you left to start Mighty, and more recently, I think about a year ago, transitioned into Playground, and you've just announced your new round. How do you like to be introduced beyond that? [00:00:34]Suhail: Just founder of Playground is fine, yeah, prior co-founder and CEO of Mixpanel. [00:00:40]Swyx: Yeah, awesome. I'd just like to touch on Mixpanel a little bit, because it's obviously one of the more successful analytics companies we previously had amplitude on, and I'm curious if you had any reflections on the interaction of that amount of data that people would want to use for AI. I don't know if there's still a part of you that stays in touch with that world. [00:00:59]Suhail: Yeah, I mean, the short version is that maybe back in like 2015 or 2016, I don't really remember exactly, because it was a while ago, we had an ML team at Mixpanel, and I think this is when maybe deep learning or something really just started getting kind of exciting, and we were thinking that maybe given that we had such vast amounts of data, perhaps we could predict things. So we built two or three different features, I think we built a feature where we could predict whether users would churn from your product. We made a feature that could predict whether users would convert, we built a feature that could do anomaly detection, like if something occurred in your product, that was just very surprising, maybe a spike in traffic in a particular region, can we tell you that that happened? Because it's really hard to like know everything that's going on with your data, can we tell you something surprising about your data? And we tried all of these various features, most of it boiled down to just like, you know, using logistic regression, and it never quite seemed very groundbreaking in the end. And so I think, you know, we had a four or five person ML team, and I think we never expanded it from there. And I did all these Fast AI courses trying to learn about ML. And that was the- That's the first time you did fast AI. Yeah, that was the first time I did fast AI. Yeah, I think I've done it now three times, maybe. [00:02:12]Swyx: Oh, okay. [00:02:13]Suhail: I didn't know it was the third. No, no, just me reviewing it, it's maybe three times, but yeah. [00:02:16]Swyx: You mentioned prediction, but honestly, like it's also just about the feedback, right? The quality of feedback from users, I think it's useful for anyone building AI applications. [00:02:25]Suhail: Yeah. Yeah, I think I haven't spent a lot of time thinking about Mixpanel because it's been a long time, but sometimes I'm like, oh, I wonder what we could do now. And then I kind of like move on to whatever I'm working on, but things have changed significantly since. [00:02:39]Swyx: And then maybe we'll touch on Mighty a little bit. Mighty was very, very bold. My framing of it was, you will run our browsers for us because everyone has too many tabs open. I have too many tabs open and slowing down your machines that you can do it better for us in a centralized data center. [00:02:51]Suhail: Yeah, we were first trying to make a browser that we would stream from a data center to your computer at extremely low latency, but the real objective wasn't trying to make a browser or anything like that. The real objective was to try to make a new kind of computer. And the thought was just that like, you know, we have these computers in front of us today and we upgrade them or they run out of RAM or they don't have enough RAM or not enough disk or, you know, there's some limitation with our computers, perhaps like data locality is a problem. Why do I need to think about upgrading my computer ever? And so, you know, we just had to kind of observe that like, well, actually it seems like a lot of applications are just now in the browser, you know, it's like how many real desktop applications do we use relative to the number of applications we use in the browser? So it's just this realization that actually like, you know, the browser was effectively becoming more or less our operating system over time. And so then that's why we kind of decided to go, hmm, maybe we can stream the browser. Fortunately, the idea did not work for a couple of different reasons, but the objective is try to make sure new computer. [00:03:50]Swyx: Yeah, very, very bold. [00:03:51]Alessio: Yeah, and I was there at YC Demo Day when you first announced it. It was, I think, the last or one of the last in-person ones, at Pier34 in Mission Bay. How do you think about that now when everybody wants to put some of these models in people's machines and some of them want to stream them in, do you think there's maybe another wave of the same problem before it was like browser apps too slow, now it's like models too slow to run on device? [00:04:16]Suhail: Yeah. I mean, I've obviously pivoted away from Mighty, but a lot of what I somewhat believed at Mighty, maybe why I'm so excited about AI and what's happening, a lot of what Mighty was about was like moving compute somewhere else, right? Right now, applications, they get limited quantities of memory, disk, networking, whatever your home network has, et cetera. You know, what if these applications could somehow, if we could shift compute, and then these applications have vastly more compute than they do today. Right now it's just like client backend services, but you know, what if we could change the shape of how applications could interact with things? And it's changed my thinking. In some ways, AI has like a bit of a continuation of my belief that like perhaps we can really shift compute somewhere else. One of the problems with Mighty was that JavaScript is single-threaded in the browser. And what we learned, you know, the reason why we kind of abandoned Mighty was because I didn't believe we could make a new kind of computer. We could have made some kind of enterprise business, probably it could have made maybe a lot of money, but it wasn't going to be what I hoped it was going to be. And so once I realized that most of a web app is just going to be single-threaded JavaScript, then the only thing you could do largely withstanding changing JavaScript, which is a fool's errand most likely, make a better CPU, right? And there's like three CPU manufacturers, two of which sell, you know, big ones, you know, AMD, Intel, and then of course like Apple made the M1. And it's not like single-threaded CPU core performance, single-core performance was increasing very fast, it's plateauing rapidly. And even these different companies were not doing as good of a job, you know, sort of with the continuation of Moore's law. But what happened in AI was that you got like, if you think of the AI model as like a computer program, like just like a compiled computer program, it is literally built and designed to do massive parallel computations. And so if you could take like the universal approximation theorem to its like kind of logical complete point, you know, you're like, wow, I can get, make computation happen really rapidly and parallel somewhere else, you know, so you end up with these like really amazing models that can like do anything. It just turned out like perhaps the new kind of computer would just simply be shifted, you know, into these like really amazing AI models in reality. Yeah. [00:06:30]Swyx: Like I think Andrej Karpathy has always been, has been making a lot of analogies with the LLMOS. [00:06:34]Suhail: I saw his video and I watched that, you know, maybe two weeks ago or something like that. I was like, oh man, this, I very much resonate with this like idea. [00:06:41]Swyx: Why didn't I see this three years ago? [00:06:43]Suhail: Yeah. I think, I think there still will be, you know, local models and then there'll be these very large models that have to be run in data centers. I think it just depends on kind of like the right tool for the job, like any engineer would probably care about. But I think that, you know, by and large, like if the models continue to kind of keep getting bigger, you're always going to be wondering whether you should use the big thing or the small, you know, the tiny little model. And it might just depend on like, you know, do you need 30 FPS or 60 FPS? Maybe that would be hard to do, you know, over a network. [00:07:13]Swyx: You tackled a much harder problem latency wise than the AI models actually require. Yeah. [00:07:18]Suhail: Yeah. You can do quite well. You can do quite well. You definitely did 30 FPS video streaming, did very crazy things to make that work. So I'm actually quite bullish on the kinds of things you can do with networking. [00:07:30]Swyx: Maybe someday you'll come back to that at some point. But so for those that don't know, you're very transparent on Twitter. Very good to follow you just to learn your insights. And you actually published a postmortem on Mighty that people can read up on and willing to. So there was a bit of an overlap. You started exploring the AI stuff in June 2022, which is when you started saying like, I'm taking fast AI again. Maybe, was there more context around that? [00:07:54]Suhail: Yeah. I think I was kind of like waiting for the team at Mighty to finish up, you know, something. And I was like, okay, well, what can I do? I guess I will make some kind of like address bar predictor in the browser. So we had, you know, we had forked Chrome and Chromium. And I was like, you know, one thing that's kind of lame is that like this browser should be like a lot better at predicting what I might do, where I might want to go. It struck me as really odd that, you know, Chrome had very little AI actually or ML inside this browser. For a company like Google, you'd think there's a lot. Code is actually just very, you know, it's just a bunch of if then statements is more or less the address bar. So it seemed like a pretty big opportunity. And that's also where a lot of people interact with the browser. So, you know, long story short, I was like, hmm, I wonder what I could build here. So I started to take some AI courses and review the material again and get back to figuring it out. But I think that was somewhat serendipitous because right around April was, I think, a very big watershed moment in AI because that's when Dolly 2 came out. And I think that was the first truly big viral moment for generative AI. [00:08:59]Swyx: Because of the avocado chair. [00:09:01]Suhail: Yeah, exactly. [00:09:02]Swyx: It wasn't as big for me as Stable Diffusion. [00:09:04]Suhail: Really? [00:09:05]Swyx: Yeah, I don't know. Dolly was like, all right, that's cool. [00:09:07]Suhail: I don't know. Yeah. [00:09:09]Swyx: I mean, they had some flashy videos, but it didn't really register. [00:09:13]Suhail: That moment of images was just such a viral novel moment. I think it just blew people's mind. Yeah. [00:09:19]Swyx: I mean, it's the first time I encountered Sam Altman because they had this Dolly 2 hackathon and they opened up the OpenAI office for developers to walk in back when it wasn't as much of a security issue as it is today. I see. Maybe take us through the journey to decide to pivot into this and also choosing images. Obviously, you were inspired by Dolly, but there could be any number of AI companies and businesses that you could start and why this one, right? [00:09:45]Suhail: Yeah. So I think at that time, Mighty and OpenAI was not quite as popular as it is all of a sudden now these days, but back then they had a lot more bandwidth to kind of help anybody. And so we had been talking with the team there around trying to see if we could do really fast low latency address bar prediction with GPT-3 and 3.5 and that kind of thing. And so we were sort of figuring out how could we make that low latency. I think that just being able to talk to them and kind of being involved gave me a bird's eye view into a bunch of things that started to happen. Latency first was the Dolly 2 moment, but then stable diffusion came out and that was a big moment for me as well. And I remember just kind of like sitting up one night thinking, I was like, you know, what are the kinds of companies one could build? Like what matters right now? One thing that I observed is that I find a lot of inspiration when I'm working in a field in something and then I can identify a bunch of problems. Like for Mixpanel, I was an intern at a company and I just noticed that they were doing all this data analysis. And so I thought, hmm, I wonder if I could make a product and then maybe they would use it. And in this case, you know, the same thing kind of occurred. It was like, okay, there are a bunch of like infrastructure companies that put a model up and then you can use their API, like Replicate is a really good example of that. There are a bunch of companies that are like helping you with training, model optimization, Mosaic at the time, and probably still, you know, was doing stuff like that. So I just started listing out like every category of everything, of every company that was doing something interesting. I started listing out like weights and biases. I was like, oh man, weights and biases is like this great company. Do I want to compete with that company? I might be really good at competing with that company because of Mixpanel because it's so much of like analysis. But I was like, no, I don't want to do anything related to that. That would, I think that would be too boring now at this point. So I started to list out all these ideas and one thing I observed was that at OpenAI, they had like a playground for GPT-3, right? All it was is just like a text box more or less. And then there were some settings on the right, like temperature and whatever. [00:11:41]Swyx: Top K. [00:11:42]Suhail: Yeah, top K. You know, what's your end stop sequence? I mean, that was like their product before GPT, you know, really difficult to use, but fun if you're like an engineer. And I just noticed that their product kind of was evolving a little bit where the interface kind of was getting a little bit more complex. They had like a way where you could like generate something in the middle of a sentence and all those kinds of things. And I just thought to myself, I was like, everything is just like this text box and you generate something and that's about it. And stable diffusion had kind of come out and it was all like hugging face and code. Nobody was really building any UI. And so I had this kind of thing where I wrote prompt dash like question mark in my notes and I didn't know what was like the product for that at the time. I mean, it seems kind of trite now, but I just like wrote prompt. What's the thing for that? Manager. Prompt manager. Do you organize them? Like, do you like have a UI that can play with them? Yeah. Like a library. What would you make? And so then, of course, then you thought about what would the modalities be given that? How would you build a UI for each kind of modality? And so there are a couple of people working on some pretty cool things. And I basically chose graphics because it seemed like the most obvious place where you could build a really powerful, complex UI. That's not just only typing a box. It would very much evolve beyond that. Like what would be the best thing for something that's visual? Probably something visual. Yeah. I think that just that progression kind of happened and it just seemed like there was a lot of effort going into language, but not a lot of effort going into graphics. And then maybe the very last thing was, I think I was talking to Aditya Ramesh, who was the co-creator of DALL-E 2 and Sam. And I just kind of went to these guys and I was just like, hey, are you going to make like a UI for this thing? Like a true UI? Are you going to go for this? Are you going to make a product? For DALL-E. Yeah. For DALL-E. Yeah. Are you going to do anything here? Because if you are going to do it, just let me know and I will stop and I'll go do something else. But if you're not going to do anything, I'll just do it. And so we had a couple of conversations around what that would look like. And then I think ultimately they decided that they were going to focus on language primarily. And I just felt like it was going to be very underinvested in. Yes. [00:13:46]Swyx: There's that sort of underinvestment from OpenAI, but also it's a different type of customer than you're used to, presumably, you know, and Mixpanel is very good at selling to B2B and developers will figure on you or not. Yeah. Was that not a concern? [00:14:00]Suhail: Well, not so much because I think that, you know, right now I would say graphics is in this very nascent phase. Like most of the customers are just like hobbyists, right? Yeah. Like it's a little bit of like a novel toy as opposed to being this like very high utility thing. But I think ultimately, if you believe that you could make it very high utility, the probably the next customers will end up being B2B. It'll probably not be like a consumer. There will certainly be a variation of this idea that's in consumer. But if your quest is to kind of make like something that surpasses human ability for graphics, like ultimately it will end up being used for business. So I think it's maybe more of a progression. In fact, for me, it's maybe more like Mixpanel started out as SMB and then very much like ended up starting to grow up towards enterprise. So for me, I think it will be a very similar progression. But yeah, I mean, the reason why I was excited about it is because it was a creative tool. I make music and it's AI. It's like something that I know I could stay up till three o'clock in the morning doing. Those are kind of like very simple bars for me. [00:14:56]Alessio: So you mentioned Dolly, Stable Diffusion. You just had Playground V2 come out two days ago. Yeah, two days ago. [00:15:02]Suhail: Two days ago. [00:15:03]Alessio: This is a model you train completely from scratch. So it's not a cheap fine tune on something. You open source everything, including the weights. Why did you decide to do it? I know you supported Stable Diffusion XL in Playground before, right? Yep. What made you want to come up with V2 and maybe some of the interesting, you know, technical research work you've done? [00:15:24]Suhail: Yeah. So I think that we continue to feel like graphics and these foundation models for anything really related to pixels, but also definitely images continues to be very underinvested. It feels a little like graphics is in like this GPT-2 moment, right? Like even GPT-3, even when GPT-3 came out, it was exciting, but it was like, what are you going to use this for? Yeah, we'll do some text classification and some semantic analysis and maybe it'll sometimes like make a summary of something and it'll hallucinate. But no one really had like a very significant like business application for GPT-3. And in images, we're kind of stuck in the same place. We're kind of like, okay, I write this thing in a box and I get some cool piece of artwork and the hands are kind of messed up and sometimes the eyes are a little weird. Maybe I'll use it for a blog post, you know, that kind of thing. The utility feels so limited. And so, you know, and then we, you sort of look at Stable Diffusion and we definitely use that model in our product and our users like it and use it and love it and enjoy it, but it hasn't gone nearly far enough. So we were kind of faced with the choice of, you know, do we wait for progress to occur or do we make that progress happen? So yeah, we kind of embarked on a plan to just decide to go train these things from scratch. And I think the community has given us so much. The community for Stable Diffusion I think is one of the most vibrant communities on the internet. It's like amazing. It feels like, I hope this is what like Homebrew Club felt like when computers like showed up because it's like amazing what that community will do and it moves so fast. I've never seen anything in my life and heard other people's stories around this where an academic research paper comes out and then like two days later, someone has sample code for it. And then two days later, there's a model. And then two days later, it's like in nine products, you know, they're all competing with each other. It's incredible to see like math symbols on an academic paper go to well-designed features in a product. So I think the community has done so much. So I think we wanted to give back to the community kind of on our way. Certainly we would train a better model than what we gave out on Tuesday, but we definitely felt like there needs to be some kind of progress in these open source models. The last kind of milestone was in July when Stable Diffusion Excel came out, but there hasn't been anything really since. Right. [00:17:34]Swyx: And there's Excel Turbo now. [00:17:35]Suhail: Well, Excel Turbo is like this distilled model, right? So it's like lower quality, but fast. You have to decide, you know, what your trade off is there. [00:17:42]Swyx: It's also a consistency model. [00:17:43]Suhail: I don't think it's a consistency model. It's like it's they did like a different thing. Yeah. I think it's like, I don't want to get quoted for this, but it's like something called ad like adversarial or something. [00:17:52]Swyx: That's exactly right. [00:17:53]Suhail: I've read something about that. Maybe it's like closer to GANs or something, but I didn't really read the full paper. But yeah, there hasn't been quite enough progress in terms of, you know, there's no multitask image model. You know, the closest thing would be something called like EmuEdit, but there's no model for that. It's just a paper that's within meta. So we did that and we also gave out pre-trained weights, which is very rare. Usually you just get the aligned model and then you have to like see if you can do anything with it. So we actually gave out, there's like a 256 pixel pre-trained stage and a 512. And we did that for academic research because we come across people all the time in academia, they have access to like one A100 or eight at best. And so if we can give them kind of like a 512 pre-trained model, our hope is that there'll be interesting novel research that occurs from that. [00:18:38]Swyx: What research do you want to happen? [00:18:39]Suhail: I would love to see more research around things that users care about tend to be things like character consistency. [00:18:45]Swyx: Between frames? [00:18:46]Suhail: More like if you have like a face. Yeah, yeah. Basically between frames, but more just like, you know, you have your face and it's in one image and then you want it to be like in another. And users are very particular and sensitive to faces changing because we know we're trained on faces as humans. Not seeing a lot of innovation, enough innovation around multitask editing. You know, there are two things like instruct pics to pics and then the EmuEdit paper that are maybe very interesting, but we certainly are not pushing the fold on that in that regard. All kinds of things like around that rotation, you know, being able to keep coherence across images, style transfer is still very limited. Just even reasoning around images, you know, what's going on in an image, that kind of thing. Things are still very, very underpowered, very nascent. So therefore the utility is very, very limited. [00:19:32]Alessio: On the 1K Prompt Benchmark, you are 2.5x prefer to Stable Diffusion XL. How do you get there? Is it better images in the training corpus? Can you maybe talk through the improvements in the model? [00:19:44]Suhail: I think they're still very early on in the recipe, but I think it's a lot of like little things and you know, every now and then there are some big important things like certainly your data quality is really, really important. So we spend a lot of time thinking about that. But I would say it's a lot of things that you kind of clean up along the way as you train your model. Everything from captions to the data that you align with after pre-train to how you're picking your data sets, how you filter your data sets. I feel like there's a lot of work in AI that doesn't really feel like AI. It just really feels like just data set filtering and systems engineering and just like, you know, and the recipe is all there, but it's like a lot of extra work to do that. I think we plan to do a Playground V 2.1, maybe either by the end of the year or early next year. And we're just like watching what the community does with the model. And then we're just going to take a lot of the things that they're unhappy about and just like fix them. You know, so for example, like maybe the eyes of people in an image don't feel right. They feel like they're a little misshapen or they're kind of blurry feeling. That's something that we already know we want to fix. So I think in that case, it's going to be about data quality. Or maybe you want to improve the kind of the dynamic range of color. You know, we want to make sure that that's like got a good range in any image. So what technique can we use there? There's different things like offset noise, pyramid noise, terminal zero, SNR, like there are all these various interesting things that you can do. So I think it's like a lot of just like tricks. Some are tricks, some are data, and some is just like cleaning. [00:21:11]Swyx: Specifically for faces, it's very common to use a pipeline rather than just train the base model more. Do you have a strong belief either way on like, oh, they should be separated out to different stages for like improving the eyes, improving the face or enhance or whatever? Or do you think like it can all be done in one model? [00:21:28]Suhail: I think we will make a unified model. Yeah, I think it will. I think we'll certainly in the end, ultimately make a unified model. There's not enough research about this. Maybe there is something out there that we haven't read. There are some bottlenecks, like for example, in the VAE, like the VAEs are ultimately like compressing these things. And so you don't know. And then you might have like a big informational information bottleneck. So maybe you would use a pixel based model, perhaps. I think we've talked to people, everyone from like Rombach to various people, Rombach trained stable diffusion. I think there's like a big question around the architecture of these things. It's still kind of unknown, right? Like we've got transformers and we've got like a GPT architecture model, but then there's this like weird thing that's also seemingly working with diffusion. And so, you know, are we going to use vision transformers? Are we going to move to pixel based models? Is there a different kind of architecture? We don't really, I don't think there have been enough experiments. Still? Oh my God. [00:22:21]Swyx: Yeah. [00:22:22]Suhail: That's surprising. I think it's very computationally expensive to do a pipeline model where you're like fixing the eyes and you're fixing the mouth and you're fixing the hands. [00:22:29]Swyx: That's what everyone does as far as I understand. [00:22:31]Suhail: I'm not exactly sure what you mean, but if you mean like you get an image and then you will like make another model specifically to fix a face, that's fairly computationally expensive. And I think it's like not probably not the right way. Yeah. And it doesn't generalize very well. Now you have to pick all these different things. [00:22:45]Swyx: Yeah. You're just kind of glomming things on together. Yeah. Like when I look at AI artists, like that's what they do. [00:22:50]Suhail: Ah, yeah, yeah, yeah. They'll do things like, you know, I think a lot of ARs will do control net tiling to do kind of generative upscaling of all these different pieces of the image. Yeah. And I think these are all just like, they're all hacks ultimately in the end. I mean, it just to me, it's like, let's go back to where we were just three years, four years ago with where deep learning was at and where language was that, you know, it's the same thing. It's like we were like, okay, well, I'll just train these very narrow models to try to do these things and kind of ensemble them or pipeline them to try to get to a best in class result. And here we are with like where the models are gigantic and like very capable of solving huge amounts of tasks when given like lots of great data. [00:23:28]Alessio: You also released a new benchmark called MJHQ30K for automatic evaluation of a model's aesthetic quality. I have one question. The data set that you use for the benchmark is from Midjourney. Yes. You have 10 categories. How do you think about the Playground model, Midjourney, like, are you competitors? [00:23:47]Suhail: There are a lot of people, a lot of people in research, they like to compare themselves to something they know they can beat, right? Maybe this is the best reason why it can be helpful to not be a researcher also sometimes like I'm not trained as a researcher, I don't have a PhD in anything AI related, for example. But I think if you care about products and you care about your users, then the most important thing that you want to figure out is like everyone has to acknowledge that Midjourney is very good. They are the best at this thing. I'm happy to admit that. I have no problem admitting that. Just easy. It's very visual to tell. So I think it's incumbent on us to try to compare ourselves to the thing that's best, even if we lose, even if we're not the best. At some point, if we are able to surpass Midjourney, then we only have ourselves to compare ourselves to. But on First Blush, I think it's worth comparing yourself to maybe the best thing and try to find like a really fair way of doing that. So I think more people should try to do that. I definitely don't think you should be kind of comparing yourself on like some Google model or some old SD, Stable Diffusion model and be like, look, we beat Stable Diffusion 1.5. I think users ultimately want care, how close are you getting to the thing that people mostly agree with? So we put out that benchmark for no other reason to say like, this seems like a worthy thing for us to at least try, for people to try to get to. And then if we surpass it, great, we'll come up with another one. [00:25:06]Alessio: Yeah, no, that's awesome. And you killed Stable Diffusion Excel and everything. In the benchmark chart, it says Playground V2 1024 pixel dash aesthetic. Do you have kind of like, yeah, style fine tunes or like what's the dash aesthetic for? [00:25:21]Suhail: We debated this, maybe we named it wrong or something, but we were like, how do we help people realize the model that's aligned versus the models that weren't? Because we gave out pre-trained models, we didn't want people to like use those. So that's why they're called base. And then the aesthetic model, yeah, we wanted people to pick up the thing that makes things pretty. Who wouldn't want the thing that's aesthetic? But if there's a better name, we're definitely open to feedback. No, no, that's cool. [00:25:46]Alessio: I was using the product. You also have the style filter and you have all these different styles. And it seems like the styles are tied to the model. So there's some like SDXL styles, there's some Playground V2 styles. Can you maybe give listeners an overview of how that works? Because in language, there's not this idea of like style, right? Versus like in vision model, there is, and you cannot get certain styles in different [00:26:11]Suhail: models. [00:26:12]Alessio: So how do styles emerge and how do you categorize them and find them? [00:26:15]Suhail: Yeah, I mean, it's so fun having a community where people are just trying a model. Like it's only been two days for Playground V2. And we actually don't know what the model's capable of and not capable of. You know, we certainly see problems with it. But we have yet to see what emergent behavior is. I mean, we've just sort of discovered that it takes about like a week before you start to see like new things. I think like a lot of that style kind of emerges after that week, where you start to see, you know, there's some styles that are very like well known to us, like maybe like pixel art is a well known style. Photorealism is like another one that's like well known to us. But there are some styles that cannot be easily named. You know, it's not as simple as like, okay, that's an anime style. It's very visual. And in the end, you end up making up the name for what that style represents. And so the community kind of shapes itself around these different things. And so if anyone that's into stable diffusion and into building anything with graphics and stuff with these models, you know, you might have heard of like Proto Vision or Dream Shaper, some of these weird names, but they're just invented by these authors. But they have a sort of je ne sais quoi that, you know, appeals to users. [00:27:26]Swyx: Because it like roughly embeds to what you what you want. [00:27:29]Suhail: I guess so. I mean, it's like, you know, there's one of my favorite ones that's fine tuned. It's not made by us. It's called like Starlight XL. It's just this beautiful model. It's got really great color contrast and visual elements. And the users love it. I love it. And it's so hard. I think that's like a very big open question with graphics that I'm not totally sure how we'll solve. I don't know. It's, it's like an evolving situation too, because styles get boring, right? They get fatigued. Like it's like listening to the same style of pop song. I try to relate to graphics a little bit like with music, because I think it gives you a little bit of a different shape to things. Like it's not as if we just have pop music, rap music and country music, like all of these, like the EDM genre alone has like sub genres. And I think that's very true in graphics and painting and art and anything that we're doing. There's just these sub genres, even if we can't quite always name them. But I think they are emergent from the community, which is why we're so always happy to work with the community. [00:28:26]Swyx: That is a struggle. You know, coming back to this, like B2B versus B2C thing, B2C, you're going to have a huge amount of diversity and then it's going to reduce as you get towards more sort of B2B type use cases. I'm making this up here. So like you might be optimizing for a thing that you may eventually not need. [00:28:42]Suhail: Yeah, possibly. Yeah, possibly. I think like a simple thing with startups is that I worry sometimes by trying to be overly ambitious and like really scrutinizing like what something is in its most nascent phase that you miss the most ambitious thing you could have done. Like just having like very basic curiosity with something very small can like kind of lead you to something amazing. Like Einstein definitely did that. And then he like, you know, he basically won all the prizes and got everything he wanted and then basically did like kind of didn't really. He can dismiss quantum and then just kind of was still searching, you know, for the unifying theory. And he like had this quest. I think that happens a lot with like Nobel Prize people. I think there's like a term for it that I forget. I actually wanted to go after a toy almost intentionally so long as that I could see, I could imagine that it would lead to something very, very large later. Like I said, it's very hobbyist, but you need to start somewhere. You need to start with something that has a big gravitational pull, even if these hobbyists aren't likely to be the people that, you know, have a way to monetize it or whatever, even if they're, but they're doing it for fun. So there's something, something there that I think is really important. But I agree with you that, you know, in time we will absolutely focus on more utilitarian things like things that are more related to editing feats that are much harder. And so I think like a very simple use case is just, you know, I'm not a graphics designer. It seems like very simple that like you, if we could give you the ability to do really complex graphics without skill, wouldn't you want that? You know, like my wife the other day was set, you know, said, I wish Playground was better. When are you guys going to have a feature where like we could make my son, his name's Devin, smile when he was not smiling in the picture for the holiday card. Right. You know, just being able to highlight his, his mouth and just say like, make him smile. Like why can't we do that with like high fidelity and coherence, little things like that, all the way to putting you in completely different scenarios. [00:30:35]Swyx: Is that true? Can we not do that in painting? [00:30:37]Suhail: You can do in painting, but the quality is just so bad. Yeah. It's just really terrible quality. You know, it's like you'll do it five times and it'll still like kind of look like crooked or just artifact. Part of it's like, you know, the lips on the face, there's such little information there. So small that the models really struggle with it. Yeah. [00:30:55]Swyx: Make the picture smaller and you don't see it. That's my trick. I don't know. [00:30:59]Suhail: Yeah. Yeah. That's true. Or, you know, you could take that region and make it really big and then like say it's a mouth and then like shrink it. It feels like you're wrestling with it more than it's doing something that kind of surprises you. [00:31:12]Swyx: Yeah. It feels like you are very much the internal tastemaker, like you carry in your head this vision for what a good art model should look like. Do you find it hard to like communicate it to like your team and other people? Just because it's obviously it's hard to put into words like we just said. [00:31:26]Suhail: Yeah. It's very hard to explain. Images have such high bitrate compared to just words and we don't have enough words to describe these things. It's not terribly difficult. I think everyone on the team, if they don't have good kind of like judgment taste or like an eye for some of these things, they're like steadily building it because they have no choice. Right. So in that realm, I don't worry too much, actually. Like everyone is kind of like learning to get the eye is what I would call it. But I also have, you know, my own narrow taste. Like I don't represent the whole population either. [00:31:59]Swyx: When you benchmark models, you know, like this benchmark we're talking about, we use FID. Yeah. Input distance. OK. That's one measure. But like it doesn't capture anything you just said about smiles. [00:32:08]Suhail: Yeah. FID is generally a bad metric. It's good up to a point and then it kind of like is irrelevant. Yeah. [00:32:14]Swyx: And then so are there any other metrics that you like apart from vibes? I'm always looking for alternatives to vibes because vibes don't scale, you know. [00:32:22]Suhail: You know, it might be fun to kind of talk about this because it's actually kind of fresh. So up till now, we haven't needed to do a ton of like benchmarking because we hadn't trained our own model and now we have. So now what? What does that mean? How do we evaluate it? And, you know, we're kind of like living with the last 48, 72 hours of going, did the way that we benchmark actually succeed? [00:32:43]Swyx: Did it deliver? [00:32:44]Suhail: Right. You know, like I think Gemini just came out. They just put out a bunch of benchmarks. But all these benchmarks are just an approximation of how you think it's going to end up with real world performance. And I think that's like very fascinating to me. So if you fake that benchmark, you'll still end up in a really bad scenario at the end of the day. And so, you know, one of the benchmarks we did was we kind of curated like a thousand prompts. And I think that's kind of what we published in our blog post, you know, of all these tasks that we a lot of some of them are curated by our team where we know the models all suck at it. Like my favorite prompt that no model is really capable of is a horse riding an astronaut, the inverse one. And it's really, really hard to do. [00:33:22]Swyx: Not in data. [00:33:23]Suhail: You know, another one is like a giraffe underneath a microwave. How does that work? Right. There's so many of these little funny ones. We do. We have prompts that are just like misspellings of things. Yeah. We'll figure out if the models will figure it out. [00:33:36]Swyx: They should embed to the same space. [00:33:39]Suhail: Yeah. And just like all these very interesting weirdo things. And so we have so many of these and then we kind of like evaluate whether the models are any good at it. And the reality is that they're all bad at it. And so then you're just picking the most aesthetic image. We're still at the beginning of building like the best benchmark we can that aligns most with just user happiness, I think, because we're not we're not like putting these in papers and trying to like win, you know, I don't know, awards at ICCV or something if they have awards. You could. [00:34:05]Swyx: That's absolutely a valid strategy. [00:34:06]Suhail: Yeah, you could. But I don't think it could correlate necessarily with the impact we want to have on humanity. I think we're still evolving whatever our benchmarks are. So the first benchmark was just like very difficult tasks that we know the models are bad at. Can we come up with a thousand of these, whether they're hand rated and some of them are generated? And then can we ask the users, like, how do we do? And then we wanted to use a benchmark like party prompts. We mostly did that so people in academia could measure their models against ours versus others. But yeah, I mean, fit is pretty bad. And I think in terms of vibes, it's like you put out the model and then you try to see like what users make. And I think my sense is that we're going to take all the things that we notice that the users kind of were failing at and try to find like new ways to measure that, whether that's like a smile or, you know, color contrast or lighting. One benefit of Playground is that we have users making millions of images every single day. And so we can just ask them for like a post generation feedback. Yeah, we can just ask them. We can just say, like, how good was the lighting here? How was the subject? How was the background? [00:35:06]Swyx: Like a proper form of like, it's just like you make it, you come to our site, you make [00:35:10]Suhail: an image and then we say, and then maybe randomly you just say, hey, you know, like, how was the color and contrast of this image? And you say it was not very good, just tell us. So I think I think we can get like tens of thousands of these evaluations every single day to truly measure real world performance as opposed to just like benchmark performance. I would like to publish hopefully next year. I think we will try to publish a benchmark that anyone could use, that we evaluate ourselves on and that other people can, that we think does a good job of approximating real world performance because we've tried it and done it and noticed that it did. Yeah. I think we will do that. [00:35:45]Swyx: I personally have a few like categories that I consider special. You know, you know, you have like animals, art, fashion, food. There are some categories which I consider like a different tier of image. Top among them is text in images. How do you think about that? So one of the big wow moments for me, something I've been looking out for the entire year is just the progress of text and images. Like, can you write in an image? Yeah. And Ideogram came out recently, which had decent but not perfect text and images. Dolly3 had improved some and all they said in their paper was that they just included more text in the data set and it just worked. I was like, that's just lazy. But anyway, do you care about that? Because I don't see any of that in like your sample. Yeah, yeah. [00:36:27]Suhail: The V2 model was mostly focused on image quality versus like the feature of text synthesis. [00:36:33]Swyx: Well, as a business user, I care a lot about that. [00:36:35]Suhail: Yeah. Yeah. I'm very excited about text synthesis. And yeah, I think Ideogram has done a good job of maybe the best job. Dolly has like a hit rate. Yes. You know, like sometimes it's Egyptian letters. Yeah. I'm very excited about text synthesis. You know, I don't have much to say on it just yet. You know, you don't want just text effects. I think where this has to go is it has to be like you could like write little tiny pieces of text like on like a milk carton. That's maybe not even the focal point of a scene. I think that's like a very hard task that, you know, if you could do something like that, then there's a lot of other possibilities. Well, you don't have to zero shot it. [00:37:09]Swyx: You can just be like here and focus on this. [00:37:12]Suhail: Sure. Yeah, yeah. Definitely. Yeah. [00:37:16]Swyx: Yeah. So I think text synthesis would be very exciting. I'll also flag that Max Wolf, MiniMaxxier, which you must have come across his work. He's done a lot of stuff about using like logo masks that then map onto food and vegetables. And it looks like text, which can be pretty fun. [00:37:29]Suhail: That's the wonderful thing about like the open source community is that you get things like control net and then you see all these people do these just amazing things with control net. And then you wonder, I think from our point of view, we sort of go that that's really wonderful. But how do we end up with like a unified model that can do that? What are the bottlenecks? What are the issues? The community ultimately has very limited resources. And so they need these kinds of like workaround research ideas to get there. But yeah. [00:37:55]Swyx: Are techniques like control net portable to your architecture? [00:37:58]Suhail: Definitely. Yeah. We kept the Playground V2 exactly the same as SDXL. Not because not out of laziness, but just because we knew that the community already had tools. You know, all you have to do is maybe change a string in your code and then, you know, retrain a control net for it. So it was very intentional to do that. We didn't want to fragment the community with different architectures. Yeah. [00:38:16]Swyx: So basically, I'm going to go over three more categories. One is UIs, like app UIs, like mock UIs. Third is not safe for work, and then copyrighted stuff. I don't know if you care to comment on any of those. [00:38:28]Suhail: I think the NSFW kind of like safety stuff is really important. I kind of think that one of the biggest risks kind of going into maybe the U.S. election year will probably be very interrelated with like graphics, audio, video. I think it's going to be very hard to explain, you know, to a family relative who's not kind of in our world. And our world is like sometimes very, you know, we think it's very big, but it's very tiny compared to the rest of the world. Some people like there's still lots of humanity who have no idea what chat GPT is. And I think it's going to be very hard to explain, you know, to your uncle, aunt, whoever, you know, hey, I saw President Biden say this thing on a video, you know, I can't believe, you know, he said that. I think that's going to be a very troubling thing going into the world next year, the year after. [00:39:12]Swyx: That's more like a risk thing, like deepfakes, faking, political faking. But there's a lot of studies on how for most businesses, you don't want to train on not safe for work images, except that it makes you really good at bodies. [00:39:24]Suhail: Personally, we filter out NSFW type of images in our data set so that it's, you know, so our safety filter stuff doesn't have to work as hard. [00:39:32]Swyx: But you've heard this argument that not safe for work images are very good at human anatomy, which you do want to be good at. [00:39:38]Suhail: It's not like necessarily a bad thing to train on that data. It's more about like how you go and use it. That's why I was kind of talking about safety, you know, in part, because there are very terrible things that can happen in the world. If you have an extremely powerful graphics model, you know, suddenly like you can kind of imagine, you know, now if you can like generate nudes and then there's like you could do very character consistent things with faces, like what does that lead to? Yeah. And so I tend to think more what occurs after that, right? Even if you train on, let's say, you know, new data, if it does something to kind of help, there's nothing wrong with the human anatomy, it's very valid for a model to learn that. But then it's kind of like, how does that get used? And, you know, I won't bring up all of the very, very unsavory, terrible things that we see on a daily basis on the site, but I think it's more about what occurs. And so we, you know, we just recently did like a big sprint on safety. It's very difficult with graphics and art, right? Because there is tasteful art that has nudity, right? They're all over in museums, like, you know, there's very valid situations for that. And then there's the things that are the gray line of that, you know, what I might not find tasteful, someone might be like, that is completely tasteful, right? And then there are things that are way over the line. And then there are things that maybe you or, you know, maybe I would be okay with, but society isn't, you know? So where does that kind of end up on the spectrum of things? I think it's really hard with art. Sometimes even if you have like things that are not nude, if a child goes to your site, scrolls down some images, you know, classrooms of kids, you know, using our product, it's a really difficult problem. And it stretches mostly culture, society, politics, everything. [00:41:14]Alessio: Another favorite topic of our listeners is UX and AI. And I think you're probably one of the best all-inclusive editors for these things. So you don't just have the prompt, images come out, you pray, and now you do it again. First, you let people pick a seed so they can kind of have semi-repeatable generation. You also have, yeah, you can pick how many images and then you leave all of them in the canvas. And then you have kind of like this box, the generation box, and you can even cross between them and outpaint. There's all these things. How did you get here? You know, most people are kind of like, give me text, I give you image. You know, you're like, these are all the tools for you. [00:41:54]Suhail: Even though we were trying to make a graphics foundation model, I think we think that we're also trying to like re-imagine like what a graphics editor might look like given the change in technology. So, you know, I don't think we're trying to build Photoshop, but it's the only thing that we could say that people are largely familiar with. Oh, okay, there's Photoshop. What would Photoshop compare itself to pre-computer? I don't know, right? It's like, or kind of like a canvas, but you know, there's these menu options and you can use your mouse. What's a mouse? So I think that we're trying to re-imagine what a graphics editor might look like, not just for the fun of it, but because we kind of have no choice. Like there's this idea in image generation where you can generate images. That's like a super weird thing. What is that in Photoshop, right? You have to wait right now for the time being, but the wait is worth it often for a lot of people because they can't make that with their own skills. So I think it goes back to, you know, how we started the company, which was kind of looking at GPT-3's Playground, that the reason why we're named Playground is a homage to that actually. And, you know, it's like, shouldn't these products be more visual? These prompt boxes are like a terminal window, right? We're kind of at this weird point where it's just like MS-DOS. I remember my mom using MS-DOS and I memorized the keywords, like DIR, LS, all those things, right? It feels a little like we're there, right? Prompt engineering, parentheses to say beautiful or whatever, waits the word token more in the model or whatever. That's like super strange. I think a large portion of humanity would agree that that's not user-friendly, right? So how do we think about the products to be more user-friendly? Well, sure, you know, sure, it would be nice if I wanted to get rid of, like, the headphones on my head, you know, it'd be nice to mask it and then say, you know, can you remove the headphones? You know, if I want to grow, expand the image, you know, how can we make that feel easier without typing lots of words and being really confused? I don't even think we've nailed the UI UX yet. Part of that is because we're still experimenting. And part of that is because the model and the technology is going to get better. And whatever felt like the right UX six months ago is going to feel very broken now. So that's a little bit of how we got there is kind of saying, does everything have to be like a prompt in a box? Or can we do things that make it very intuitive for users? [00:44:03]Alessio: How do you decide what to give access to? So you have things like an expand prompt, which Dally 3 just does. It doesn't let you decide whether you should or not. [00:44:13]Swyx: As in, like, rewrites your prompts for you. [00:44:15]Suhail: Yeah, for that feature, I think once we get it to be cheaper, we'll probably just give it up. We'll probably just give it away. But we also decided something that might be a little bit different. We noticed that most of image generation is just, like, kind of casual. You know, it's in WhatsApp. It's, you know, it's in a Discord bot somewhere with Majorny. It's in ChatGPT. One of the differentiators I think we provide is at the expense of just lots of users necessarily. Mainstream consumers is that we provide as much, like, power and tweakability and configurability as possible. So the only reason why it's a toggle, because we know that users might want to use it and might not want to use it. There's some really powerful power user hobbyists that know what they're doing. And then there's a lot of people that just want something that looks cool, but they don't know how to prompt. And so I think a lot of Playground is more about going after that core user base that, like, knows, has a little bit more savviness and how to use these tools. You know, the average Dell user is probably not going to use ControlNet. They probably don't even know what that is. And so I think that, like, as the models get more powerful, as there's more tooling, hopefully you'll imagine a new sort of AI-first graphics editor that's just as, like, powerful and configurable as Photoshop. And you might have to master a new kind of tool. [00:45:28]Swyx: There's so many things I could go bounce off of. One, you mentioned about waiting. We have to kind of somewhat address the elephant in the room. Consistency models have been blowing up the past month. How do you think about integrating that? Obviously, there's a lot of other companies also trying to beat you to that space as well. [00:45:44]Suhail: I think we were the first company to integrate it. Ah, OK. [00:45:47]Swyx: Yeah. I didn't see your demo. [00:45:49]Suhail: Oops. Yeah, yeah. Well, we integrated it in a different way. OK. There are, like, 10 companies right now that have kind of tried to do, like, interactive editing, where you can, like, draw on the left side and then you get an image on the right side. We decided to kind of, like, wait and see whether there's, like, true utility on that. We have a different feature that's, like, unique in our product that is called preview rendering. And so you go to the product and you say, you know, we're like, what is the most common use case? The most common use case is you write a prompt and then you get an image. But what's the most annoying thing about that? The most annoying thing is, like, it feels like a slot machine, right? You're like, OK, I'm going to put it in and maybe I'll get something cool. So we did something that seemed a lot simpler, but a lot more relevant to how users already use these products, which is preview rendering. You toggle it on and it will show you a render of the image. And then graphics tools already have this. Like, if you use Cinema 4D or After Effects or something, it's called viewport rendering. And so we try to take something that exists in the real world that has familiarity and say, OK, you're going to get a rough sense of an early preview of this thing. And then when you're ready to generate, we're going to try to be as coherent about that image that you saw. That way, you're not spending so much time just like pulling down the slot machine lever. I think we were the first company to actually ship a quick LCM thing. Yeah, we were very excited about it. So we shipped it very quick. Yeah. [00:47:03]Swyx: Well, the demos I've been seeing, it's not like a preview necessarily. They're almost using it to animate their generations. Like, because you can kind of move shapes. [00:47:11]Suhail: Yeah, yeah, they're like doing it. They're animating it. But they're sort of showing, like, if I move a moon, you know, can I? [00:47:17]Swyx: I don't know. To me, it unlocks video in a way. [00:47:20]Suhail: Yeah. But the video models are already so much better than that. Yeah. [00:47:23]Swyx: There's another one, which I think is general ecosystem of Loras, right? Civit is obviously the most popular repository of Loras. How do you think about interacting with that ecosystem? [00:47:34]Suhail: The guy that did Lora, not the guy that invented Loras, but the person that brought Loras to Stable Diffusion actually works with us on some projects. His name is Simu. Shout out to Simu. And I think Loras are wonderful. Obviously, fine tuning all these Dreambooth models and such, it's just so heavy. And it's obvious in our conversation around styles and vibes, it's very hard to evaluate the artistry of these things. Loras give people this wonderful opportunity to create sub-genres of art. And I think they're amazing. Any graphics tool, any kind of thing that's expressing art has to provide some level of customization to its user base that goes beyond just typing Greg Rakowski in a prompt. We have to give more than that. It's not like users want to type these real artist names. It's that they don't know how else to get an image that looks interesting. They truly want originality and uniqueness. And I think Loras provide that. And they provide it in a very nice, scalable way. I hope that we find something even better than Loras in the long term, because there are still weaknesses to Loras, but I think they do a good job for now. Yeah. [00:48:39]Swyx: And so you would never compete with Civit? You would just kind of let people import? [00:48:43]Suhail: Civit's a site where all these things get kind of hosted by the community, right? And so, yeah, we'll often pull down some of the best things there. I think when we have a significantly better model, we will certainly build something that gets closer to that. Again, I go back to saying just I still think this is very nascent. Things are very underpowered, right? Loras are not easy to train. They're easy for an engineer. It sure would be nicer if I could just pick five or six reference images, right? And they might even be five or six different reference images that are not... They're just very different. They communicate a style, but they're actually like... It's like a mood board, right? And you have to be kind of an engineer almost to train these Loras or go to some site and be technically savvy, at least. It seems like it'd be much better if I could say, I love this style. Here are five images and you tell the model, like, this is what I want. And the model gives you something that's very aligned with what your style is, what you're talking about. And it's a style you couldn't even communicate, right? There's n
What does artificial intelligence have to do with an outdated practice that involved blasting shoe stores with radiation? Trust me, there's a connection. Or at least I make one. See omnystudio.com/listener for privacy information.
Jennifer and Pumps are joined by the hilarious Ross Mathews for one extremely "sh*tty" conversation - from the Delta flyer incident to the time Pumps took a dump in styrofoam cup on the highway. Jennifer brings up a rather barbaric childhood trend and Pumps has had it with scam calls and their sneaky new tricks. Come see I've Had It live on the Hot Sh*t Tour! More info & tickets available at https://linktr.ee/ivehaditpodcast and subscribe to I've Had It wherever you get your podcasts. Thank you to our sponsors: L'Oreal Paris: This episode of I've Had It is brought to you by the new L'Oreal Paris Bright Reveal Dark Spot Serum: https://bit.ly/3tmvJmB and Broad-Spectrum SPF 50 Daily Lotion: https://bit.ly/3ruWMLW. Dark Spots? Game Over! Discover the new Bright Reveal Dark Spot Duo! Visit Target online and in-store to buy yours today! BetterHelp: This episode is sponsored by BetterHelp. Give online therapy a try at betterhelp.com/HADIT today and get 10% off your first month. Happy Mammoth: Listener, you can get your first bottle of Hormone Harmony for 15% OFF if you use the code HADIT on the checkout page. Go to HappyMammoth.com and enter the promo code HADIT on the checkout page. Valid till November 30th. Shopify: Sign up for a $1 per month trial period at shopify.com/hadit and take your business to the next level today. Quince: Take the drama out of planning an outfit and upgrade your closet with Quince today! Go to Quince.com/hadit for free shipping on your order and 365-day returns. Bombas: Go to Bombas.com/HADIT and use code HADIT for twenty percent off your first purchase. Lume: Control Body Odor ANYWHERE with @lumedeodorant and get $5 off off your Starter Pack (that's over 40% off) with promo code hadit at lumepodcast.com! #lumepod Just Thrive: Get 20% off your first 90-day bottle of Just Calm and Just Thrive Probiotic today – Visit JustThriveHealth.com and use promo code: HADIT Blueland: Blueland has a special offer for listeners. Right now, get 15% off your first order by going to Blueland.com/HADIT Apartments.com: Apartments.com is making Earth more habitable than it's ever been, no matter where you're from. Apartments.com The place to find a place. Subscribe to our Patreon: https://www.patreon.com/IveHadItPodcast Follow Us: I've Had It Podcast: @Ivehaditpodcast Jennifer Welch: @mizzwelch Angie "Pumps" Sullivan: @pumpspumpspumps Special Guest: Ross Mathews @helloross
For the last 46 years, Dally in the Alley has been a vibrant celebration of Detroit's Cass Corridor neighborhood. On a new Daily J, WWJ's Zach Clark learns what drives the organizers to put on the free event without any corporate sponsorships. (PHOTO: Eric Seals/USA TODAY NETWORK)
Michigan lawmakers introduced legislation on Thursday to restrict gun access to those convicted of domestic violence misdemeanors. Plus, Dally in the Alley returns to Detroit's Cass Corridor this weekend, and more. Do you have a community story we should tell? Let us know in an email at detroiteveningreport@wdet.org.
The Dynasty War Zone, Season 1, Episode 9, Memphis and Jerry identifying trade targets for success in 2023. -- Check out our new Dynasty Deluxe Strategy Mode ➔ www.playerprofiler.com/dynasty-league-rankings -- Memphis (@DWZMemphis) and Jerry (@JerrySinDWZ) provide valuable insights into coveted strategies for being better dynasty players. They bring a wealth of knowledge, combined with their unique perspectives on game theory, trades, player value, and much more. From uncovering hidden gems to analyzing trade scenarios, they'll guide you towards making the right moves to secure that league championship. Join the community of dedicated dynasty enthusiasts and engage with the hosts by rating the show with a 5-star review on iTunes. Don't forget to follow them on Instagram and Twitter @DynastyWarZone for the latest updates, insights, and interactive discussions. In this episode, join Memphis and Jerry as they share the players they're looking to add to their "Win Now" dynasty rosters. They'll be revealing potential game-changers that could lead you to a league championship and make dynasty trades even more worthwhile. Let's revolutionize the world of dynasty fantasy football together. Tune in to The Dynasty WarZone and gain the winning edge you've been seeking. Your league championship dreams are within reach, and we're here to make it happen. Happy Wednesday, and thank you for being part of our journey. -- Get the Breakout Finder: Apple: https://apps.apple.com/us/app/breakout-finder/id1494942321 Google: https://play.google.com/store/apps/details?id=com.rotounderworld.breakoutfinder --
We can find power and strength in service, involvement, and commitment during turbulent times. Access the speech page here. Support the show: https://ldsp-pay.ldschurch.org/donations/byu/byu-speeches.htmlSee omny.fm/listener for privacy information.
Sam Newman, Mike Sheahan and Don Scott - 'You Cannot Be Serious'
Siwar Assili e Maria Dally nasceram num lugar a que chamam territórios de 48. Ocupados em 1948. Ou Israel, como reconhecido por uma maioria dos Estados. Identificam-se como palestinianos, mas o seu passaporte é israelista. Moram a 30 quilómetros de Tel Aviv e aos olhos do Estado são israelitas árabes. Então, o que é ser uma pessoa palestiniana aqui? E o que significa viver com o seu opressor? Vem celebrar connosco o 7.º aniversário do Fumaça no dia 25 de junho. Inscreve-te aqui para sabermos com quem contar: https://calendly.com/fumacapt/aniversario-fumaca-7-anos?month=2023-06Ajuda-nos a ser a primeira redação profissional de jornalismo em Portugal totalmente financiado pelas pessoas: https://fumaca.pt/contribuir/?utm_source=podcast+appSee omnystudio.com/listener for privacy information.
Our destination is fly-fishing the mighty White River in Northern, Arkansas with Chad Johnson, owner of CJ's White River Outfitter. The White River below Bull Shoals Dam offers over 55 miles of quality trout water and is world-famous for Brown trout the size of your leg, including 38 lb. brown trout caught just a few years ago. Chad grew up in Mississippi catfishing on the Perl River until he was invited on a weekend fly-fishing trip to Arkansas. Even after catching only 4 trout, he became obsessed with fly fishing, moved to Arkansas to work at Dally's Ozark Fly Fisher, and now is the head guide. Chad shares his expertise on targeting big Browns, springtime hatches, and nymphing year around, along with some funny stories about his good friend, Dave Whitlock. Bonus: where to wade when the water is high. With host, Steve Haigh Be the first to know about new episodes. Sign up at https://www.destinationanglerpodcast.com Top Flies for the White River: @DestinationAnglerPodcast on Facebook and Instagram Contact Chad: CJ's White River Outfitter: https://cjswhiteriveroutfitter.com/ | (601-557-2653) flyhead73@yahoo.com FOLLOW Chad on Instagram @flyhead and Facebook Please check out our Sponsors: Angler's Coffee - elevating the coffee experience for the fly-fishing community & anglers everywhere with small-batch coffee delivered to your doorstep. https://anglerscoffee.com/ | Facebook & Instagram @anglerscoffeeco New England Nets - makers of handmade hardwood landing nets for anglers everywhere. Exceptional quality that will complement your catch of a lifetime. newenglandnets@gmail.com | Instagram: @newenglandnets TroutRoutes - the #1 Trout Fishing app, helping you find new trout water so you spend less time on the road and more time fishing. https://troutinsights.com/ | Facebook @troutinsights Instagram @TroutRoutes Destination Angler: · NEW WEBSITE! https://www.destinationanglerpodcast.com · Get updates and pictures of destinations covered on each podcast: @DestinationAnglerPodcast on Instagram and Facebook · Join in the conversation with the @DestinationAnglerConnection group on Facebook. Comments & Suggestions: host, Steve Haigh, email shaigh50@gmail.com Available on Apple, Spotify, or wherever you get your podcasts Recorded April 21, 2020. Episode 90 Music on the show by A Brother's Fountain, “Hitch Hike-Man”. Podcast edited by Podcast Volume https://www.podcastvolume.com/
See what happens when 2 suburban dads (and 1 sister) chat about their lives. We covered a lot in this discussion, and we're excited for you to hear it!
Sometimes it feels like I'm dipping my toes into several ponds at once - and sometimes... that works for me. Other times, however, I thrive in a bit of structure. Routine is something I rarely seek out, but daggone it's valuable as hell for us neuro-spicy types.WATCH: www.youtube.com/juliemericaGET A MONTHLY NOTE FROM ME: www.makeyourdamnbedpodcast.comBUY SOME MERCH: https://www.teepublic.com/stores/make-your-damn-bed-podcast?ref_id=27657TUNE IN ON INSTAGRAM FOR COOL CONTENT: www.instagram.com/mydbpodcastOR BE A REAL GEM + TUNE IN ON PATREON: www.patreon.com/MYDBpodcastThis show is sponsored by BetterHelp. Get 10% off your first month of online therapy at www.betterhelp.com/makeyourbed The opinions expressed by Julie Merica and Make Your Damn Bed Podcast are intended for entertainment purposes only. Make Your Damn Bed podcast is not intended or implied to be a substitute for professional medical advice, diagnosis or treatment. Get bonus content on PatreonSupport this show http://supporter.acast.com/make-your-damn-bed. Hosted on Acast. See acast.com/privacy for more information.
Isabella Janovick was born in Oceanside, CA and grew up doing team sports (basketball and swimming) then later got into extreme sports, snowboarding, surfing and skateboarding. After breaking her back snowboarding at the age of 19 and almost being paralyzed, she switched to a “safer” sport - running. Isabella began to run local road races and in 2014 met Nickademus Hollon at an outdoor bootcamp. Nick had just launched his coaching business and Isabella was one of his first clients. They ran together often, he introduced her to the trails and the ultra world and in 2015 Isabella ran her first ultra, the Lake Hodges 50k. This same year Isabella was also diagnosed with Ménière's Disease, which causes hearing loss, imbalance and vertigo. Isabella aka “IzzyB” continued to grow with the sport, increasing her mileage and started her own running club Ultrabuds. In 2016 she raced her first Spartan, and usually participates in 1-2 races a year in between ultras. Professionally Izzy owns her own marketing and communications company where she works with running brands and race directors such as Injinji, Spring Energy, Nathan Sports and the Endurance Race Series, to name a few. She is also a Professor of Marketing part-time at two universities and advises senior experience projects. When Izzy isn't running or posting on Instagram, she spends quality time with her family in both San Diego and Idaho, husband and two rescued pitbulls Dally and Romanoff. More from me @hilsport55 Go to www.athleticgreens.com/TRWP to get 5 FREE travel packs and one year supply of Vitamin D! Check out oladance headphones at www.oladance.com/trw and use discount code TRW20 for 20% off! Get free shipping on any order from Goodr at goodr.com code TRW
In this episode, Ava and Aaron interview the returning LI Champion, Lady Bruins basketball team. This interview features Dally Moreno, Katelyn Simpson, Renelle Grannum, and Peyton Dulin. They discuss their season, past performances and the pleasure of working under the legend Coach Catapano.
Body builder and muscle goddess Ranya Dally joins us to talk about life in the gym, growing up as a dancer, how to find your niche, and so much more! Give Ranya a follow on Instagram @ranyamdally Leave a voicemail for Molly & Laura to answer on the show. Share your stories, questions, comments on episodes, anything you want us to hear! 702-900-6446 is the number to leave your voicemail! You can also EMAIL the show at twndpodcast@gmail.com - we love hearing from our listeners, so get involved! WE ARE NOMINATED FOR AN AVN AWARD! Vote for us to win! https://avn.com/awards/voting/favorite-adult-podcast WE WILL BE AT AVN 2023 Jan 4-7 at Resorts World in Las Vegas! TWND has a booth! We will have live interviews, VIP meet-and-greet, photos, signings, merch, and so much more! Get your tickets: https://www.eventbrite.com/e/avn-adult-entertainment-expo-january-4-7-2023-tickets-301117078287 You can follow the podcast on Instagram @twndpodcast and follow Molly on Instagram/TikTok @thisredheadissfw and follow Laura on Instagram/TikTok @whamglamartistry - SPONSORS - Support the show you love so we can keep bringing you episodes! Check out our amazing sponsors: Get hydrated and STAY hydrated with the amazing taste of LiquidIV with a special deal for my listeners! 15% off when you use code TWND or this link: https://glnk.io/75r0/totallywholesomenotdirtypodcast | #LIVPartner Do you need a new ring light? Want to gift something to that cute streamer that is just as adorable as they are? Use my affiliate link and help support this podcast! Shop Kawaii Lighting for all your RGB ring light needs: http://kawaiilighting.com/discount/TWND?redirect=%2F%3Fafmc%3DTWND #mentalhealth #girltalk #twnd #twndpodcast --- Send in a voice message: https://anchor.fm/twnd/message Support this podcast: https://anchor.fm/twnd/support
On May 6th 1996 Sherri Renee Guess Dally disappeared in California, leaving her two young sons without a mother. Her loving husband was the epitome of a grieving spouse and did everything he could to help police bring Sherri home or was that all an act?Support the showAs always we want to thank each and every one of you for the continued support! If you haven't done so already please follow us on Facebook and Instagram. We would also LOVE if you would write us a review on apple podcasts! XOXO,Ashley & Sierra CLICK HERE⤵️⤵️⤵️https://linktr.ee/weeklydoseofwicked?utm_source=linktree_profile_share<sid=6148575e-7853-4821-ae73-dc352c3340ab
Richard Freeman returns to the show to update us on the exciting expedition he has just returned from on the hunt for the Orang Pendek in Sumatra. Richard, Carl Marshall and Geordie Jackson, along with their guide Dally spent a month staking out several locations on behalf of the CFZ on the first of several new expeditions the CFZ are undertaking in the next couple of years. We here about the new witnesses they interviewed, the new locations they explored, a tantalising close encounter and the amazing footage of something totally unexpected that the team stumbled across by accident. Richard also gives us a sneak preview of his new book, which sounds very weird, to say the least! Thank you as always to Richard for joining me on the show. His last book can be found here: The CFZ can be found here: Richard can be contacted via doctor3uk@yahoo.com Our Patreon is now live, with bonus content, early release of the regular show, articles and more. Join here now for the flat fee of $4 a month which is a bargain! Don't forget, you can now show your support with our Merchandise shop on Redbubble! Check it out here! We are also now on Vburl - check us out here: You can join us on Facebook, Twitter and Instagram as well. You can also subscribe to our YouTube channel! Email us at mysteriesandmonsters@gmail.com with any feedback, guest suggestions or if you'd like to appear. All artwork by Dean Bestall and the show was produced by Brennan Storr of the Ghost Story Guys. Mysteries and Monsters is a part of the Straight Up Strange Network #Cryptids #Monsters #RichardFreeman #OrangPendek #CarlMarshall # #CFZ #InSearchOfRealMonsters #Cryptozoology #CentreOfForteanZoology #GeordieJackson