In short
Podcast Summary: This Week in Startups - E1873
Episode Title
2024 AI Predictions and Bets with Sunny Madra
General Overview In this episode, host Jason Calacanis and guest Sunny Madra discuss their predictions for AI in 2024, making specific bets on various developments in the industry. They engage in friendly wagers, each betting $1,000 on the outcomes of their predictions.
Key Segments
Introduction
- Podcast Introduction: Jason welcomes listeners to the first episode of 2024, jokingly referencing the holiday break.
- Guest Introduction: Sunny Madra joins the discussion to explore AI predictions for the year ahead.
Bets and Predictions Overview
- Bet Structure: They decide to make a series of bets regarding AI, with a $1,000 wager on each bet.
- Review of AI Developments: Discussion about AI advancements that occurred during the previous year, setting the context for their predictions.
Detailed Bets and Discussions
Bet 1
AI-generated Billboard Top 100 Song
- Prediction: An AI-created song will make it to the Billboard Top 100.
- Arguments:
- Sunny (Over): Believes it’s technically possible, citing advancements in AI lyric generation.
- Jason (Under): Skeptical about the music industry's acceptance of AI-generated content, suggesting industry politics may prevent it.
Bet 2
AI Mimicking Jason's TWiST Segment
- Prediction: AI will replicate Jason's style in a TWiST segment that listeners won't recognize as AI-generated.
- Sunny (Under): Thinks people will not be able to recognize it as AI.
- Jason (Over): Believes the distinctiveness of Jason's delivery will be hard to replicate authentically.
Bet 3
AI-generated Pixar Short
- Prediction: An AI-generated short film comparable to Pixar's quality will gain mainstream attention.
- Sunny (Over): Anticipates the rapid advancement of generative AI will make this possible.
- Jason (Under): Doubts its acceptance in mainstream recognition, citing industry standards.
Bet 4
Deployment of Small AI Models on Smartphones
- Prediction: Small AI models will be available on iPhones and Android devices.
- Arguments:
- Sunny (Over for iPhone): Given Apple's advancements in AI.
- Jason (Under for Android): Suggests Google might lag in deploying such technology effectively.
Bet 5
Popularity of Open-source Multimodal Models
- Prediction: An open-source multimodal model will become the most popular AI model by usage.
- Arguments:
- Sunny (Over): Confident in the rise of open-source accessibility.
- Jason (Under): Believes that proprietary models like ChatGPT will maintain their lead due to established resources and branding.
Additional Discussions
- AI in the Music Industry: Highlights concerns regarding the music industry's gatekeeping and its implications for AI.
- Legal Implications: Discussion on the New York Times lawsuit against OpenAI, discussing copyright issues and the evolving landscape of content creation.
- Technology and Society: Insights into how advancements in AI, particularly reasoning capabilities, could impact various industries and society.
Conclusion
- Wrap-Up and Future Outlook: Jason and Sunny express optimism about the future of AI, emphasizing the potential for innovation while acknowledging the challenges that come with rapid advancements.
Key Takeaways
- The conversation reflects the evolving nature of AI and its integration into creative industries.
- There is a tension between technological advancement and traditional industry practices.
- Potential legal ramifications regarding content ownership and copyright in the age of AI are significant.
- Both guests agree on the exciting potential of AI but recognize the need for careful consideration of the ethical and economic implications.
Links and Resources
- Suno AI: [suno.ai](https://www.suno.ai/)
- TinyLlama: [TinyLlama on Hugging Face](https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0)
- LLaVA: [LLaVA Project](https://llava-vl.github.io/)
- Miro: [Miro Sign Up](https://miro.com/startups)
- NetSuite KPI Checklist: [netsuite.com/twist](http://netsuite.com/twist)
- Vanta SOC 2 Discount: [vanta.com/twist](http://vanta.com/twist)
Follow the Hosts
- Sunny Madra on X: [@sundeep](https://twitter.com/sundeep)
- Jason Calacanis on X: [@jason](https://twitter.com/jason)
- This Week in Startups: [twistartups.substack.com](https://twistartups.substack.com)
This episode encapsulates a lively exploration of AI's future, blending predictions with insightful commentary on the industry dynamics.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Transcript
Automatic transcript. May contain errors.0:00I never told anybody this but Fox, Ruper Murdoch, put in a late bid to buy weblogs Inc. okay when you already tried when you had the aol deal on the in the red zone and they were like hey how about plus three million and i was just like how about no like and i just told them like if you're going to give an offer that's 30 more than what they're offering which would have been nine million dollars and i'll consider it but if you're just gonna we're not playing the prices right here sorry respectfully rupert yeah you sold early on that one jacal wow 18 months from zero to 30 million yeah wow dude i needed to secure the bag i mean that was that was the Setup of all setups.
0:35Once you get your first 10 milli, you're dangerous. That's what I always tell people. Get that first 10 million in the bank and man, you have to give money. Just to think of how it's been. It's all good though. All good. This Week in Startups is brought to you by Miro. Miro helps take ideas from in your head to out there in the world with its ability to democratize collaboration and input. Sign up for free at Miro.com slash startups. NetSuite. Once your business gets to a certain size, the cracks start to emerge. Things you used to do in a day take a week. You deserve a customized solution, and that's NetSuite.
1:13Learn more when you download NetSuite's popular KPI checklist absolutely free at netsuite.com slash twist. And Vanta. Compliance and security shouldn't be a deal breaker for startups to win new business. Vanta makes it easy for companies to get a SOC 2 report fast. Twist listeners can get$1 ,000 off for a limited time at vanta.com slash twist. All right, everybody, welcome back to This Week in Startups. It is 2024 and it's Monday. No, it's not Monday. It's Tuesday. But we're going to treat Tuesday like it's a Monday because you love, you all love this. This Week in Startups AI. Every Monday, and we'll do it on Tuesday this time because Monday was a holiday.
1:57happy new year i get together with my good friend sandeep bonjour there he is sunny how you doing sunny happy new year i am doing great happy new year i have my mid journey generated ah yes mid journey yeah you know it's interesting to bring up mid journey we with the big news that dropped this past week while we were on break was of course this new york times lawsuit and then i saw on tiktok and this tiktok that said mid journey uh for audio yeah had specifically sampled all of the major artists to make the styles. Now, I don't know if this claim is true or not, or if we have any backup to this.
2:33Did you see a story break about MidJourney and musicians? No, I didn't know about MidJourney doing the musician stuff. I didn't see that. It must have been during the break. So, yeah, we're going to keep an eye on that. But there are some MidJourney artists lists. Now, is MidJourney a company or an open source project or both? Because this is another one of these situations where I'm a little confused. It is not an open source project. It is a company, and it is a not venture-backed company. and it is gone from zero to 200 million in revenue without any venture dollars wow i was totally unaware of this yes and so okay they are leveraging open source to build what they're doing but they are not open sourcing their model so you can give a prompt for mid journey on artists i think so anyway this is good i'm just curious when you saw that new york time story and by the way today we're going to do some predictions and bets yep for what's going to happen in ai in 2024 so we're going to make bets we're going to make five bets explain to the guys what we're going to do yeah what we're going to do is we're going to make the mid-year because like year is like really long so let's make the mid-year and then and we're going to do over under so you know person has to take either side of it and they're going to be things that we've touched on throughout the last couple of months um and they're all going to be pretty relevant, I think, to how 2024 shapes up.
3:53So the bets are also encapsulations of our predictions or my predictions more accurately, but I think you would align there. Well, yeah. And this is going to be great. So we're going to say, hey, six months over under, we each pick over under, we do five bets,$1 ,000 a bet. So if somebody sweeps, it's five dimes. If somebody doesn't sweep, it could be two dimes, three dimes, four dimes, who knows what's going to happen here, but this is big money on the line here. This is not inconsequential, so we got to think about this and that means on august 1st we're going to check in on the bets july 1st oh sorry i meant july 1st july 1st yes the seventh month right yeah perfect okay so let and then at the end of this episode let's talk about this new york times lawsuit if we have time because everybody wants to hear your comments on it sandeep because okay you're the technologist and i am the content creator so we have this beautiful peanut butter chocolate here we could have a really interesting granular discussion what's our first bet okay the first bet so there There will be a billboard top 100 song created by AI.
4:53A billboard. Yes. Top 100 song created by AI. Yes. Okay. Now, what I'm going to do is we just don't want to do predictions. I'm going to show you the latest in AI lyric and song creation. To give us a little warm up here to just level set where we're at and how much further we'd have to go. And we're going to talk a little bit about what happened earlier in the year as well. And we don't think there's been one already. Because if you told me there's been one and one of the Grimes mashups had made it, I would have not. I would have believed it. But I don't think one has because we would know that.
5:31Yeah. And I don't think it went into the top 100. Got it. So it has to be in the top 100. And it has to be 100 % created by AI. Not this is Taylor Swift and then they do a little AI on it. it's an AI artist or an AI representation of with the AI representation of Grimes count? I think it would. Yeah. We're going to allow that to be part. That's okay. That's sufficient. If someone takes an AI voice. And so I think we have Nick on with us, Nick, you have just for a quick recap, let's go backwards. The AI Drake collaboration. Can you throw that on for us, Nick? Yeah. So this is earlier in this year, remember there was a song hard on my sleeve that everybody thought was real.
6:12And this was sort of in the early stages of generative AI where people were kind of confused about if Drake actually did this or not. And here's just a video. This isn't the person who made it, but here's just a guy basically recreating how they did it very simply. Vocals. She know what she need. Oh, I need. Oh, she bless. Uploading those vocals to an AI generator. She know what she need. Oh, I need. Oh, man. A de-esser. She know what she need. Auto-tune artist, making sure that he's in key. Some light reverb on top. So this is the final result. Got my heart on my sleeve and a knife in my back.
6:42was with that 821 i love him that's my brother that's my i kind of like producer bay i know i mean what's up with producer bay that guy should be i mean he's handsome he's got the accent and he's got i mean do you see his hand eye coordination the speed at which he's editing that stuff yeah man producer ai bay has got potential man he's got to drop some more tracks so what you're saying here by the way is his name shout out that was six months ago right so that was uh you know So that's what, you know, it's out there. People have figured out how to do it. Anybody can take the tools. It's indistinguishable.
7:16It's indistinguishable today. So let's just put that out there. So it's indistinguishable. So what we're really betting on here is, and I guess I'll go first on this one. You go first on the next? No, no, no. Before we do the bet. Oh, it's one more piece of evidence here. Okay. Now, at the end of last year. Exactly. Because we've got to have, yeah. It's okay. It's well produced. Yeah. You're very excited. This is awesome. So at the end of last year, just a couple of weeks ago, a week ago, Suno.ai was released and they already have their own trending list. Like you would see in a Spotify, exactly.
7:49And, you know, people have created different, you know, obviously there's a Christmas tune up here. Here's a bluegrass inspirational lively. And so I'll just kind of rewind this back and hit play.
8:06okay and so what this does is you can create and what i did and i'm just going to keep going forward here i actually uh made a paw patrol drill rap song and so that was my uh my prompt was using the characters of paw patrol make a you know breaking all copyright violations here you You got the Paw Patrol IP, you've got whatever artist. I got their names. Rap style 50 Cent in here. I love it. I'm going to play this and let's go back here. Just rewind this. Cease and desist coming. We run this game. Paw Patrol, yeah, we got the fame. We're not sure we never back down. Just rush and bubble, we run this out.
8:43We're walking here, it's never been a clear. We got the bus number. Okay, so. Oh yeah, man, I can see kids digging that. Yeah, and so again, now we've gone from what the, we saw in the YouTube video where you You had to at least put the lyrics in yourself, write the lyrics. This is all just prompt-based. Yes. The music is prompt and the lyrics are also generated just on the theme. Full service here. AI, what we call AI all the way down. All the way down. Got it. Yeah. So that's what we have. You can do sort of be really bespoke or you can just say, type in prompts. There's a Spotify. One of those songs makes it in the top 100.
9:19Over and under, July 1st. Right. Okay. So I'm going to take the over. Now let me explain why. Okay. it's clearly this is easily technical technically possible in fact it's probable and in fact we probably could have had a billboard 100 in the last year or so so what you have to understand when making this bet this is my thinking behind the back now wait have you picked your choice over under well no what we're going to do is you got to go first so next time i get to go first so okay so i have to take the other okay great sorry so okay so you have the under by default i took over over the reason it's the over and the reason i'll win this one hands down is because nothing happens in the music industry without the music industry making it happen this is an industry it's kind of like china right like they'll decide what the rem and b is they'll decide the stock price they'll decide inflation they'll decide what gdp is and yeah everything is decided by the people in charge so the music industry runs like a communist country or you know any other you know dictatorship so the music industry you'd have to ask would they want this would they want it now i think the music industry is a little scared of doing this right now and opening up this can of worms with artists so the only way this could happen is if a true artist like taylor swift um who has more leverage over the music platforms than the platforms have over her right so how many artists have more power than the music industry itself it's taylor swift it's jay-z it's beyonce and you you could tell this because remember beyonce jay-z and taylor swift all took the music off of spotify at different points in time to make a point yep yeah so i don't know who else is in that rarefied air but i think it's those three so unless one of those three says i'm putting out this ai jay-z track and i don't think jay-z would ever do it i don't think beyonce would ever do it and i don't think taylor swift would ever do it unless they were doing it as some sort of a goof or proof of concept so i think the music and i don't think the music industry is ready for this um because it would create all kinds of union issues contract issues and other things that would trigger artists to not want to participate in the music industry or it just creates a fight and tension that they don't want it's smoke they don't want yet so that's why i'm picking the over and that's why i'm going to win a thousand dollars okay well i'm going to take the under because think about remember rebecca black friday yes that that kind of one hit wonder and we have those all the time and i think given these tools someone will create at least a one eight one hit wonder to begin with that'll get into the top 100 this year so i'm actually okay with the under on this one okay that's fine you think that the music industry actually that that 100 list is not picked and sorted over by the music industry you think you could actually break into it it can be an independent an independent song right it doesn't have to be on a label yeah right any song in the billboard 100 the billboard 100 though i think is bought and paid for it that's all i'm saying i don't have evidence of that i just think this is like a tony soprano kind of situation how did rebecca i think they let her on i think they let her on i think anybody gets on that list it's because they let them on i think it's okay you can look at the history of the music industry it's a little bit of the inside i don't know i don't know that side of the world there but i think it's like one of these the few remaining industries that is run by a cartel okay and from what i understand from friends of mine who are in the music industry and you can look up stories of people putting guns to people's heads and sign over these royalties and you should get rob goldberg on for an episode because he knows this stuff really well he knows this music industry but i just think the music industry is still like one of the few remaining gangster yeah ask driven industry that's why i'm picking the other but i could be wrong who knows okay awesome all right so i'm gonna be under great and with action item let's come into me all right okay awesome all right so we're done the first one yum founders always ask me for pitch deck punch-ups how to make my pitch deck a little better well i've got some great news for you we work with the team at miro uh you know like hero miro m-i-r-o that awesome whiteboarding software i've told you all about and when we worked with them to create an amazing pitch deck template for founders which you can see if you're watching the video right now or if you're listening you can see this pitch deck at miro.com slash miroverse and do a search for Pitch Deck.
13:28M-I-R-O dot com slash Miroverse. M-I-R-O-V-E-R-S-E. I want you to go check this out because our founding university of participants, they love this template. They use it all the time. And so go ahead and check out that Pitch Deck template. You're going to love it. And if your team is hybrid or fully remote, Miro is incredibly useful for you because it's like an old school whiteboarding session. You know, those great ones, but this one is distributed and asynchronous. Let's everybody brainstorm ideas, collaborate on projects from anywhere in the world. This software is so popular. And when you think of Miro, I want you to think of zero to one, but faster.
14:01If you get all those ideas out on your whiteboard with Miro, you're going to have all kinds of creative breakthroughs. And it's hard to do creative breakthroughs when you're just in text. So Miro is much more than just a simple digital whiteboard. Your team can collaborate on everything, planning, research, design, feedback cycles, and faster inputs equals faster outcomes. And velocity is how startups win. So to access our new Miroverse template and thousands of others sign up today for a free miro account at miro.com slash startups m-i-r-o.com slash startups that's miro.com slash startups to sign up for free okay so the next one is um ai jcal doing a short episode of twist okay define short like 15 minutes like a 15 15 minute episode of twist like a first segment like a segment let's say a segment because a segment is 12 minutes let's you go 12 minutes 12 minute segment or less 12 minute segment exactly so a 12 minute segment and and basically you know we'll put it out there and no one will know that it was um done via um ai okay so this is a setup this is a setup we're gonna make an ai of me yeah we're going to have me comment on a news story yep and then we're going to see if the audience we won't tell the audience we're doing this we'll put it in one of the random segments of a news program and then we'll see if the audience can tell the difference that's the bet will the audience be able to pick it up yeah okay so this what we're about to watch is a video of a vc speaking yes using hey gen using hey gen gen already congrats to the hey gen team on their rocket ship growth this year.
15:43We at Conviction are so excited to support Joshua and Wayne in their mission to make AI the new camera. They are democratizing expensive video production. I love seeing all the creator, marketing, sales, and education use cases for HeyGen, as well as the ability for everyone to extend their reach by speaking fluently. Okay. Well, wow. So that is, look, from the video, it's perfect. The audio is a little bit, you know, you can tell it's a little robotic, but again, no mistakes. The problem is no mistakes. yeah so on this one i'm gonna even toss you the over under you can pick if but where you're at but if you want me to go first i'll repeat repeat what the bet is repeat what the bet is the bet is we're gonna have a segment okay of which is it on control this weekend startups you have to wear the same clothes i'll wear like this polo shirt um actually no let's not do that because then people can it'll be like a segment be like hey this is a segment we recorded previously or something we'll find a way not to like trip people off on it and we'll put it in and it'll be you talking about a subject so not you and a guest or anything like that but maybe like copyright solo news solo news ai copyright something like that and we'll put it out and we'll see if people can pick it up great so we'll do what we do is actually here's the better way to do it we just release it on the twitter we do that we do segments so we just do like a news up hey is jake i'm talking about this yeah and we'll just see in the replies in the replies if people say is that ai yeah that's it so the replies if they say that is that ai yeah so wow hmm no i'm gonna let you you we have to do this fairly i'm gonna take the under i'm gonna take the under okay i would have taken the under as well okay the under you're taking is that people don't notice people don't notice they interact they comment on it you know yeah they don't notice okay no no one per and look yeah um now people are gonna be looking out for it so you gotta have uh so we'll go with the majority majority yeah majority yeah so the majority of replies or something or let's just say it's significant it can't just be one but it's like yeah okay well nick how do we make this bet a little bit tighter here how would you make it tighter how about most people don't notice yeah i will say i do think this is this is going to be over by a lot and the reason is because i've done this before and i've trained jason's voice on a couple of different platforms 11 labs by the way for audio only is the best one by far jason has a very unique cadence of speaking and he has a unique pitch and it's very hard for any model that i've tried to train on jason for it to get it right so like for instance remember the jason peter jason jordan peterson thing we did yeah that was that perfectly it nailed jordan peterson but for you it wasn't exactly perfect and i think that's because you have a really interesting you have a unique yeah of cadence yeah yes way of talking i change it too because i could be excited to be up here yeah sometimes i can lay back i can lay back and be thinking about that stuff you know i have a yeah because a performer you can see i do different things keep the audience's attention all right i like it i like it i like it i get the over i will see what happens here uh very interesting let's go to our third bet our third bet i feel like right now i got two dimes in my pocket yeah okay so the third bet is ai pixar ai pixar right we had a bet on this yeah yeah and so um you know there's there's a couple things we're going to just pull up here just to get kind of state of the art level set we're going to level set to level set this is just you know kind of state of the art someone made a little short here oh i remember this yes was this mid journey or who is this this is a combination of tools i think like pika there's a bunch of tools that people we're using this now because you're you're gonna generate the characters in probably in kind of one of the big image models and use a video models to there's a few different ways people are going to go at it one would be straight text to video others will be like stitching together two or three different tools and then um you know in relation to this uh we saw something else over the holidays which i'm going to try to pull up here as well uh so give me a second this is amazing yeah i mean yeah this this looks like a i would say cable tv movie more than a pixar but you know like uh the next level down yeah okay but i have to say that the that gap i just want to point out the gap between what's a movie and what's a tv show you know in terms of budget of special effects that gap is closing and so you see this in the star wars films uh and the star was tv shows like and or or like mandalorian there are moments that are indistinguishable between those two or ashoka where like especially when they're doing space battles and stuff like that you're kind of like that could be a movie that could have been you know rogue one star wars empire you know whatever or it could have been a tv show yeah yeah and i'll tell you one there's a there's this thing called the monster series um with titans in it so like godzilla king kong Godzilla vs.
20:41King Kong, Kong, Skull Island, and so my daughters and I have been watching that. And then there's a TV show out on Apple TV called Monarch, which is about the secret organization that since the time of Godzilla appearing when the nuclear bombs went off in Hiroshima and Nagasaki, that's kind of the folklore of it all, till now there's been this Monarch agency, this like stealth agency that tries to protect humans and monitor these titans, these giant monsters like moth or etc wow anyway my point here is the tv show is so well produced and the apple tv has got unlimited cash obviously so that factors into it but having just watched the movies at the same time as watching the tv show you cannot tell the difference difference yeah yeah there's no difference and so that's kind of a watermark moment movies that cost like the thing you're talking about is it like pacific pacific rim is yeah got monsters and stuff in it but they kind of copied and did something derivative from the godzilla series and i really like godzilla and there's a godzilla movie out that's a japanese language godzilla movie oh yeah it's been in the apple trending i've seen it right it is awesome i went to see the movie theaters with subtitles and i took my 14 year old to it and she loved it it is this and it really is about the japanese culture yeah uh and how they uh you know um lost the ability to go to war and not have an army and wow i don't want to spoil it for anybody but it is a very soulful deep personal story yeah with godzilla in it yeah it's awesome um like they're really kind of reimagining what godzilla means in terms of like nuclear war war in general uh you know and you know all this stuff it's really interesting anyway it might be godzilla year zero or something nick you can look it up here here's an example like just to build on that jake i'll do the last example then we'll do the prediction on this one here's an example where the voices are definitely done with a minus one by the way godzilla minus one well worth your time here's one where the voices are definitely done with ai because they are the voices of characters but the video is questionable whether it's cgi or not which is you know we're that means we're trending in the right direction so this is like a short rap video that uh this user username blurb on youtube created and it's like spongebob drill rap so we're just going to get into it right now
23:19hilarious
23:23this might win two awards because that seems like it could chart on billboard It's kind of a quick aside. So I don't know, like you see this on Jan 1st, the Disney trademark on Mickey Mouse expired. And now people have taken Mickey Mouse and created first person shooters and all kinds of stuff. Wonderful. Yeah, it's a Steamboat Willie version of him, I think is what they call it. The original, original one, right? And so, but yeah, you know, when you watch that video and the lyrics and especially, you know, played it for Nick, who is definitely in the SpongeBob generation, it's pretty wild because the lyrics, the voices, and then I don't know if the animations are AI or not, but definitely someone will be able to pull that off.
24:06So what's our bet here? What's our over under bet for July 1st? We're just going to build up on the other one and we'll solidify it in here. By July 1st, a Pixar short. that is also made it into the mainstream and people are talking about it that's why the billboard 100 thing is there people are talking about it people are excited by it it's it's coming up on nightly news short that is indistinguishable from an actual pixar short correct and it's circulating in non-technorati site circles it's it's you know people are talking about it on the national news so it's so good that the national okay that's good so it's so good it gets national recognition yes okay so it's not enough that it comes out it has to get national regular recognition i take the over i take the over i take the over i think it's not going to get national recognition i think it could like these things spread in on x slash twitter yeah i don't think it's going to spread anywhere else yeah i i'm okay with the under on this one too i think we're gonna seems like the best bet this is like the closest one for me i think there's so many creatives out there they have access to so much tooling and compute power that they are really going to push to create something uh for society and i it's gonna be fun so i'm good well here's here's here's something for everybody i'm gonna give you an idea okay um i don't think i i think what we're doing here is we're kind of challenging the ai community so this is a challenge the ai community take one of the pixar's uh films and make a short about a character from a pixar film that you think should have gotten more play i'm going to give you an idea there's the character anton ego in my favorite pixar film ratatouille and there is a scene at the end when it's his childhood and how his mom made him um ratatouille and how that dish you know changed his life anyway they show a quick scene of him it's just for a brief second they show him going back to his childhood in the final scene uh of the film when he says hey listen the life of a critic is not very important unless we're supporting new things so here we are you could make a film about anton ego as a child making ratatouille itself or learning how to make ratatouille from his mom or coming home from a rough day at school and then his mom version of the mandalorian you're doing a little you're doing a little spin-off character so and then i think in toy story you could do a spin-off character because i don't think they ever had woody or buzz lightyear have a child right there was never they never pursued that right like a woody had a son or a daughter they were ageless right they were ages so here's another idea for you because they did have woody get a girlfriend right yeah at one point what about if woody jr and woody has to like raise a kid and just have woody being a dad yeah so there's another great one woody's a dad and he's got like a little son or daughter who is you know dealing with whatever issues kids are dealing with these are trust me i have ideas i should have been a movie i always thought i should have run a movie studio i feel like it's not too late for me i think in in this you know next era that we're in jacob you've always been a content producer writing books and obviously silicon alley and you know twist and all in but i think now you can expand to that i think it's going to be your fingertips i love it i love it yeah so those are two good ideas for you somebody make them okay yeah here we go okay so uh to be clear i have the over you have the under under i don't feel very good about this bet so i feel like i got two out of three right now okay and i just inspired these lunatics in the audience to go do one of these ideas okay yeah i know and actually if you do either of those ideas i'll take you to lunch in the bay area so if anybody actually does those ideas and it's reasonably good i'll take you to lunch okay let's keep going okay your business gets to a certain size and we know the cracks start to emerge things that you used to do in a day are now taking a week or a month.
27:55You have too many manual processes. You know that. And you don't have one single source of truth. If this is you, you should know these three numbers. 37 ,000, 25, 1. 37 ,000? That's the number of businesses which have upgraded to NetSuite by Oracle. You know NetSuite. It's the number one cloud financial system. Streamlining accounting, financial management, inventory, HR, and more. 25. NetSuite turns 25 this year. That's 25 years of helping businesses do more with less, close their books in days, not weeks, and drive down costs. One, because your business is one of a kind. So you get a customized solution for all of your KPIs in one efficient system with one source of truth.
28:42Manage risk, get reliable forecasts and improve margins everything you need to grow all in one place so here's your call to action right now download netsuite's popular kpi checklist designed to give you consistently excellent performance absolutely free at netsuite.com twist that's netsuite.com twist to get your own kpi checklist netsuite.com twist so over the holidays a new model was released I'm going to pull it up here, which was called Tiny Llama. And it's a 1.1 billion parameter model run super fast. And so it's a 1.1 billion parameter model. It's tiny and it was trained just in 90 days using just 16 A100 chips.
29:27So a couple of things here, it's very small to run. It was trained with, you know, basically compute that's accessible by any person in the world. and you can see that on a local device this thing runs at like a blazing speed all right so here i'm going to just show the speed of one of these tiny models running and you'll just see here someone's taken that tiny llama and you can see how quickly it's producing results right and so describe what you see on the screen there so what we're seeing is someone on their m3 max which is just an m3 mac which is the latest chip you can get it in a laptop or an iMac yeah only available in the pros right now the iMac and the Air only have M2 but the M3 is so powerful yeah that you could train well it's not training this is just running inference now so so running it explain the difference between since we're here you know we like to educate everybody one more time training versus inference so training is what you do to create a model so you take all of the information available to you, including copyright on copyright information.
30:34Sure, why not? You feed it into your training infrastructure to build a model. And it takes sort of a rule of thumb you can use is the number of parameters a model has. So in this case, this tiny llama I just showed earlier is 1 billion parameters. It probably took 1 trillion tokens to train it. And when you get those bigger models, like we see with llama 70b then you have to have a hundred times that amount right sorry a thousand times that amount so seven that would be 70 trillion right and then when you get into the and define what a token is it a token a word is a token a paragraph a sentence an image it's not quite a word but let's just say a word for for all intents and purposes it's a word right and so that's what you use to to train in so got it so here we're running inference on the m3 now you wouldn't be able to train this model on an m3 right now how you need those it's said on the previous page it took 16 uh a100s which you can go get in amazon they're not like the h100s this is the kind of previous generation of nvidia chip and what i'm showing here is running on your laptop this is like writing a story about the universe i'll just maybe zoom this out so instead of using chat gpt4 where you're sending your prompt up to a bunch of servers that are very powerful and it's sending the results down to you edge computing we would call that client server computing this is your computer is the server and the client in one you're running it locally so if you turn off internet yes and you're you have no starlink which by the way shout out starlink i was getting 200 megabit down i put it at my ski house yeah i have unobstructed oh my lord wow yeah it just took a jump i got over 200 megabit down more and more satellites right you know and so i don't know what to say i mean i'm kind of starting to think the fcc and the united states government is corrupt i mean the fact that they canceled that contract yeah i don't know if you've heard that episode of all in where we talked about it and i also had on twist uh last week i had the action commissioner on brendan car on shout out to my guys two times he's been on the pod i now actually think that there are people at work in our government i don't get all deep state here but i think there's a contingent of people in our government who are doing corrupt things i hate to break the news to people but people might have ulterior motives inside of our government yeah wild what yeah it definitely is i mean that one makes no sense no no one can explain it well here's the thing if you cancel those people's free starlink because they're getting it at a discount or free they're going to order starlink in the interim anyway yeah because it's the best solution so this would be like saying to somebody like i don't know what's the best uh hands down best thing oh look an iphone right so let's say you were going to subsidize people to get an iphone you know two years before the iphone was available whatever because you had the digital divide you want to get him an iphone and then you're like you know we're canceling this screw steve jobs he's too successful we don't like him his politics whatever and then all those people had a choice of what phone to buy and then you know when that two-year contract came up 80 of them would buy the iphone anyway yes it's the best product of course so literally we're sitting here and it's like this pizza is incredible uh oh no i don't think this pizza is great it's like no no this is like the best pizza you know dave port and i gave it a 9.6 it's the highest rated pizza ever in pizza reviews and they're like yeah no and then everybody goes to that pizzeria anyway and buys it it's it's a total farce i know i'm talking about my guy but i'm it's jason i would mention that the key point here is that the free market is going to figure out what the best product is anyway but the reason that there's contention here is because the alternative is installing fiber lines which is going to take 10 more years and cost five times the price yeah yes for no reason and that money is going to be burned for no reason that's exactly correct producer nick coming in in his final 60 days here producing for this week and start us before he goes full time at all in and providing massive value in year three what is this year four for you working for me i don't know i honestly don't even know i don't want to trigger him wow crazy for a while he's been here for a while but now he's going to be working for chaman the chairman dictator good luck to you producer nick march first sunny do you have any uh do you have any experience in that realm i think it's gonna i think you're gonna have a really good you're gonna have a time you know he's very generous he's uh very kind um you know despite his uh gruff sharp elbows A lot of people, a lot of money.
35:03It's good. Let's keep, keep at it. It's good. I love that man. Don't get on the wrong side. I love that man. I know a couple of people got on the wrong side at Jamal. It wasn't pretty. It went to the mat with a couple of people. And just be on top of it. You always are with information, right? You know, he's, he's always very fair when you bring information to the table. So yeah. Yeah. He was your original investor, right? So deep in your, uh, he's been an investor in all my companies. Right. So yeah. All right. Chairman dictator. Let's do it. Okay. Um, yeah, just on Starling. i'll say i'll say the same thing i just i want to add one thing there which is it's i i just think the beauty now is uh for those folks that are struggling for connectivity i saw that star league's available at costco did you see that yeah no i didn't but that's like yeah nick i don't know you want to pull it up all right you know it was yeah no when you when you wind up at costco yeah it's game over yeah that's when you know you're a mass market product and everybody's aware of it like um you can get like i knew that wagyu and kobe beef was going to be a thing when it's at costco yeah look at this yeah there it is costco what i have i have a better one i have a better one hold on i have one where it's someone took a picture inside of costco i'm going to show you that look at this yeah yeah right there just grab one grab one you know what also get the also get the locks there's a really good deal on um nova scotia locks yeah but i mean my wife goes My wife goes to Costco.
36:25So my counter argument to the FCC thing. Look, I'm not supporting it because I think it's terrible. Is that go to your Costco, go buy one, set it up. That's it. And you don't need the government. I mean, that was my feeling on it. Listen, I'm happy for my friend and their company to save the government money and to get a great contract. That's awesome. I think that actually makes sense to me if the other option. Because I do believe the digital divide is real. Yeah. And I do believe that taxing people who are in cities, a modest amount, in order to make sure that we have broadband to people who are not in cities, that to me seems like a fair use of government, right?
37:03You don't want people on farms and in rural America. It wasn't fair when you had to pay several thousand dollars to get DSL or whatever it is, or phone lines. Now, again, my point is if you just go to Costco, no one should get that contract. If it's gone from SpaceX, it should be gone from everyone. Yeah, yeah, no, it should. Nobody gets it. and the government would be like hey you know what you want internet like a cell phone go get it from you know at &t yeah but i mean if you were a libertarian you're like hey every you chose to live on a farm therefore you don't get broadband i think that is like a very unnuanced uh position it's like a very brute force like i'm a libertarian government get out of our lives if we live in a society yeah and we want society to be fair and lift everybody if if one percent of Americans, let's just say 3 % of Americans, let's, you know, out of 100 million plus households, or some bunch of million households, there's 3 million people who can't get access to broadband.
37:55Would it not be okay for the 97 % to pay an extra$10 a month, or$5 a month, or$3 a month, whatever it is, in order to subsidize those 3 million, just to get some fiber out there and make sure they don't get left behind, because it is such a critical, essential piece of infrastructure, I think that's worth it. I also think it's worth it for electricity, and clean water and public education right so that seems like a fair thing to do in society while we wait for technology to make those things possible and if technology is not possible yet that's a great use of government dollars or our tax dollars we can decide as a society if that's a good use of our tax dollars i don't think anybody's arguing that kids on a farm or living in rural america don't deserve to have the same broadband as kids in you know and families who are in cities who have easer access so anyway shout out to the team at starlink and also the team at amazon and hughes who are also putting up low earth orbit satellites shout out to all three of them and the team at costco and the team at costco shout out to costco i'll never go to a costco that's too much craziness especially with my level of micro celebrity now yeah i don't know if you know what it's like for me to be in a public space anyway you do because you've hung with me i've been with you yeah it's very it's very disturbing you could have two or three people ask for a selfie every day it could be as much as three selfies a day after you realize how crushing that is on my email to have to do it for the people go to costco let them happen it happened you and i were out and somebody came over and they were the biggest fans and i was like happy to take a selfie they're like maybe later i was like excellent all right listen selling software is hard it's hard right now right 2022 2023 it's been a grind In 2024, it's going to be hard too.
39:37Everybody's making very thoughtful decisions. And the last thing you need is to slow your sales team down because you don't have your SOC 2 dialed in. So if you're a SaaS or services company that stores customer data in the cloud, you need to check out Vanta. Vanta will get your startup SOC 2 compliant easier and faster. Vanta makes it really easy to get and renew your SOC 2. On average, Vanta customers are SOC 2 compliant in just two to four weeks. Compare that to three to five months without Vanta. Vanta can save you hundreds of hours of work and up to 85 % on compliance costs. And Vanta does more than just stock too.
40:09They also automate up to 90 % compliance for GDPR, HIPAA, and more. You can't afford to lose out on major customers because of silly stuff like lacking compliance. Just work with Vanta. Get your compliance automated and tight. Tight is right. Close those big deals, the lighthouse deals that send all the other customers to you. Here's the call to action. It's very simple. Fanta is going to give you$1 ,000 off at vanta.com slash twist. That's vanta.com slash twist to collect$1 ,000 off your sock too. What is the bet? The bet is, do we see small models in Apple's phones or Android before, you know?
40:47Those are two very different bets. I know. Well, okay, we can refine it, but that's where it's going. I was like, and mobile devices, local models on local devices. pre-installed or installable let's say pre-installed okay so i think we need to break this into two yeah okay i will propose all right there is two flagships you have the pixel from google google fi google pixel which they're still producing and they will they put their latest and greatest on that and then there's the iphone so i think we're breaking into two bits yeah and am i supposed to go first yeah okay so i'm taking the over for the iphone okay and then should you make the bet on android to make it fair yeah or should i make the under on android yeah see it now so i think this is going to be a push this is a classic push yeah well because you know you know apple takes time to do stuff yeah and they're thoughtful and you know google just you know is getting better at throwing stuff out there so great i can give you just a little that's great bet i think this is the most interesting one yeah on you know again just before the holidays apple released uh like a version of a model called ferret and put it into you know hugging face and yeah so whoa whoa whoa whoa whoa breaking news apple has a language model this is a first it's it's not their own so what they've doing what they've done is they've been working with existing models and modifying the fact that they're putting this out there is wait wait on a github account I don't mind putting it out there.
42:15There's a GitHub account from Apple employees of something called Ferret. Ferret, yes. Refer and ground anything, anywhere at any granularity. Yes. Okay, so this is in order to do what? This is for images? It seems like they've done this for images. Exactly, right? And so they've done this on, and the reason they've probably been forced to open source it, and I don't know how much time you spend on open source, there's certain models, and certain open source licenses that if you use them to build with, you have to open source what you're doing. So what I am taking this as Apple is using this and they're kind of being forced to open source it because the license, it was just a good thing for open source.
42:55And so this tells us they must be close if they're already pushing it in the open source. So they are using an open source model. The open source license requires you to give to get. So if you want to build on an open source model, in some cases, the licenses mean you have to give back any changes you make. Correct. Fair is fair. So then Apple is putting this out there and this is to identify images and to make changes on images using AI. No, no. This particular model is like, you can see the example of what is the relationship between object and object? In this case, there's like a, you know, a dog and a ferret, right?
43:31And so it's basically, what it's going to allow you to do is like kind of maybe ask questions about stuff on your phone. So this is for Google. This is for Apple Photos. Apple Photos sucks. when you're like bulldog and it's like it gives you like a piece of pizza yeah so this they're obviously working on ai for google photos yes got it and instead of having everyone upload all their photos and do it in the cloud they're like we'll just do it on your device with all the compute that we have there okay that makes a lot of sense if you got an m3 if you eventually have oh no they're using the a chips on there i think they're up to a14 or 15 or something these are just kind of like tuned down versions of them chips nowadays yeah so i mean what an incredible thing steve jobs did in his final uh years to get off of intel and make his own silicon and what a tremendous advantage what a visionary thing to do what a great thing to do with your money yeah you know when you think about being an entrepreneur when you build up a chip stack you know what can you do with the chip stack uh you can't buy things anymore obviously adobe couldn't buy figma and they're trying to block every acquisition so you know what can you do with the money amazon built out data centers and warehouses uh apple decided to build their own silicon yep uh elon over at tesla built uh supercomputers and his own silicon and yeah with dojo so you know in deploying capital if you're not allowed to buy things then build things and build things that are unbelievably expensive and insane long passes i love this idea like if you're the new york times right and you're starting to build up the subscription business yep what could you do with the money well you can buy little things like wordle and then keep adding them to your subscription or buy the athletic a small purchase or buy wire cutter they should be doing you know and i don't know the new york times never offered to buy a gadget uh etc but let's just say they were dancing around and aol just had the cojones to come in quicker than the new york times but fox i never told anybody this but fox uh rupert murdoch uh put in a late bid to buy weblogs okay when you already tried when you had the aol deal on the in the red zone and they were like hey how about plus three million and i was just like how about no like and i just told them like if you're going to give an offer that's 30 more than what they're offering which would have been nine million dollars and i'll consider it but if you're just gonna we're not playing the prices right here sorry respectfully rupert yeah you sold early on that one jacal wow 18 months from zero to 30 million yeah wow dude i needed to secure the bag i mean that was that was the setup of all setups once you get your first 10 milli yeah you're dangerous that's i always tell people get that first 10 million in the bank and man just to think how it's been it's all good though all good well no i mean yeah but i mean in gadget right now if i owned in gadget it'd probably be doing 50 million in revenue yeah whatever it is 15 20 years later almost like yeah that's not actually that big of a business it was a great business don't get me wrong yeah but you know i i'm building something with inside.com that i showed you briefly that i think is going to be the best editorial product to ever make awesome um and that's my 2020 that's one of my 2024 resolution and you're bringing ai into it so it's kind of it's an ai based community news kind of product that i'm i'm grinding on and i just got a very talented person to work with me on it so i need a collaborator i need like a sandeep you know a technical collaborator because i'm not in the weeds but i think i might make something that's very interesting you know and i've always had this obsession with the inside.com domain name and then i think i finally have the concept yeah well with ai it fits really well because you want to be i think so yeah based and the stuff I've shown you, I think we might have something.
47:01We might have a winner here. Let's finalize this bet. So we said - I got the over. Yeah. Apple, you got the under - Under Google. Google Pixel. Yes. And what we've seen - That they bundle the model on the phone. On device. Yes. On device. If they announce it, does that count? If they announce they're bundling, I think it should. I think it counts. Yeah. Yeah, announced or executed. Either one of those. So if they announce they're going to have this model - Apple always has this weird six, nine month window sometimes between announcements and releases. wait aren't the vision supposed to be on sale yet then they say in the new year they were going to have the the goggles on sale to be honest that's one thing i'm just not tracking very well nobody's tracking and it's like isn't an amazing yeah it's really great ai is got so many practical consumer applications yeah that ar vr yeah and crypto seem unimportant by comparison i will say and i i didn't pull it up here and i i want to do it maybe next week but the new ray-bans from meta have been coupled with the language model that they have and so you can wear them you can turn on record mode and you can say like hey what am i looking at and it will explain it to you so that's what i've always felt would be the best use of those hands down is to what's in the world right so i felt like if i'm looking out at a a building if it told me hey in that you know like when you're in japan yeah uh they'll have very tall buildings and the buildings will um uh when you're when you're looking at the building there might be a 10-story building and there's three restaurants on each floor yeah and there's all little signs and it's very hard to navigate and you have to go to the directory but if i just looked at it and it said and i said just show me the four and a have five-star yelp review google reviews yeah and i just looked around on the block and it just lit them up man that would be powerful or what about you were wearing them at ces this year and you're like you know ah yes show me the give me information on this information about this or just you know summarize what i just looked at here and book markets imagine real world bookmarks i'm just looking at something and i'm like hey uh bookmark that and send it to sunday that i look at a dish in a restaurant i'm always i use food as the use case because we eat food all the time but imagine i'm looking at that french toast at the central bakery in japan yeah and i just looked at and i said send this to sunny and it just sends you a video in the link man the bookmarking shared bookmarks on ar are going to be sick yeah that's going to be the number one use case where you're like oh what was the name that place and it pulls it back out and you know french toast let's go yeah all right here we go okay so last one last bet the last this is going to wind up being six bets because we have the double bet yeah so you know it was supposed to be five but we turned that into double but whatever we'll throw another throw another dime out there now wait are we on the last one are we 500 each or a thousand each no just do a thousand it's like you know makes it easier so it's six thousand at stake okay six thousand at stake now all right like it yeah so the last one is um and um we have some thing i can pull up but i want to just lay it out first last year was a year of like large language models text and what we saw at the end of the year is multimodal and you've started doing this with talking um you know we had a poker game recently and the settlement i don't know and share this but the settlement was done by gpt you know you took the screenshot of the notes and did the settlement with chat gpt and so um and you know like typically we have someone do that and all that so like the the bet here is over under by july 1st do multimodal models become multimodal does an open source multimodal model become the most popular model on the planet does an open source multimodal yes become the most popular on planet and so this would be you're basically saying would it be better than chat gpt's app yes that's basically what you're saying because that's the number one and we assume it's not going to get displaced so will chat gpt4's multimodal app be displaced yes by an open source model by july 1st do you have anything to set this up or not are you going to show something to set this up i i am i'm going to show this uh so okay let's show it now we we're not saying by number of downloads or paid subscribers it's just a better product qualitatively you know like how everyone talked about chat gpt and gpt4 through all of 2023 everyone starts to talk about um everyone begin to talk about um okay okay so i think the way to say this is will an open source model displace multi-modal models very specifically but yes okay so multi-modal i uh and is it my bet i think you did the android one last so it's my bet okay yes your bet go for it my bet um i'm gonna take the over i'm gonna share my screen here i'm taking the over here because i just think open source takes a long time to bake and get stuff out okay but maybe so this is a um a project called lava uh large language and vision assistant and it's a combination microsoft research columbia and i remember this uh lava wisconsin L-A-V-A, Large Language and Vision Assistant Visual Instruction Tuning.
52:18Yes. Got it. Wow, this is impressive. Okay, wait, I may have to change my bet. Okay, hold on. Based on this. You didn't, yeah, you got, you had something here. Okay, hold on. So here's a picture of a guy ironing a shirt on the back of a taxi? Yeah. And the question that I, you know, this is in the demo. It says, what is unusual about this? And it says, the unusual aspect of this image is that a man is ironing clothes while standing on the back of a moving car. this is not a typical scene as ironing clothes, yada, yada, yada. Right. It's done in a more indoor environment, not. And so the person can't balance and it'd be really hard to do it.
52:49How does the model know that this is what's the old J. Cal, like you took the, how does the model know? I mean, cause you know, I understand how it regurgitates, like what's the best coffee machine to the New York times versus opening. I know I get that. I get when it makes an image and it's like, Hey, make me an image of a happy new year image. Yeah. and it just goes finds other happy new year images and and you know builds off of it but what really trips me out about language models yeah is when they can interpret an image that always freaks me out this is the part where i start to think it's thinking well here's here's unless it knows this image does it know this image and somebody described it on the internet somewhere no it's actually figuring this out from the context that's what freaks me out about And so the way to think about this is, and this is, there's a lot of, there's a lot of energy around this particular topic.
53:43People believe that LLMs are actually not, should not be used as knowledge retrieval systems, which is what they, people use them for. Like, Hey, you know, tell me where, you know, exactly. But more their reasoning engines and because of their understanding of the language. And so in order, what is reasoning, right? Reasoning is like an understanding of language and they have an incredible understanding of language. And so if you think about what happens when this image is presented to a model, it basically, the first thing that happens, the model has to describe what the image is. And it'll say, well, it's a car and there's a person ironing on the back of it, just a description.
Read the full transcript
54:20And then the reasoning capabilities of the model say, hey, that's just not practical. Like it's, you know, because of what it's saying here, because it's able to reason that from its understanding of every, all the training material it's ever seen. So there's some query going on that says… Not a query, per se. Yeah. It knows scenarios about driving. It knows scenarios about ironing. And then it's able to kind of infer from that. You look and it tells it. It's a car is moving. You want to be balanced if you're doing this. So it's, you know… So it knows that this task requires balance and it's done indoors and that this is outdoors.
54:55So it has in its mind… It knows that it's all that in training material when it learns about ironing, right? And it's like, hey, you want to have a stable surface. you want to make sure it's not tilted you know like uh you want to make sure the stable the surface is very flat like don't you wouldn't do it and it also knows that you it's dangerous to be in the middle of the road yeah yes yeah so this is i mean i really think the reasoning part of ai is the part yeah that i find so impressive i do think it's a bit of a parlor trick when it rewrites stuff because i you know it's very easy to understand that part the reasoning part is where I always get impressed.
55:29I have to be honest. Yeah. And at the all in summit, I spend time with Steven Wolfram and he's really big on the reasoning aspect of it as well. So here we go. You're showing an image. What are the things I should be cautious about when I visit here? And it's showing a lake with a dock that walks out, you know, it's like a pier. Yeah. When visiting the pier over the lake, there are a few things you should be cautious about. First, ensure that you have appropriate footwear with good traction as the pier may be wet or slippery, especially if it has been raining or if there is algae on the surface well it's a deep pull second be mindful of the weather conditions the image shows a cloudy day oh here we go which could lead to sudden changes in weather such as rain or strong winds it's essentially be prepared for these conditions and to have a plan in case of an emergency lastly be aware of any potential hazards in the water such as submerged rocks or debris and avoid swimming or wading lakes without proper safety measures so it knows it's a lake it knows it's a dock it knows that there are since you said what should i be cautious about it does a reverse and says we're dangerous what's dangerous on docks and lakes and yes and then it goes into to this one i get the reasoning a little bit better when you do something completely like the ones that always get me is when you put a meme in there and you say explain the mean yeah they do a great job most of the time or sometimes it gets it nails it so i find that also very fascinating this has been an incredible year here um we didn't vote on this one yet it's my vote i'm taking the over i'm taking the over i'm taking the under like we saw that okay all right there we go all right everybody on the page it'll be up this week in startups.com slash bets is going to redirect to a bets page if you remember any of the long bets we've been making over the last decade here on the show or if you want to go search using our new ai website the new ai website powered by podcast ai is thisweekinstartups.com all the ai episodes are available at thisweekinstartups.com slash ai the bets will be available at thisweekinstartups.com slash bets startup basics is at thisweekinstartups.com slash basics that's kind of how we do things here because we just put a slash and a simple keyword uh sandeep can be found on x.com slash sandeep yes two e's and a p sandeep uh we call him sunny he's awesome i don't think you have any room for more customers or clients for definitive intelligence you're all we always do oh no you have room for one more okay so there's one more client engagement for 2024 available uh definitive.io yes is that the url or sunday that definitive i uh dot i was your email i believe and uh listen we're out of time but on the new york times thing uh it's going to be an incredible lawsuit most of these things get settled they tried to settle this one already but i think there is a i think this is landmark it's sort of like what happened in the internet at the beginning you know with similar um yeah my advice to new york times is to not back down on this one and go to the mat on it that's my best advice to them because this is the Google crucible moment.
58:32If all the publishers had gotten together and said to Google, if you want to index us, and you want that snippet there, because when you look at 10 snippets, you might get 50 % of the value of clicking. And that's what a lot of us do. You do some search on Google, you get the one box, you get the snippets. And then they went and they boiled the frog basically. But if the top 500 publishers had said, Okay, we'll no index, Google would have been worthless. Google would have been worthless. Let's be honest, it would have never found all the interesting information. I think New York Times should then team up with Reddit, Quora, Stack Overflow, and a bunch of other people.
59:05And they should all sue OpenAI, and all put these examples out there. And they should just target OpenAI until OpenAI comes up with a licensing model that makes sense. And I think the licensing model should be reasonable, so that it encourages people to do it, but it should be people's right to opt out of it, and they should be able to set their own price. So I believe what should happen is if you want to license the New York Times, you should pay a fee of$1 per year per user of your service. $1 per year per user of your service. If OpenAI gets 100 million users a year, they should pay$100 million to have the New York Times per year forever.
59:48That's my belief. Okay, and they make$240 per user. So they would be making$24 billion, right? And so they would just have to fork out, you know,$20 ,30 billion. A very small amount. Yeah. If they're making, you said they're making$20 per month or$240 per year? Yeah,$240 per year. So they're paying less than 1 % to the New York Times, one of the top websites. So they're paying 40 bps to them. Yeah. And then let's say they paid Reddit and Quora 10 bps, 15 bps. And let's say 30 % of OpenAI's revenue went to licensing deals. That seems pretty fair to me. Yeah. If 30 % of Google's revenue went to publishers right now, hundreds of billions of dollars revenue, we wouldn't have the problem with publishers being underfunded right now.
1:00:33So I think I'm going to give it the, this is my best JCal judgment. It's a JCal judgment call. Okay. 70-30 split. The language models keep 70, 30 % goes to who they trained on. And you got to show up, you got to have a license. you got to have newyorktimes.com slash license.txt and the license.txt should if you make it available then you should have in there what it costs and you say it's a dollar per user per year forever and it don't do a one-time fee i think the one-time fee is the stupid stupidest thing you could ever do it should be a yearly licensing fee if you want my book in your index angel i should get paid i don't know what's a reasonable amount 25 000 a year what about all the content of twist yeah okay so if listen if they said if you if youtube wants to train their model on twist 2 000 episodes i would say i don't know a hundred dollars an episode per year ten dollars an episode i would take 25 000 to 200 000 because once it's done it once you don't need to go backwards right no but i want to because just like artists should get paid residuals for their music so that it encourages more people to keep making music so that they can make music for the rest of the lifetime because youtube and open ai are going to be making money on it for those years for those years so if they make money on it for those years why should not yeah okay so that's what this is where fairness comes into play for artists this is my belief yeah i get royalties on my books uh i can have continuing revenue from this week in startups if the i can take my stuff off of youtube anytime and i can just let people go direct if youtube wants my stuff if they want to do a language model on my stuff if they want to index it it should just be they should cut a deal with me and people are like oh that's not possible you know what open ai cuts a deal with every employee yeah right google has tens of thousands of employees they cut a deal with every employee doesn't mean it doesn't have to be standard but they can do it they can do it And, you know, maybe if you're in the long tail, if you have under X number of pieces of content, maybe you just have to take the standard deal.
1:02:41But maybe when you're over a thousand podcast episodes or over 10 million listeners, then you have the ability to negotiate. So, again, I would just put it in my license.txt. Here's my license terms. I like your approach. I'm in like a different technological headspace on it. But going back to this reasoning thing, I think we are at a place now where we have models that can reason and we don't need to use them for information retrieval. So, you know, if you're trying to get the best review of an espresso maker, you can basically have the model reason to go out to something I have access to and pay for it and go get it.
1:03:18So if I'm already subscribed to New York Times, it should be able to use my subscription to go get that from Wirecutter. A hundred percent. Right. And New York times will provide that. And so I don't provide that. Yeah, exactly. And I don't think models need that information anymore. I don't think models get any better by feeding them all the wire cutter information. And so he adjusted it. Well, I think even if you took it out and retrain the model, I think we're at his place now where their reasoning capabilities are so good that if you going back to that image, right. That we just showed with the multimodal model, I do think a different business model will emerge, will emerge which is more like connect it to things i have access to rather than that licensing approach i think that's what'll end up happening i think for people who have large amounts of data and that data is continuing to grow because the new york times said like their goal is to like increase the amount of content they have by 50 like the number of articles by 50 in the next four years or something wow okay you you probably want to have access to those next 50 you know percent increase i don't think it is freshest information what if the advice on cancer treatments what if the political landscape and history of ukraine ukraine ukraine changes you know and you want to have that information in your in your model like and it's in the new york times and that's the freshest data it might be worth it to you to give them 100 million a year to have that new access yeah like you may want your model to have that when you comment on ukraine you know is david sachs right about ukraine or not and like yeah well we'll find out in the next five years if who was right biden nato sacks it's it's so it's a real intersection between technology and publishing and then human knowledge and like how does that all work together it's a really and make being respectful of content creators is the best model for technologists we have not always been and i sit right at the i sit at the crossroads of both yeah i want to see the most powerful ai and i want to see humans who create great content be compensated for it in the long term yeah and it's so unfair um and this is the great thing about the new york times lawsuit and i applaud the new york times for filing it in such a deliberate way because what they've done is they've proven beyond a shadow of a doubt that these things are getting massive value and taking value away from the new york times franchise because the new york times is behind a paywall they require subscriptions open i require subscriptions and so i think actually as a human being you will be making your decision to get this information is it better to use the open i interface or the new york times interface and i find myself doing searches about i literally did one on coffee grinders i was looking for a coffee grinder i did it on chat gpt and i got everything i wanted then i went to wire cutter and i was like oh wait a second this looks familiar and it was like somebody had copied their homework you know when you take the two multiple choice tests and you hold it up to the light and you can see like oh my god you picked all the wrong answers as well as the correct ones yeah yeah and i bought the oxo 99 or whatever it was great coffee grinder i love it it's great for drip coffee it may not be what espresso people want but i'm paying for both subscriptions i think people will pick one yeah and i think they proved it perfectly and then i don't know if you saw over the weekend there was a person who did what i did with the jedis remember i did the jedi bulldogs somebody said make me two italian cartoon characters in a race or whatever and it made luigi and it made mario yeah and then they did it again like a couple of days later yeah and it wouldn't let them make mario and luigi anymore yeah and so open ai is going through right now and testing and removing characters they know they don't have the ip for because they know they're going to get smashed with a disney lawsuit and a nintendo lawsuit any day it is nintendo's right to make an ai generative ai where you can make the characters and make birthday cards out of them or you know tweets out of them that's their ip they have the right to make those derivative products not dolly chat gpt 20 a month yeah well the other ones are doing something similar i read something somewhere about mid journey where people were create recreate in the latest version they were recreating like you know marvel superheroes and stuff and so of course and you know what i want to do that yeah but if i'm going to do that i kind of feel like it should be part of my disney plus subscription i shouldn't be doing it for free yeah now i know everybody wants everything for free but why would stan lee create the next marvel characters if there's no residuals or no rights to it everybody can just have a free-for-all why should open ai or mid journey or stable diffusion get to make marvel characters for my birthday party invite you know if i want to do it an avengers birthday party invite i should be paying a license to or it should be part of my disney plus subscription that's only fair it should be part of my open ai subscription yeah yeah 100 now if you wanted to make a parody parody is allowed but if you are providing a service to create parody that's not allowed so if i made a parody so you can go and create memes well if i made memes on my own using a local version and i'm not trying to make money from it it's under parody it's fair use okay if i make a tool to make parodies and i charge for that tool then i've like a meme generator you're in then if i'm the meme generator now i'm in trouble so this is what people don't understand about copyright law it's so nuanced that people are like well why can't i could draw it myself it's like yes but if you started charging for drawing it now you start to get into a different place so if i did um if i started charging people on the web a hundred dollars to make versions of iron man or versions of jedi yep uh and i named them and i put star wars on it that's not i'm not allowed to do that right and here's to mario brothers dolly doesn't need so this guy gary marcus is just doing what i did yeah like a month ago right he copied my thing here but he said video game italian and it made mario brothers of course it made mario we're getting into the place where these things are going to need to block certain characters and not have the output as well.
1:09:21So I think and I think that's right. I think that's the right thing to do. And I'll be honest, I said something in my newsletter, I think I may have to do a clarification on I said something to the effect when I wrote about this, the day it happened, that a bunch of people that opening I used the stolen content to make a bunch of money in the secondary offering. Yeah, I actually don't know. I was kind of reading their minds. I don't know that each individual opening person is acting that way. So I take that back, I'm going to clarify it, that sentence, because I don't want to mind read them or their intentions.
1:09:49They might be just doing AI because they think it's good for society, or it might just be a job and they don't feel particularly either way about it. But the fact is, it's unfair that open AI is selling billions of dollars based upon of shares. Yeah. They might sell more shares in their secondary than the entire value of the New York Times. if they sell 8 billion in secondary which is completely possible over the coming years they sell eight because a company let's say the employees own 30 percent of the company is 30 billion dollars new york times worth 8 billion so when those employees liquidate they're going to get four times the value of the new york times and they built it on the new york times content that's unfair and that's where they're going to get so demolished in court for damages the damages on this could be unbelievably high because so much commerce if open ai wasn't so successful and it was a non-profit it would be so hard to do damages but once the employees are selling billions of dollars in shares and they have a hundred million paid subscribers or whatever they have that's when it becomes profoundly unfair i just did thanos as a farmer in mid journey yeah that should be done on disney plus that should be a feature of disney plus subscriptions make your own characters on disney plus and then it puts a logo there made by disney plus there's going to be a language model made by new york times and this is my thesis what if the new york times competes with open ai heads up they take a language model that was built uh on the new york times content and then they have a model where having access is pretty dope and it increases the number of new york times subscribers and then they go buy engadget from yahoo they go buy tech of they go by ready well they would just take one of the big open source models and then to fine-tune it with their own data or hopefully that model wouldn't be built on stolen content either so they'd have to find a clean model on the open source models we can see where all the training data came from so if they did that and then they made the new york times model as long as they don't have to give to get like we talked about earlier the program the new york times is a competitor to chat gpt and that's what the courts will realize and i am the tip of the spear i pay $300 to both companies.
1:12:01I will not pay$300 to both companies forever. I will pick one. And if it goes the way ChatGPT is going, I'll probably cancel my New York Times because I'll get all the value from that. Yeah. Yeah. I mean, Apple does this really nicely with the news, Apple News Plus thing. Yeah, they pay. You have to have, and I have, what's it called? Apple One or something for my family. That has 40 bucks a month. Exactly. I spend$500 a year with Apple. They made it much more expensive. It's like, it's more expensive. No, mine's$38 a month. So it's$500 a year, basically. I get Apple Arcade, Apple Music, Apple News, everything, Apple TV Plus for my whole family and two terabytes of storage.
1:12:41It's the greatest deal ever. It's$100 per family member. It's great. Yeah. And it's simple. It's phenomenal. So this is what ChatGPT should do. They should have part of your subscription be for one. Yeah, Apple did it right because they respected that. Apple did it right. Yeah. Now imagine if ChatGPT said, include New York Times for$2 more a month. include disney characters for five dollars a month for ten dollars a year one dollar a month that could be a great way for them to make cut commercial deals with people and they're clearly they want a deal i think you know if if this is my challenge to sam waltman if you're such a great deal maker you know people are saying it's such a great deal maker well then make a deal let's make a deal for content creators make a deal you know like make a revenue model that's sustainable all right this has been another amazing episode we went a little overtime uh happy new year again and we'll talk soon bye bye
From the publisher
This Week in Startups is brought to you by…
Miro. Working remotely doesn’t mean you need to feel disconnected from your team. Miro is an online whiteboard that brings teams together - anytime, anywhere. Go to https://miro.com/startups to sign up for a FREE account with unlimited team members.
NetSuite. Once your business gets to a certain size the cracks start to emerge. Things you used to do in a day take a week. You deserve a customized solution - and that's NetSuite. Learn more when you download NetSuite’s popular KPI Checklist - absolutely free, at NetSuite.com/twist
Vanta. Compliance and security shouldn't be a deal-breaker for startups to win new business. Vanta makes it easy for companies to get a SOC 2 report fast. TWiST listeners can get $1,000 off for a limited time at vanta.com/twist
*
Today’s show:
Jason and Sunny make a bunch of predictions and bet $1000 on each for the upcoming year in AI.
*
Timestamps:
(0:00) Sunny Madra joins Jason to make some 2024 AI bets
(4:46) Bet 1: There will be a top 100 song created by AI
(10:22) Power dynamics in the music industry and its intersection with AI
(13:01) Miro - Sign up for a free account at https://miro.com/startups
(14:34) Bet 2: AI will be capable of replicating Jason doing a TWiST Segment without listeners noticing
(18:51) Bet 3: There will be an AI-generated Pixar short that has made it into mainstream
(27:46) NetSuite - Download your free KPI Checklist at http://netsuite.com/twist
(29:06) Bets 4 and 5: We see a small model deployed locally on an iPhone/Android
(39:31) Vanta - Get $1000 off your SOC 2 at https://vanta.com/twist
(49:34) Bet 6: A multimodal model will be #1 by usage, AND open-source multimodal will beat out closed models
(52:00) LLaVA: Large Language and Vision Assistant
(58:04) The potential implications of the New York Times lawsuit
(1:10:01) OpenAI's financial success, copyright lawsuits, and the potential of Disney and the New York Times in AI
*
LINKS:
Midjourney: https://www.tiktok.com/@controllaxyz/video/7319184981987839263?_r=1&_t=8igobwQ1dzq
AI Drake: https://www.youtube.com/watch?v=NhVuQ06tO3M
AI Spongebob: https://twitter.com/Teridax/status/1741205766046888149
AI Pixar: https://twitter.com/eyishazyer/status/1741418522448986492
TinyLlama: https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0
Llava: https://llava-vl.github.io/
Suno AI: https://www.suno.ai/
*
Follow Sunny
X: https://twitter.com/sundeep Check out Definitive Intelligence: https://www.definitive.io/
*
Follow Jason:
Instagram: https://www.instagram.com/jason
LinkedIn: https://www.linkedin.com/in/jasoncalacanis
*
Great 2023 interviews: Steve Huffman, Brian Chesky, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland
*
Check out Jason’s suite of newsletters: https://substack.com/@calacanis
*
Follow TWiST:
Substack: https://twistartups.substack.com
Twitter: https://twitter.com/TWiStartups
YouTube: https://www.youtube.com/thisweekin
*
Subscribe to the Founder University Podcast: https://www.founder.university/podcast




