Latest / Newcomer Pod / Inside the AI Gold Rush (One Year Later)
Transcript
- 0:00Artificial intelligence. AI AI AI AI AI AI AI.
- 0:04AI 2025 has been another crazy year for artificial
- 0:08intelligence. We've seen the rise of thinking
- 0:10models. We've seen video creation,
- 0:13including Open AI's fictionalized version of it.
- 0:16We are about to host the Cerebral Valley AI Summit, our
- 0:20ultimate insider AI event for startup founders and investors.
- 0:25I host that event with Max Child and James Wilsterman, my dear
- 0:28friends and the cofounders of the voice AI games company
- 0:31Volley. They're coming on the show to
- 0:33pick up where we left off on our AI startup draft.
- 0:37We've been picking the biggest names in AI startups, and it's a
- 0:41good way to learn about the top Silicon Valley startups that
- 0:44insiders are paying attention to.
- 0:45So we're going to have a fun time picking up new companies
- 0:48and dropping them from our list. But first we will discuss some
- 0:51of the key themes in AI going into Seruville Valley Volley
- 0:55thinking through what's really happened over the last 12
- 0:58months. So give it a listen.
- 1:07All right, let's get into it. We've got the Srebul Valley AI
- 1:12Summit coming up November 12th. I'm here with Maxwell and James
- 1:16Wilsterman, my Srebul Valley Co host, good friends and the Co
- 1:20founders of the voice AI games company Volley.
- 1:25We have two main objectives for this episode 1 themes, which is
- 1:30half preparation for the event for ourselves and half just sort
- 1:33of like, OK, where where are we in sort of the zeitgeist of AI
- 1:37going into this event? And then we'll see what comes
- 1:41out of it afterwards. And then to our favorite part of
- 1:45the show, we're going to update our our drafts in a second.
- 1:49So hey guys, welcome to the show.
- 1:51Hello, Eric. Glad to be here.
- 1:52Hey, Eric, hope you're doing well.
- 1:55Yeah, I've got a screaming child.
- 1:58We're a day. We're 28.
- 1:59Days the draft gets. Heated you get in.
- 2:02In trouble over there. Yeah, keep it down.
- 2:04No, I have enough distance that I think I'm far away, but a
- 2:07screaming child can can carry quite far.
- 2:11Great. And you guys have been dialed in
- 2:13thinking about sort of the big theme.
- 2:16So let's let's get into it. Yeah, I think the goal here is
- 2:21everybody come with at least two, if not sort of three themes
- 2:24that you think will stand out in AI generative AII mean we do
- 2:30Cerebral Valley twice a year. We did London mid year, we do
- 2:34San Francisco in November. So these are themes, you know,
- 2:38meant to, I think look at like a six month window or sort of
- 2:41what's, what's the mood for the next 6 months in AI Max?
- 2:46I know you've thought about this a lot you.
- 2:47Want to go first. I mean, I have a bit, I have a
- 2:49bit of a look back and a look forward here in terms of 2025
- 2:53and beyond. I think that what's interesting
- 2:55about this year is there were a lot of kind of industry
- 2:58prognosticators trying to call certain themes for 2025 back in
- 3:02like January, February, March. I remember distinctly I had some
- 3:07very intimate conversations with folks who worked at Open AI.
- 3:10Who? Intimate.
- 3:11Intimate they. Were intimate.
- 3:13You're like in Asano. Yeah, yeah.
- 3:16Yeah, Yeah. Let me tell you.
- 3:19The Gary Chan sauna. A couple cocktails, sauna
- 3:23action. It was a great time.
- 3:26I won't say whose house it was at anyway.
- 3:29Yeah. And they basically were like,
- 3:31there's going to be two big themes for 2025.
- 3:34And I don't know if you guys want to guess these in advance.
- 3:37They were like the two big themes are 2025 is the year of
- 3:40agents and 2025 is the year of voice.
- 3:45And my contention would be that both of these themes turned out
- 3:49to be bullshit right too early. I mean, like, I think that we
- 3:54talked about this a year ago on the podcast.
- 3:56I don't even think we'd seen a quote UN quote agent at that
- 3:58point. And I think at this point you
- 4:00could argue maybe we're starting to see the contours of what
- 4:03people would define as an agent, but it's almost exclusively
- 4:05encoding and then voice. Look, James and I are obviously
- 4:10voice bandwagon ears. You know, we may have hand
- 4:13constructed the voice bandwagon out of Wood and Nails about 8
- 4:17years ago, but I don't really think 22 in the voice either.
- 4:21I think there's been progression in voice recognition in TTS, 11
- 4:25labs, evaluations, things like that.
- 4:27But I certainly wouldn't say we're all spending every day
- 4:29talking of voice AI and even the products that we use I don't
- 4:33think have made maybe as much progress as we would have
- 4:35expected. So that was my initial framing
- 4:38for this, which is those were the two expected themes for.
- 4:40The what didn't happen? Yeah, it didn't happen, James.
- 4:42Your reaction to the ones. That did happen, but yeah, go
- 4:44ahead. Yeah.
- 4:45I mean, I I agree, I think, I think that a lot of companies
- 4:49built out foundations for agents.
- 4:53I think Open AI just came out with their kind of agent's drag
- 4:58and drop framework and there have been a lot of other
- 5:01companies entering that space. There's like an agent's SDK from
- 5:06Open AI and there's other companies that have these now.
- 5:09But I agree, I haven't seen like deployment or scale of these in
- 5:16any meaningful way. I, I think the fact that we're
- 5:20talking, that people talk about agents is itself a sign that
- 5:24things aren't where they need to be, right?
- 5:26Like coding customer support, like if you're succeeding, you
- 5:29just say, oh, there's this great use case.
- 5:31And like, I, I feel like you don't start talking about SAS
- 5:34and software as a service until it's like, OK, you've got some
- 5:37products people can point to. It's like you vertical SAS
- 5:40becomes a thing because you've got like mind body and you say,
- 5:43oh, you can, you know, provide for this sector, We'll do it for
- 5:47everybody else. But I feel like with agents and
- 5:50you know, this isn't the first time this ever happened in
- 5:52technology, but people are like, you know, sort of in concluding
- 5:56how all this can play out without sort of the first really
- 5:59successful use of agents. Because we say coding, but even
- 6:04with coding, it's like a lot of it is a sidekick, right?
- 6:07And not necessarily like, you know, Devon running, running on
- 6:11its own. Yeah.
- 6:13I mean, I think that Andre Carpathy obviously had a very
- 6:16notable podcast a couple weeks ago on Dwarkish podcast where he
- 6:19talked about development and agents.
- 6:20And I think the analogy he used was to self driving cars, which
- 6:24I thought was a good one where but more like self driving cars
- 6:26back in 2016 or 2017 where he's saying, OK, if I think of an
- 6:30agent as essentially a smart intern who can go off and do
- 6:33basically any sort of task that I assigned them.
- 6:36Like, you know, it can do a little bit of stuff and you give
- 6:39it a very specific use case and a very focused set of, you know,
- 6:43check boxes it needs to check off.
- 6:44It can occasionally go off and do interesting things.
- 6:47But you know, his point was what happened in self driving cars is
- 6:50we, we, we ended up having to go through what he called like the
- 6:53March of the March of nines, the March of nines being what
- 6:57percentage accurate is this thing right?
- 6:59And we started with 90% and then we got to 99% and then we got to
- 7:0399.9% and then 99.99% and so on and so forth.
- 7:06And ultimately to be superior to a human, you know, Waymo, you
- 7:09maybe need 5 nines or six nines or 7 nines or whatever, right?
- 7:13And he's saying, hey, agents are maybe at the first nine, you
- 7:16know, today, maybe this, you know, you could argue the second
- 7:19time, but I don't even think you could argue that really.
- 7:21And we're probably still 234 more kind of nines away and and
- 7:25maybe things are getting 10 times better every year.
- 7:27So that's only three or four or five years, but he put more like
- 7:30a 10 year timeline on sort of fully, you know, human capable
- 7:34level agents. Why do you think companies like
- 7:37Open AI are moving so fast in this direction?
- 7:43Building out the SDK, is that the the no code interfaces the
- 7:48browser agents within their, their their Atlas browser.
- 7:53I guess like why? What is it marketing or is is
- 7:57there a belief that this is further alignment?
- 7:59It is or or you know, a different take on how the
- 8:03timelines might look. I mean, my argument would be
- 8:06it's competitive defense, basically.
- 8:07Like there's people doing these little agentic drag and drop
- 8:10frameworks and they're like, we better do that.
- 8:12And then there's people doing, you know, browsers and they're
- 8:14like, maybe the browsers will be a thing.
- 8:16I guess we better fucking do a browser.
- 8:17You know, like it's like, that's my argument.
- 8:20I could be wrong. You know, obviously they're much
- 8:22smarter than we are. They're smarter than I am.
- 8:24Like they figured out a lot of stuff that we would never
- 8:26figured out. So maybe they see the future,
- 8:28they see the 2-3 year agentic vision and it's it's, you know,
- 8:31they're building towards that right now.
- 8:33I think. Yeah, maybe the more charitable
- 8:34interpretation is they can see where this thing's going to be
- 8:37in three or five years and that we're putting all the
- 8:39foundations in place today, but it doesn't seem like they work
- 8:43that well. Like I have not heard of anyone
- 8:44using this agent drag and drop composer thing for to build an
- 8:48agent in the last four weeks since they launched it.
- 8:50Have you? I mean, I'm I'm it like
- 8:52basically vanished without a whisper in in my my social
- 8:55feeds, I guess. I'm going to offer a theme, I
- 8:59guess in it feels like it it follows the same pitfalls of the
- 9:04ones that you said didn't come true.
- 9:06It's sort of like it's you know, it's going to be voice, it's
- 9:09going to be agents. And I'm going to say like
- 9:13fiction frenzy, you know, it's with the Sora creation and just,
- 9:17you know, and, and also speaking to your 999 issue, it's like,
- 9:21OK, what doesn't require stuff that's like, you know, totally
- 9:27accurate works well. Like, you know, the problem with
- 9:29agents is compounding. It's like you, you make a
- 9:31mistake here. You have even more mistakes down
- 9:33the line. So just just the, the errors are
- 9:36are huge. Where stuff where it's like
- 9:38create, you know, regular people create content, fictionalized
- 9:43stuff. I think, I think that's has a
- 9:47lot of promise. And so, and then everybody
- 9:49copies open AI and so they have their Sora app.
- 9:52It got a lot of downloads and it's like, oh, that can work.
- 9:55You know, we should, we should lean into that.
- 9:58And the last thing I'll offer on this theme, maybe it's an overly
- 10:01expansive view, but I'm trying to also capture the clearly
- 10:05serious like mental health thing and the sort of like people
- 10:09spinning up their own worlds and how that's going to loom in the
- 10:13AI story too. It's like you've got, it's sort
- 10:15of, this is sort of my consumer, consumers going wild with what
- 10:19is possible with AI and who cares how accurate it is at the
- 10:23moment. Why is that a tale of things
- 10:26that didn't work out? Or how are you thinking?
- 10:29Of that just because it's a very like oh, we're going to build a
- 10:31sore app or you know, it's like you see a glimmer of something
- 10:34that's like you see coding and so we're all going to have
- 10:37agents in every category you see you know, maybe a little funding
- 10:40going to voice. I just, I just think it's the
- 10:42sort of like you got, you got a sign of some success and then
- 10:46the AR world is desperate to see more.
- 10:49And so you're saying, oh, there's going to be so much much
- 10:51building there. So I think it's just reading a
- 10:54little bit too much maybe from one area of success.
- 10:58But I mean, you guys are in the you're trying to do fiction AI
- 11:01stuff. It's your.
- 11:02Yeah, it's possible. Yeah.
- 11:03I mean, I was gonna sort of say, I think that AI video is the one
- 11:06that kind of came out of nowhere this year as like an actual
- 11:09theme of the year in some ways. Whether or not you think it's
- 11:11good enough yet, I think that's an interesting argument.
- 11:14But I think that VO3 in May and then SORA a month ago definitely
- 11:20like sort of felt like minor nuclear explosions in this sort
- 11:24of consumer scene, right? Especially the especially the
- 11:27Sora era, where you could RIP off classic copyrighted IPS.
- 11:32Yeah, for like a week or five days or whatever it was.
- 11:35Make a SpongeBob episode of yourself and any character from
- 11:40history you wanted. And I just felt like that those
- 11:44seem to achieve actual consumer breakouts in a meaningful way.
- 11:47And and they fell off a big Cliff, I think when they pulled
- 11:50all the copyrighted material out.
- 11:53But I do think that AI video is something that has kind of
- 11:56crossed the chasm into actually useful now, whereas maybe a year
- 12:00ago it was sort of more of a curiosity.
- 12:02I think again, you can make the case that it's not a part of our
- 12:05everyday lives or whatever, but it feels like it made a big
- 12:07progress. And then I do think the AI
- 12:11psychosis theme, which is a little more of a negative theme,
- 12:14is a real 1 and probably a big part of the future of our
- 12:17lifetimes is this sort of AIAI induced psychosis and AI induced
- 12:22dream worlds and virtual friends and can they drive people off
- 12:26the ledge and all this kind of stuff.
- 12:27So kind of a sad one, but probably a very real part of the
- 12:32future. And do you guys think that based
- 12:36on what you're seeing that is happening at large scale like
- 12:39the AI itself is the cause of psychosis or do you think?
- 12:44Well, wasn't Sam Altman? Seems like he's owning this
- 12:46issue A. Little bit, yeah.
- 12:49Yeah. I mean, I think he was saying
- 12:50that very recently. Yeah.
- 12:52I mean, I was trying to read what he was saying, and a lot of
- 12:56it was kind of couched in this idea that they have massive
- 13:01scale. They have 800, right?
- 13:04At scale, all humans problem. Human problems are yeah, open AI
- 13:08problems like. What percentage of non AI
- 13:12conversations every day show signs of psychosis in the world?
- 13:15Right? Right, A lot.
- 13:17Exactly. It's just, yeah, we just have an
- 13:19active psych exam. Yeah, now we have someone
- 13:22running. All the time we.
- 13:23Can actually monitor this. So I'm just curious to see the
- 13:29causality evolve. I think the current chat bots
- 13:33would be net good in that they're trying to ground people
- 13:36in reality. If anything, I'm concerned about
- 13:39where we're going, which is this embrace of what the social media
- 13:43companies did, which is moving. You know, Open AI, Anthropic and
- 13:48everybody else wanted accurate answers.
- 13:50You know, that works in the enterprise.
- 13:52That was sort of what researchers building.
- 13:54But obviously the social media world learned, oh, we want
- 13:56engagement. Who cares?
- 13:58The humans can decide if it's accurate.
- 14:00And so, yeah, we get things can get worse.
- 14:04But right now, if the models are trying to be accurate, often to
- 14:07me, it's bringing me back towards reality.
- 14:10You know, it's like, OK, no, don't worry too much about that.
- 14:13This is what's normal. I think they're that's true, but
- 14:16I think clearly like especially pre GPT 5 and maybe in some
- 14:24other apps, like you definitely saw models that had that
- 14:27sycophant C problem that would like, you know, eventually go
- 14:31off the rails as the conversation and the context
- 14:34window got longer and just start agreeing with you.
- 14:38So that I think is the risk vector that still exists, you
- 14:43know, but even even if the models you know on your first
- 14:48attempt are pushing back. For anyone running a startup
- 14:51right now, Mongo DB for Startups is one of those programs that's
- 14:54looking to make the life of a founder easier.
- 14:57It's basically a shortcut for founders who want to move faster
- 15:00and spend less. You get free credits on Mongo DB
- 15:03Atlas, so your products ready to scale from day one without
- 15:06worrying about database costs. They also throw in 200 million
- 15:10Voyage AI tokens so you can start experimenting with AI
- 15:13features right out of the gate. You'll get one-on-one time with
- 15:16actual Mongo DB engineers, people who can help you
- 15:19troubleshoot, optimize, and think through your next stage of
- 15:22growth. There's even Co marketing and
- 15:24partner opportunities, which means more visibility and
- 15:26potential customers without extra spend.
- 15:29If you're building, scaling, or just trying to stretch your
- 15:32runway a little longer, Mongo DB for startups can actually help.
- 15:36Head to mongodb.com/startups to learn more.
- 15:41I think a theme for the conference this year will be how
- 15:45much AI is holding up the US economy and whether I think
- 15:55there's a lot of downstream questions from that.
- 15:56But like, you know, whether it's a bubble, especially in the AI
- 16:01infrastructure build out, whether AI is going to cause
- 16:07bunch of massive job losses and anytime soon, you know, how
- 16:12does, how does the the music stop or how does, how does this
- 16:16whole thing wind down? And if you didn't have these
- 16:21massive data center build outs, would the entire political
- 16:26situation in the US be different?
- 16:28Would you know, the right, you know, would we be in a
- 16:32recession? And then also like to what
- 16:35degree is there, to what degree is there round tripping of
- 16:38investments between all these companies that is risky?
- 16:44I mean, the debt stuff is getting crazy huge amounts of
- 16:47money. You know, the first Cerebral
- 16:49Valley was March 2023. And there if you were in that
- 16:54room, I should have been more entrepreneurial or, but you, you
- 17:00could have made a lot of money just being like, oh, these
- 17:01people really see where the world's going.
- 17:03Let's like invest in NVIDIA. You know, people did make a lot
- 17:06of money doing that, you know, as you'll sort of see from our
- 17:10draft picks later in the episode anyway.
- 17:14But I think this I think this one, I don't think this is
- 17:17necessarily the end of the bubbleimean.com ran for years
- 17:20and years. And obviously like, you know,
- 17:22some of the things sustained in bubbles, you know, it's like you
- 17:25people start shouting bubble when you're when you're at
- 17:27evaluation that might not, you know, they might be below where
- 17:33the bubble ultimately falls too. So it's like there's a run up
- 17:36and then there's a big fall. People hate the fall.
- 17:38So I don't know where we are, but I do think this one, it'll
- 17:43be interesting to see who plays the role of like cheerleader.
- 17:47It's like, yeah, things are good.
- 17:49Like don't the party, we need this party to keep going here.
- 17:51You know, this sort of, there'll be the true believer
- 17:54cheerleader. There'll be this sort of
- 17:55disingenuous cheerleader where it's sort of like my job in this
- 17:58whole business and my employees hang on me saying this thing's
- 18:01great. And then I think, but I think we
- 18:03attract and I think this will be a large segment of sort of the
- 18:06realists. And this is one of the reasons I
- 18:08like AI more than crypto. I think it's less cult like.
- 18:11And I think there are lots of people that are like, I'm doing
- 18:14my smart enterprise thing. We're selling to reasonable
- 18:17customers. And yes, obviously I'm happy
- 18:20that open AI might get, you know, 10% better because of
- 18:26wasteful, you know, infrastructure spending, but
- 18:28it's not necessarily how I'd run my business.
- 18:30So I I think there'll be voices of caution.
- 18:31I don't know what do you guys expect I guess in terms of.
- 18:36Hype or bubble? Bubble caution.
- 18:38I mean, I do think it's become like the intellectually en vogue
- 18:41thing to like talk about the bubble and like complain about
- 18:43the bubble and everything. I'm not totally sure that
- 18:47anyone's like investments are really indicating that they
- 18:53believe there's really a bubble or that they believe it's going
- 18:56to pop anytime soon, I guess would be.
- 18:57Investments in terms of venture funds companies venture.
- 19:00Funds LP's, you know, to your point about crypto, it's not
- 19:05sort of all about like exiting with the, you know, with the
- 19:08bags before everyone else like it gets scammed, right?
- 19:11Like I don't think people are selling out of these really hot
- 19:14companies very much and I don't think that people are, you know,
- 19:19not doing the next round at the next craziest evaluation yet in
- 19:22any meaningful way, right. I think that, I mean, Tyler
- 19:26Cowen always has this comment about how, well, if you actually
- 19:29believe the US political system is gonna collapse, why aren't
- 19:32you short the market? Basically, it's a big comeback,
- 19:34right? And it's like, there are
- 19:35different ways to analyze that question.
- 19:37But I think that in AI, it's like, okay, well, if you
- 19:40actually believe we're near the peak of a bubble, you can't
- 19:43really be short necessarily, but you could be secondary ING out.
- 19:46You could be not participating in the next hot funding round.
- 19:49You could like, you know, you could be doing various things
- 19:51that would indicate that you actually believe we're near the
- 19:53top. And it doesn't really seem like
- 19:54anyone's doing that. I mean, you're more of an
- 19:56insider than we are, Eric, but that from the word on the
- 19:58street, that's not what. I'm my, I think this thing's
- 20:01still got juice, you know, and I because I talked to, I talked to
- 20:05skeptical people, you know, I have, you know, lots of the
- 20:09investors I like to talk to investors who.
- 20:11Do care about market timing and sort of say, Oh yeah, you needed
- 20:15to not deploy a ton in 2021 if you want to make a bunch of
- 20:18money. You know, like those types of
- 20:19investors are super useful. And like, I think they're like,
- 20:24the party's still still roaring, you know, I mean, just, you know
- 20:29what? Amazon's laying off like 10% of
- 20:31their corporate staff. Like if it does, if new jobs are
- 20:36going to AI, if consultants are being taken by AI, you know,
- 20:40like there there's you can do huge market sizing on this
- 20:44stuff. There's a funny case study I saw
- 20:49this is one of these it was on Twitter things.
- 20:51So I should be a little more responsible.
- 20:52But somebody was like some, you know, a company was going to buy
- 20:56AI startup and then the consultants like as part of the
- 21:00acquisition made a demo product that was like supposedly as good
- 21:06as the start up in like 2 weeks. And I is that pro AI or anti
- 21:11anti AI? Because on the one hand it's
- 21:13like, OK, you don't need to buy this.
- 21:16You know, there are lots of like fake companies that could be
- 21:18bought. On the other hand, it's like,
- 21:19oh, you think it's a valuable thing that can be built in just
- 21:22two weeks? Like it.
- 21:24Yeah, I don't know. It seems like there's a lot of
- 21:25stuff stuff coming. Well, yeah, I mean, the numbers
- 21:28that people are able to raise for compute and scaling and
- 21:31everything are still just out of control.
- 21:32I mean, there was that like rumor 18 months ago.
- 21:35Remember that? Like Sam Alman was trying to
- 21:36raise $7 trillion for scaling, right?
- 21:40And people were like, ha, ha, ha.
- 21:42That's so ridiculous. Like $7 trillion is like, is
- 21:46like 1/3 of USGDP. It's like 1/3 of the US economic
- 21:51output. It's just like Sam Alman
- 21:52personally, it's gonna like fundraise 1/3 of U.S. economic
- 21:55output for some some microchips. And then I saw someone check in
- 22:00on this recently and they're like, well, if you put together
- 22:02all these like Stargate commitments, NVIDIA and AMD and
- 22:04Broadcom and all this shit, they're like they're up to like
- 22:071.5 trillion now. And it's.
- 22:09Like right? And Sam Ullman, I think he
- 22:11basically said like, yeah, the 7 trillion people thought that was
- 22:15too absurd. So I had to like, come up with
- 22:16like a smaller one that people, you know, it's like clearly it's
- 22:19just like, yeah, we need a lot of money.
- 22:21I'm trying to reset expectations. 1.5 trillion in
- 22:25fundraising right over whatever 12 to 18 months.
- 22:29Yeah, there are a lot. It's also sloppy.
- 22:31Like sure of. Course, but still a lot.
- 22:33I think it's big, Yeah, we can all agree.
- 22:34That and definitely the economy is, you know, the stock market
- 22:37is very is in, you know, upon it, yeah.
- 22:40We've just never seen this level of fundraising in any of our
- 22:43working lifetimes, right? You know?
- 22:46So TBD if that means they can only go.
- 22:49Higher you saw the cursor. But.
- 22:51This is this is the the ultimate insider bubble work.
- 22:54The cursor lost one of its Co founders or I do think there are
- 22:57some of these where it's like such high flying valuations they
- 23:00raise around and then like they're going to lose a Co
- 23:03founder. I mean, I don't know the inside
- 23:05story there, but there, there clearly are going to be, you
- 23:09know, some of these AI startups that get to raise on.
- 23:13It's a good, good time to be a journalist, though this doesn't
- 23:15happen A lot of juicy stories that we'll.
- 23:18Log on, yeah. I like that your sign of the the
- 23:23the party continuing, Eric, is the more layoffs we we see in
- 23:27the US economy. They're pretty hilarious.
- 23:30They're driven by AI, right? I don't know.
- 23:32That's a good question. Like if it's Amazon, you know?
- 23:37Yeah, or or is it just? I do have like, I mean, I do
- 23:39have two more like sort of real themes I actually think did
- 23:43happen this year, if that's helpful.
- 23:44And try to stick with the stick with the term.
- 23:47So I called mine Fiction Frenzy, I think.
- 23:50I told mine I called mine AI. Is AI holding up the US economy?
- 23:56Too. Long.
- 23:57Too long. Well.
- 23:58Mine are. Yeah, U.S. economy hangs in the
- 24:02balance. Yeah, anyway.
- 24:03My themes are all internally consistent, which is they're all
- 24:06the year of X, but it's is it the year of X?
- 24:09We really didn't coordinate on editorial style.
- 24:12All right, Max. So, so I had, I had the Year of
- 24:15Agents and the Year of Voice both as being kind of bullshit
- 24:18themes, but I think there are fake themes.
- 24:20Yeah, there are two themes that I think were very real themes
- 24:23this year and saw, you know, real contributions to the way
- 24:27people work. Maybe ultimately to Amazon
- 24:29layoffs, like, I don't know. But I think, you know, the the
- 24:33number one thing to me that was sort of thematically important
- 24:37this year was the year of the thinking models, right, the
- 24:40reasoning models. And I know this isn't new news
- 24:43anymore, but you know, O one was only released like 11-12 months
- 24:47ago O3, which is the first thinking model that could
- 24:50incorporate links and actually start referencing real
- 24:52information. And then ultimately, you know,
- 24:54we got GPT 5 with the thinking version of that, you know,
- 24:58Claude Opus, all this kind of stuff, right?
- 25:00We, we actually made this dramatic architectural
- 25:02transition in, in terms of how LLMS work in the last, you know,
- 25:066 to 12 months that I think took them to a whole nother level of
- 25:10intelligence and utility. And at least in my personal
- 25:13life, you know, as a startup founder, whether it's, you know,
- 25:15doing product planning or writing memos or researching
- 25:19things or writing emails or whatever, you know, any of the
- 25:22million things, you know, you could use an LM for.
- 25:24I just, I feel like my personal usage has gone up like 10X since
- 25:273 launched essentially in March or April because I feel like the
- 25:30utility of these models with thinking combined with sourcing
- 25:36leads me to find them. I've seen an inflection point in
- 25:40the value I'm getting out of these models.
- 25:43And I think that, you know, if anything sort of real happened
- 25:45to this year, I, I think it was the transition to thinking
- 25:48models that could, you know, reference, make references and,
- 25:52and actually sort of be accurate in a lot of ways since.
- 25:54And maybe that's what's leading to these job losses or these
- 25:57crazy things you could do with these models now.
- 25:59Astute No, no argument. I mean it it it's just how hey,
- 26:04I move so fast. I'm like old news and you know,
- 26:06like what, 11 months ago? And it's true.
- 26:09It's. Transformational in a 12 month
- 26:11time frame, so. Transformation London.
- 26:14I mean again, I I consider O3 just.
- 26:17Even O3 which is less like 8 I think 8 months or.
- 26:20Something. Yeah, yeah, eight months ago,
- 26:22right, So. And this was one of Carpathy's
- 26:25sub points, right? That there, there have been
- 26:28these improvements and we still need to link all this data and
- 26:31get all this. I mean, it's the, the theme
- 26:34every year is sort of like, man, there, there's a lot of value
- 26:37right there that we have that feels like we're not explaining,
- 26:41you know, here's how you should use it.
- 26:43I mean, at Dave's sex Medicina, the healthcare health event, we
- 26:46did, you know, some of the some of these companies, I don't
- 26:50know, are they that different than Chachi BT but they're
- 26:52wrapping it in like doctors feel comfortable with it.
- 26:56They're showing doctors the sources, they're making it work
- 26:58for medicine. And like, this wasn't one of my
- 27:01themes, but medicine is one of the areas where AI is going
- 27:04super hot. Yeah.
- 27:06And. And so.
- 27:07Yeah. I think, yeah, I think, I think
- 27:09similarly, you know, Harvey as the legal AI, the sort of, you
- 27:13know, scuttlebutt 2 years ago was that it, it actually sucks
- 27:16and it's not very good, you know, as a legal model.
- 27:19And then I do think basically it got a brain transplant from
- 27:22these thinking models in the last 9 to 12 months.
- 27:25And I mean, my parents are both lawyers.
- 27:27My dad's like, this is awesome. Like, I mean, like, it's
- 27:29amazing. You know, I just feel like the
- 27:31the buzz you're hearing on Harvey or similar legal models
- 27:34is just way more positive in the last 6 to 12 months.
- 27:37And I would argue it's all kind of downstream of how good these
- 27:40thinking models are. I mean, I think ultimately it
- 27:43gets into this question of if you have a text in, text out
- 27:48only model, right? You know, do we reach AGI in two
- 27:54to three years, right? Because I think that I think
- 27:56that I'm pretty confident that we're nowhere near AGI when it
- 28:01comes to visual interpretation, when it comes to multi step
- 28:06processes, when it comes to audio and video, I think we
- 28:10haven't really reached sort of quote UN quote AGI levels,
- 28:12right? I mean, James and I were talking
- 28:14about recently like, you know, I do these little basic visual
- 28:17tests with GBT thinking, you know, the most advanced model.
- 28:20I took a picture of a bunch of pomegranate seeds on a plate
- 28:23because my daughter's willing to pomegranate seeds right now,
- 28:26because we counted all the pomegranate seeds and there were
- 28:27like almost exactly 100 pomegranate seeds on this plate.
- 28:31You know, it was like 99. And I took a picture and it was
- 28:33like, there are 76 pomegranate seeds on the plate.
- 28:36And I'm like, no, they're not like we counted.
- 28:37There's like, you know, a hundred, 101.
- 28:39And then I was like, hey, we counted.
- 28:41There's like 101. It's like, you're right.
- 28:43It's probably a little more. I think it's like 82 pomegranate
- 28:45seeds, you know? And you're just like, OK, it's
- 28:48not 82, it's not 75. You're like like my 3 year old
- 28:51can count pomegranate seeds better than AGI or whatever you
- 28:54want to. Call it.
- 28:54Do you think? Do you think that like will get
- 28:57solved by models or do you think that will get solved by tool
- 29:01use? Right?
- 29:02Where you could imagine that GBT doesn't have a connective tool
- 29:05to the exact right ML model that can count pomegranate seeds, but
- 29:12like, it probably exists. Deeply wrong that it's so
- 29:15confident when it doesn't have the tools.
- 29:17So totally something is wrong there, like if you were just
- 29:21dealing with another human or another.
- 29:23They would say like, I can't count, I can't count.
- 29:25They would just tell you. You don't know.
- 29:27You know you get it wrong all the time.
- 29:28People are shouting at you. They get it wrong.
- 29:30Like no, it speaks to some breaking something.
- 29:33Wrong. It was like this same day there
- 29:35was some tweet about how it got a gold medal on the Math
- 29:37Olympiad or whatever, which is like the hardest math test we've
- 29:40invented for human beings. And then I'm like, it can't
- 29:42count. Fucking pomegranate sea.
- 29:43It's like, it's like, it's like worse than a three-year old.
- 29:45This was solved by Justin Hoffman in The Rain Man Like.
- 29:49This yeah. So my point to come back to this
- 29:53is that I think for visual audio tool use whatever, you know,
- 29:56James's category, there's just a long way to go in a lot of these
- 29:59areas, right? But I think if you're like, OK,
- 30:01text in, text out, you know, we're already at this gold
- 30:04medal, you know, US math Olympiad level.
- 30:06I think that, you know, O5O6O7, you know, I know they renamed
- 30:11all these fucking things. So whatever numbering scheme you
- 30:13want to take. But the next two or three
- 30:15generations of models, I can see us getting to a text in, text
- 30:20out, you know, AGI level of thinking, right?
- 30:23Where I for me, the leap forward we took between GBT 4-O and O3
- 30:27was the first time that I really believed, you know, I got AGI
- 30:32pill that like, Oh yeah. As at least as a chat bot, this
- 30:35thing is going to be, you know, much, much smarter than the
- 30:38average human and basically able to do anything and potentially
- 30:41break ground in terms of scientific discoveries and other
- 30:43things like that. Related theme, My last theme
- 30:45related. And I just wanted to add it on
- 30:47as a sort of postscript. We sort of forget again another
- 30:52old news thing that the term vibe coding was coined this
- 30:56year. The idea?
- 30:58Yeah, what? Karpati's tweet was like January
- 31:01this year or something? January is not this year.
- 31:03That's like a different. Year.
- 31:04No, it was it. Was this year?
- 31:08Isn't it crazy? Like, so I would just say as a
- 31:11postscript, as an extension of the thinking model progress and
- 31:14as a post, an extension of what we've seen with these text in
- 31:17textile models. And I would call this the year
- 31:19of vibe coding, right? I mean, this is the clearly the
- 31:22mimetic phrase of the year. There's a million startups that
- 31:25are getting crazy evaluations off of.
- 31:26It, it, it is. And we were having Guillermo at
- 31:28Versal and Amjad at Repplit, both on stage, different panels.
- 31:32We had lovable at CV London, you know, exactly right.
- 31:35I would say internally to our company that all of our
- 31:38designers and product managers now believe they can code, you
- 31:41know, websites and prototype things for internal tools like
- 31:45we have. Clearly, you know.
- 31:47Cerebral Valleys. Logos 3D AI generated or like
- 31:50the design was human, but I think the animation was yeah.
- 31:54So I would just say vibe coding as a phrase that has entered our
- 31:56lives, you know, representing the idea of telling the coding
- 32:00bot to what to do and then getting back results and sort of
- 32:03riding the vibes, as Karpati said when he coined it was a
- 32:07big, big deal and I think was a huge part of what we actually
- 32:11saw happen in the world this year based on large models in
- 32:15AI. So yeah, those are my 2.
- 32:16I love how your strategy is look at the dates and be like some of
- 32:21these things we take for granted are we were so greedy.
- 32:24You know, it's like a lot has happened and it's still very
- 32:27current. And so we're gonna be unpacking
- 32:29those those themes. Well, well.
- 32:31You spent all this time complaining about things that
- 32:32were too early. I'm.
- 32:34Saying no, no, no, no, I like it actually here.
- 32:36I think it's it's a good point and you're differentiating
- 32:39obviously agents and other stuff.
- 32:41You know, we got I'm not I'm not criticizing.
- 32:43This is just funny. Actually worked.
- 32:44This year in my opinion, right? Yes, and date them to 2025.
- 32:48Yeah, James. Yeah, I, I have a, a tangent out
- 32:53of that theme for Max, I guess I'm calling it the the jagged
- 33:00edge of AI. Essentially, if you're not
- 33:04someone who's using these coding models or vibe coding all the
- 33:08time or playing with the latest VO model like or Sora like,
- 33:14you're just not really seeing the dramatic improvements.
- 33:19So I think most people in the world are are, you know, maybe
- 33:23using chat GPG 5 now and getting a little of this thinking and
- 33:28reasoning, you know, experience, but not much beyond that.
- 33:35And then they are thinking, oh, this is going to be another over
- 33:38promising technology. This, you know, is kind of a
- 33:45Silicon Valley pipe dream AGI. And then what you hear from the
- 33:50labs themselves is like, Nope, everything's going pretty well.
- 33:54Like on track to AGI, you know, we're gonna have automated
- 33:58science in a couple of years. And but Dario made some
- 34:01predictions that aren't really holding up right, like some of
- 34:03these predictions. I think that.
- 34:04What, didn't you say 90% of code was gonna be written six months
- 34:08ago? I think that's not that far off
- 34:10inside of anthropic like I think like.
- 34:12This is like self driving. Just like definition question or
- 34:16something or? No, no, I self driving where is
- 34:18just like tomorrow, tomorrow, tomorrow.
- 34:19We're almost there. I'm literally saying like I
- 34:21think today you could would make an argument based on what I've
- 34:24heard from Anthropic that you know 90% of the code is written.
- 34:28You're saying it's a human problem.
- 34:30The humans just haven't really figured out how to use the good
- 34:32technology. But it's like positive.
- 34:34Yeah, yeah, exactly. Like and I you hear from open AI
- 34:38researchers and engineers that they like vibe coded or you
- 34:43know, used codecs, their own, you know, version of cloud code
- 34:47to build like a lot of the products that came out at
- 34:49Devday. So I actually think there's just
- 34:53like more adoption happening within labs than people realize
- 34:58and like the people who are AGI pilled in lat in the labs have
- 35:02like a good reason to be and and we're not seeing all of that.
- 35:05Yeah. I mean, I, I agree with that.
- 35:07I agree with the jagged frontier in general.
- 35:09I think that it's, it's already not evenly distributed how good
- 35:14these models are at some things. And, and I think particularly I
- 35:17think the sort of GPD 5 launch to your point was sort of this
- 35:21anti PR moment where people are like, well, AI like it's not
- 35:24really any better now. It's just sort of like it's just
- 35:26kind of sucks. And it's like, well, try
- 35:28clicking the little. It's insane.
- 35:30Yeah, yeah, it takes some. You have to do the work, you
- 35:34know, you have to find the value, right, Chaji?
- 35:37But it's insane, this picture that you're painting that it's
- 35:40like, Oh yeah, they just have access to Chaji Buti and they
- 35:43can't see the value. I'm like what?
- 35:45Like they're. Do you agree with my my?
- 35:47Sort of Astoria or version of that, that most people are,
- 35:51yeah, kind of, yeah, to Max's point, disappointed by the the
- 35:55ChatGPT 5 launch. Yeah, I think there are a lot of
- 35:59sort of skeptics. It takes too much work.
- 36:01Yeah, to get the value out of it.
- 36:04And people need to be led to water.
- 36:06That's what that's what I keep saying.
- 36:08You know, it sort of needs to be handed to them.
- 36:09Here's the thing it can do. I hear a lot from in even people
- 36:13in Silicon Valley who say, oh, the labs are lying about AGI.
- 36:18You know, if you look at what's opening eyes releasing it's, you
- 36:23know, Sora and they're gonna just hack attention.
- 36:28And now Sam is talking about erotica in ChatGPT.
- 36:34So like, you know, it's all, it's all, yeah, it's all we're.
- 36:39Talking about AGI less right the public commerce, it doesn't feel
- 36:43like the labs are talking about AGI.
- 36:44Don't agree with that. I think they, I think they still
- 36:46talk about it a lot. Like I mean.
- 36:48I think yeah, I think, I think they we.
- 36:50Should gauge that I'm. Talking about it.
- 36:51What? Live streaming?
- 36:52Yeah, yeah. I mean one thing.
- 36:55I mean Microsoft right? Like the they just announced
- 36:57this? Yes, it's contractual.
- 37:00Yeah. Microsoft had this deal with
- 37:02Open AI and you know, some of the triggers are some of the
- 37:05deal terms are conditional on AGI, but in some ways it goes
- 37:08both ways because it's a deal. So Microsoft could be like AGI
- 37:11is not close, that's good for us.
- 37:13We have a better deal, you know, so maybe Open AI thinks it's
- 37:16close. Now I guess Open AI can declare
- 37:19AGI to get out of the contract and then there's like an
- 37:22independent panel of reviewers that will decide if if AGI has
- 37:28actually been achieved, which I thought was pretty interesting.
- 37:30Here's the last theme that I'll be quick about.
- 37:32Don't say EA, like don't say effective altruist.
- 37:37I mean that not like we're bringing that up all the time,
- 37:40but I think, you know, anthropic is obviously run away from that
- 37:43term. You know, we had what we had
- 37:45Holden right, who is a former kind of yeah, speak at one of
- 37:51them. So it's been come up.
- 37:52He's married to Danielle at Anthropic.
- 37:54She spoke like Anthropic doesn't want to talk about effective
- 37:57altruism anymore. And then there's this sort of
- 38:00Trump world sort of thing. You know, you saw David Sachs
- 38:02fighting with Jack Clark over having any sort of safety ISM.
- 38:07And so it's beyond just don't say yeah, it's will people talk
- 38:10about safety at all and how much in sort of a Trump world is even
- 38:16just worrying about AI safety Lib coded and therefore nobody
- 38:19wants to talk about it. Yeah, David Sachs went like
- 38:23further and said the all of this was regulatory capture, which I
- 38:29thought was just a very. There was like performative.
- 38:31Jump like, like it's clearly not performative.
- 38:34To your point. Like they've been talking about
- 38:36this for a decade. Everything he does is
- 38:39disingenuous. All his arguments are motivated
- 38:41for some end, so he cannot comprehend someone having.
- 38:44Yeah, yeah, argument just on principle.
- 38:48All right, those are our themes. I'm giving myself the last word
- 38:50on that one. Very excited for this next
- 38:53segment. We're going to go into the
- 38:54draft. We've been doing this for a
- 38:56couple years now, so we already have our picks and we're going
- 38:59to be sort of dropping and then picking and picking up some new
- 39:03ones. And I'm very excited to say with
- 39:06the upped production value on the Newcomer podcast, we have a
- 39:09hype video to catch you up on what we've done to try and make
- 39:12this comprehensible. So you can follow along.
- 39:15This ultimate feat of nerddom, which is taking what people do
- 39:19for cool things like sports and using it to pick not just AI
- 39:23nerdy things, but companies. I feel like it's a double whammy
- 39:27of Loserdum and that we're we're rooting for companies to succeed
- 39:31and we're interested in in tech. So anyway, here's that hype
- 39:34video. In living rooms everywhere, die
- 39:37hard fans chase glory through fantasy football.
- 39:40In newsrooms, armchair strategist bet the house on who
- 39:44wins the Oval Office. But in Silicon Valley, these
- 39:48insiders draft generative AI startups.
- 39:52This is the AI Fantasy Draft League IN2023G Miss Eric
- 39:57Newcomer, Max Child and James Wilsterman chose their five
- 40:00startups. Open AI, Inflection, Character
- 40:03AI, Glean and Mistral AI. Databricks, Pine Cone, Cohere,
- 40:07Modular, and Imbue. Hugging Face, Anthropic AI 21
- 40:11Labs, Replit and Adept. After a year of celebration and
- 40:15regret. Dominating right now.
- 40:16Bad pick, bad pick. In like 6 months that has just
- 40:19totally changed the. GMs went back to the table to
- 40:21evaluate their teams and prepare to add 2 second round draft
- 40:25picks with Eric in the lead in 2024, followed by James and Max
- 40:29struggling to hold on to his franchise death.
- 40:32Defying public IPOs, market fluctuations and many rounds of
- 40:35funding later, the boys are back in the hot seat once again.
- 40:39All in the hopes of glory on November 1st, 2028 when the
- 40:42Champions will be crowned by total market cap valuation.
- 40:45Welcome to the 2025 AI Fantasy Draft League, presented in
- 40:49partnership by Newcomer. Wow, production value is just
- 40:55just skyrocketed. If if if that had come out
- 40:59before TBPN existed, I would say we should start.
- 41:01TBPN, not like TBPN invented production like Youtubers have
- 41:07been doing this forever. Everybody's yeah.
- 41:09But anyway, I'm we're hooked. Up ESPN before they did OK.
- 41:13I know and Riley, my business guy, literally one of his, the
- 41:18his main editorial insight the whole time he's been here, he's
- 41:21been with me three years. He's been like, you need to be
- 41:23more like sports. You need to be more like sports.
- 41:25He's like and he's like be more correct about.
- 41:27Yeah, I told. You I take this seriously.
- 41:32So I feel like the sports thing is a little glib sometimes.
- 41:34Anyway, I love I love that. That's great.
- 41:37We should get into it. James.
- 41:38You're the both a player, but also sort of the game master of
- 41:43this. I don't know what the.
- 41:45What do you call it when you run?
- 41:46Missioner. Isn't there a term commissioner?
- 41:48Commissioner. Yeah, OK.
- 41:49Yeah. Yep, this is year three-year
- 41:55three of the draft. We've been working on this for a
- 42:02while now. We all have teams just to set
- 42:05the the rules again. You know, we drafted a few teams
- 42:09originally. Now there's like waiver pickups
- 42:11and you can, you can add and drop teams.
- 42:14We're drafting companies that have raised over $100 million
- 42:18have a core use of generative AI or a core role in the generative
- 42:21AI ecosystem. There are some excluded
- 42:24companies if they're really focused on bio health, defense,
- 42:30cloud computing or or chips or silicon we.
- 42:33Block the whole chip infrastructure that which is
- 42:35sort of funny in retrospect and now that everything's flowing
- 42:38there, but but we we excluded that.
- 42:41Yeah, we could always bring it back if you guys are don't think
- 42:44we have enough companies to expect.
- 42:46Yeah, I know we're not. We're not focused on robotics
- 42:49companies or China companies based in China.
- 42:54And yeah, it's a snake draft this this season, this year.
- 42:59I always like to say I had to. I insisted that we have an
- 43:04auction for who goes first, and we auctioned and I paid $75
- 43:08billion so that I could pick Open AI first because my view
- 43:11from the beginning was that Open AI was going to swamp it all and
- 43:14you needed the winner. It's a home run business.
- 43:16So far that strategy is doing well though, so.
- 43:19Far, so far, Open AI is worth more than James and my team
- 43:23combined this year, which is a perfect expression of power law
- 43:27dynamics. Now you did pay a $75 billion
- 43:29handicap, so it's a little bit under, but it's pretty damn
- 43:32close, which is kind of mind. Boggling but it wasn't like oh I
- 43:35just happened to go first by some random chance I paid for
- 43:38the right knowing that we would fight for opening and.
- 43:41You, your team's worth close to 510 billion right now, including
- 43:47that handicap of 75 billion Max, your team's worth 150 billion
- 43:53and my team is worth close to 400 billion.
- 43:57Just to catch you up on last season, we we, we dropped a few
- 44:01companies and we had some exit. So that opened up new slots on
- 44:05our teams. I drafted Xai with the first
- 44:08pick last year and that's been a huge win.
- 44:11Now rumored to be valued around 200 billion.
- 44:15Great fall in your lap sort of opportunity, I mean your first.
- 44:18Is this the moment where I get to say I basically got Coin
- 44:21flipped out of being in second place because if I had drafted
- 44:24XAI I would be in second place by a mile just like James for
- 44:29the. Record.
- 44:29Fair, fair. Well, fortunately for you Max,
- 44:32you get to draft first this year.
- 44:35Yeah, I know. And this is the year where
- 44:36there's no. Obvious wins.
- 44:39Yeah, I know. This is oh wow, I get the first
- 44:41pick of the year where everything's worth $10 billion
- 44:43and nothing's worth 20. So I guess first to start, Max,
- 44:48do you want to just quickly go, you know, run through your team
- 44:52and then also let us know if you're dropping any of your
- 44:55team? Sure.
- 44:57My team today is Data Bricks, Cohere Modular Scale, Sierra
- 45:03Sakana and Hebia. You know, it's not great.
- 45:07Other than data bricks, I did get an exit with scale at $29
- 45:12billion. Are we calling that an exit?
- 45:15We basically I think. I'm calling that we, we've,
- 45:17we've gone through this a couple times where there's been these,
- 45:20yeah, pseudo exits count. So I'm counting that.
- 45:25Yeah, again, I'm in last place by a huge margin.
- 45:31Not not, you know, my team out on the field has just not really
- 45:33been performing. Sierra is a good company, but
- 45:36and Co here I think is gonna have a second I think they're.
- 45:39Yeah, yeah, I think we are. There's some value here.
- 45:41I think ultimately, you know, as I said earlier, had I been, had
- 45:45I won the coin flip and gotten to pick Xai, I would be
- 45:48breathing down Eric's neck kind of like James.
- 45:51And instead I'm I'm a joke, you know, looming far in the
- 45:54background. So with that, I'm gonna do some
- 45:57aggressive restructuring on my team this year and do some heavy
- 46:02dropping and try to pick up a lot of value in a hurry.
- 46:05So I'm out on scale. Yeah, I'm also going to drop
- 46:09modular. I'm also gonna drop Sakana and
- 46:12I'm also gonna drop heavier, so I'm gonna make 4 picks in
- 46:16addition to my 22 picks that I get as part of the draft this
- 46:19year. So you get 4 picks or?
- 46:21Yeah, you get 4 + 2. So you get 6 picks?
- 46:24Yes, that's a lot, yeah. Yes, I'm taking, I'm shooting my
- 46:27shot here at restructuring. Look, sometimes you gotta have.
- 46:31A rebuilding year? Yeah, I think.
- 46:32You gotta you gotta go all in, which is what I'm doing this
- 46:35year, so we're really going for it.
- 46:38Eric, you're gonna pick second this year.
- 46:41Yeah, so my team, first of all, I've dropped nobody.
- 46:45All my exits are from actual deals.
- 46:48So I think that's another point in my favor.
- 46:50But OK, I currently have Open AI, Glean, Mistral, Perplexity,
- 46:57Safe, Super Intelligence and Harvey Kodium just exited.
- 47:03So I'm slash windsurf, whatever you want to call them.
- 47:07So James has that at 2.4 billion.
- 47:10I've been scrutinized those small potatoes and then I've
- 47:15already exited inflection and character.
- 47:19So we I replaced those last time.
- 47:21So the I have an extra spot thanks to windsurf.
- 47:25But otherwise, yeah, I'm I'm holding on.
- 47:28I I like I like mine. I mean, who knows what's
- 47:31happening at safe super intelligence.
- 47:33We're just betting on Ilia there perplexity, obviously lots of
- 47:37zeitgeist glean. You know, I really believe in
- 47:39the business. I just don't know, you know, I I
- 47:42think that one has the lowest to fall.
- 47:44It's just like, does it have the true insanity?
- 47:46You know, it seems like this this thing is supposed to end
- 47:49what we said 2028, November. It's like the bubble bubble
- 47:53could still be going for all we know.
- 47:54But glean, I feel like bubble or not, I'm, I'm, I'm betting on
- 47:58that one. Yeah.
- 48:01So I I like this team in Harvey. You know, we're putting them on
- 48:04stage at St. Row Valley.
- 48:05They're getting, they're getting just another boot, the little
- 48:07boost that they need, you know, to really go to the point.
- 48:11To your point, it'll be it'll be hilarious if in 2028 where the
- 48:14bubble has popped and the stock market has collapsed and the
- 48:17final scoring is really just about like whose businesses
- 48:20collapsed the least in the. Yeah, bubbles look like human
- 48:24made this fucking fortune zone at the right time, you know,
- 48:26like yeah, yeah, your scale deal could be.
- 48:29Like open AI will be like -400. Billion and like.
- 48:32You know, James and I all have these like, little like cobbled
- 48:36together. Stones like.
- 48:37Companies can only go to 0. You can't be -400 billion.
- 48:41I don't know what you're. I mean on your current valuation
- 48:43like from 500 to 1. 100 Oh, I see.
- 48:45Yeah. I'll draw from you.
- 48:46I thought, I thought like, I just meant I thought like they
- 48:48were gonna be in debt, you know, to, to the Saudis or something.
- 48:52New to that? No.
- 48:53No, no, I just meant, I just meant 400 down from today, not
- 48:57not -400 valuation. We'll see.
- 49:00Yeah, yeah. OK, so 3 picks for Eric, 2 new
- 49:03picks and 1 exit. My team is Hugging Face,
- 49:08Anthropic, Replit, XAI Runway 11 labs and poolside.
- 49:15Getting some value from XAI obviously and Anthropic the rest
- 49:22sort of small potatoes, but poolside is kind of rumoured at
- 49:26that $12 billion evaluation recently, which I think is
- 49:29interesting. 11 Labs just had a tender at 6.6 billion and runway
- 49:37rumoured at around 5 billion. So some value here.
- 49:41I'm definitely thinking about dropping Hugging Face this year.
- 49:45That's I don't have any exits this year so I'm drop.
- 49:47I'm going to drop Hugging Face. They just haven't raised.
- 49:50Brutal. They just haven't raised, you
- 49:54know? I never thought that pick made
- 49:56sense. We talked.
- 49:57We we should talk a little bit. When you we did kind of shit
- 49:59talk that pick so. But so you guys were.
- 50:01Ready. We also should talk to other
- 50:02picks that I mean, I don't know what's going on there.
- 50:08It's an interesting question cuz they were the hottest company in
- 50:11the universe kind of when you picked them at the small scale,
- 50:14like they were like really, really, really yeah, Zeitgeist
- 50:16in their early stages and I don't know, they're sort of out
- 50:20there. People put their models up on
- 50:21Hugging Face, but maybe for some reason they have not gotten that
- 50:25AI premium like everyone else. Yeah.
- 50:27I mean, they did rate, I think they where, where were they at
- 50:31when I drafted them? I mean and they've.
- 50:33Weren't they at? Four, yeah, four, 4 billion, so
- 50:374 1/2, yeah. So I don't know, I don't, it
- 50:39just seems like maybe hard to get another round done at a
- 50:42higher valuation there. So I think that I'm going to
- 50:46have to drop and free up a spot there.
- 50:51The other ones are a little trickier because they're all
- 50:56doing pretty well, but are they the next open?
- 51:01AI poolside is a black box. Yeah, poolside's a black.
- 51:04Poolside still feels in the bigger big good.
- 51:07Blood still hot, obviously Amjad's coming to the
- 51:10conference. XAI is great.
- 51:13Runway's pretty exciting in the video space 11 Labs is raising
- 51:19at seemingly every year at increasingly large valuations,
- 51:25so that's a good one. I think I'm just going to leave
- 51:27it at that and take 3 new 3 new picks.
- 51:31Great. All right, let's do this thing.
- 51:32All right, time to pick. Time to pick Max with the first
- 51:37pick of year 3. I'm going to go with my gut here
- 51:41even though this technically isn't the highest valuation on
- 51:43the board and just go with the strongest vibes I think
- 51:48available on the board and say cursor the coding application.
- 51:54You're locked into that. You're locked into that.
- 51:56I'm. Locked into cursor you wanted.
- 51:57That I think that's insane. I'm I'm overjoyed and I'm like,
- 52:01that was not one of the ones I was going to pick.
- 52:03I just think that. I just think that my other
- 52:05options are cursor competitors, essentially scale AI
- 52:09competitors, questionable foundational model companies.
- 52:14You know, I don't know, I just the heat on cursor is so hot
- 52:18right now so I'm happy to hear the negative case now that I
- 52:20picked it since Eric clearly has a strong.
- 52:22I just think it couldn't be hotter than it is.
- 52:24Yeah, I mean, but that's what we said.
- 52:26About but it's like 20 billion, right?
- 52:29Why do we have? We have a listed here at, not.
- 52:31Yeah, okay. I think it is 20, so maybe it is
- 52:33number one right now. Yeah.
- 52:35I mean, has there not been a lesson from this draft other
- 52:38than a draft the highest valued company for.
- 52:41The first pick, certainly, yeah, we're going for market caps,
- 52:44not, not a sort of climb. So or two for two.
- 52:48On the highest valued company being the best pick in
- 52:51retrospect so far. So I don't know.
- 52:53I can't. I can't, can't talk myself out
- 52:55of taking. It, it is, it is rumoured at
- 52:57around 30 billion. It's just not raised 30.
- 53:00It's not raised at that level. So it's Max.
- 53:06It's a good Max. Remember when the last two years
- 53:09you've told us like that we were dumb for picking all the hypiest
- 53:12companies and. No, I wanted xai.
- 53:14I wanted xai, I wanted, I wanted XA clear, definitely.
- 53:18Pull the tape. I I wanted xai.
- 53:21So don't. Don't back.
- 53:22Don't. Retcon that.
- 53:24If Tesla and XAI ever merge, we're going to have like a big
- 53:27challenge in evaluating that that I also believe.
- 53:30That I Yeah, yeah, I believe I bid up Eric farther with open AI
- 53:35than you did, James. So I was more.
- 53:36Of a believer that's. True.
- 53:38All right, so I'm next. With the next pick, Eric.
- 53:44This is a hard one. Not so easy now, is it?
- 53:47My go. It's hard to be the man in the
- 53:50chair now. I am picking cognition labs.
- 53:57OK. All right.
- 53:58So you, so you literally. Spike for the Spike Pepsi Dick.
- 54:02Or a joke. They're they're they're like you
- 54:05bet What? You went to Pepsi, I picked Coke
- 54:08and you picked Pepsi and you were shitting on.
- 54:10My I'm betting on this is fucking Red Bull.
- 54:13This is the next generation. They're like doing the new
- 54:15thing. You guys are stuck in the pads
- 54:17trying to sell sugar water, you know?
- 54:19Like yeah, wait, how is Devin Red Bull?
- 54:25Please explain. Just because it's like they've
- 54:28got a new, they have a new growth strategy, they're doing
- 54:30new things. I'm just saying it's like it's
- 54:32not Pepsi. I mean, obviously, you know
- 54:34what, Pepsi, I assume is bigger than Red Bull.
- 54:36I don't actually know, but I'm just saying, you know,
- 54:38technicians got some juice. They got new things going there.
- 54:41They've, I feel like there's upside.
- 54:43That's what I'm trying to say. They are the new, new coder on
- 54:46the block. Even though they've been around,
- 54:47it feels like, I feel like their momentum is is gonna be hitting
- 54:51over the next 12 months. All right, sounds good.
- 54:55Good pick. Good pick.
- 55:00I have some tough choices to make here For my first pick.
- 55:06You get 2 picks, I get 2 picks. Yeah, that actually helps a
- 55:08little bit. We do a snake dress.
- 55:10Yeah, yeah. I'm going to take for my first
- 55:13pick Mercor which is the data labeling startup that.
- 55:18That was my second choice. Isn't, isn't that also basically
- 55:21a scale AI client? It is, yeah.
- 55:23I'm OK, All right. Which did exit at $30 billion,
- 55:27right? Get me wrong?
- 55:27And and apparently when Windscale kind of got Aqua hired
- 55:31by Meta, like a lot of business seems to have gone to Mercor
- 55:35like because all the other labs were like, I don't want to send
- 55:37all of our data to Meta. So just raised I think in
- 55:42October, a $350 million round is my what my research says here.
- 55:48And then valued at close to 10 billion.
- 55:51So yeah, going with the the data labeling play, which is another
- 55:55hot area of AI right now, definitely could come up as a
- 55:58theme in the conference. I'm I'm happy either way with
- 56:01what you go. I I think I'm I'm feeling you're
- 56:04making my life easy for my next. Pick Really.
- 56:06You're like but but. Because I don't have to make the
- 56:10decision I'm saying. I think there are two obvious
- 56:13ones to choose from. No.
- 56:14And you're saying that I haven't choose one or the other, OK.
- 56:18Yeah, and I don't have to. I'll just get the other one.
- 56:21OK, well let me take a look here for my second pick.
- 56:25I'm like, nervous. I take this very seriously.
- 56:31I'm gonna have to go with thinking Machine Labs.
- 56:33I just think it's another foundation model play and who
- 56:37knows? Those are those are the only
- 56:39ones that seem to matter in this draft in terms of potential
- 56:43upside. Ride with and the insane acquire
- 56:48potential there is. True, true.
- 56:50And there seems to be a lot of true believers there because
- 56:52like, it seems like Mark Zuck can't, can't poach anyone for
- 56:5810s of billions. He got, he got 1.
- 56:59He got, oh, he eventually got that guy.
- 57:01Got the founder of. Thinking Oh yeah, yeah.
- 57:03Yeah, that's right. Yeah, yeah, yeah.
- 57:04No, he grabbed the founder of. Thinking fair enough.
- 57:08The Yeah. And nobody knows what they're
- 57:11doing, but yeah. They come out with interesting
- 57:14research it seems like. Yeah.
- 57:17But I agree it's a little bit of a hot just a Mira.
- 57:21Murati, CTO of Open AI, founded this company.
- 57:25Andreessen Horowitz is now a big backer.
- 57:27But it's totally mysterious. And when this thing goes to 200
- 57:31next year, you guys can't say that I got a steal cuz you had
- 57:34the shot at this, you had the shot at this.
- 57:36So I'm excited. I already have the ILIA 1.
- 57:39So on one hand I was like, oh, safe super intelligence, have
- 57:42thinking machines and have, I don't know, the pair of
- 57:45speculative super hyped, you know, but.
- 57:49It made it a little easier to pick right after Mercor, cuz I'm
- 57:52like, all right, Mercor, we got real revenue and a real business
- 57:55and something's happened in there.
- 57:58But yeah, passing it back to Eric.
- 58:01I'm back. It's back to me.
- 58:02Yeah. All right.
- 58:03This is a good one. I think you guys know what I'm
- 58:05gonna pick, right? I mean.
- 58:07I mean, I have a guess, yeah, but I don't want to this this.
- 58:09Founder can eat the fucking world.
- 58:11I sit down with him and I'm like what business aren't you going
- 58:14to destroy Guillermo Co Versal I'm I'm picking up versal.
- 58:19They first of all, even if they don't succeed in vibe coding
- 58:23themselves, they are like doing a lot of the infrastructure work
- 58:27for a lot of the vibe coding. So it's like lovable winds fine
- 58:30like bill still use versal. So I I think Versal is like in a
- 58:34cool position and I think Guillermo is a very talented CEO
- 58:39and yeah, I think they're going, they're going place so happy to
- 58:42pick up for a sell a sort of infrastructure play with, with,
- 58:46you know, consumer aspects. Yeah, that definitely would have
- 58:49been the next pick for me. Now it gets a lot more
- 58:53interesting, I think, because I think there's a bit of a Cliff
- 58:57here I get. Two, I'm lost now, yeah.
- 59:01Yeah, I'm pretty lost, I mean. This is when it gets fun.
- 59:04Yeah, I think thematically in the thinking machines category,
- 59:10I feel like I, I just have to grab reflection AI here just
- 59:13because open source foundational model company, it's American
- 59:18apparently that's part of their story and we're going to have
- 59:20some sort of national security defense by having another cool
- 59:24foundational model company. I I don't know, it's got big
- 59:27time backers, you know, Sequoia and DST and all these folks.
- 59:30So, yeah, ultimately, I think just, I honestly know very, very
- 59:35little about this, but yeah, I'm like, it's worth $8 billion and
- 59:39it's foundational model company. Sign me up, Scotty.
- 59:50Man, it's easy to be a VC. It's easy, especially when you
- 59:54don't have to really pay. Like we're not like buy, you
- 59:56know, we're not caring about the cost.
- 59:58We don't worry about dilution. We only care about market cap.
- 1:00:01So if you were in a VC some I think our incentives would be
- 1:00:05different. I agree.
- 1:00:06Like, OK, sure, I'm gonna get diluted like crime.
- 1:00:08Do I want it at this price versus like am I?
- 1:00:11Just trying. So the the way we set this up
- 1:00:14incentivizes foundation models and it would have also
- 1:00:17incentivized infrastructure companies.
- 1:00:19Well, this is like this is a classic debate in fantasy
- 1:00:22football, right? Is like you do snake draft or
- 1:00:24you do auction draft where you actually have to like pay out of
- 1:00:27your own budget and stuff like that.
- 1:00:29So next, next, next time we can be a little bit more, a little
- 1:00:33bit more savvy. Yeah.
- 1:00:35OK. This next one I'm I'm actually
- 1:00:38going to take what I consider to be a real risk on here because I
- 1:00:42think we're starting to get into the risk part of the draft here.
- 1:00:45Cerebra Valley alumnus here. I'm going to take lovable.
- 1:00:49I just feel like the momentum around lovable as a brand name
- 1:00:54or a meme or whatever the fuck you want to call it is just so
- 1:00:57incredibly strong. And I don't necessarily think
- 1:00:59the product is like that much better than a bunch of other
- 1:01:02similar competitors or even ultimately cursor, you know, but
- 1:01:06I think that it's just got that sort of like mimetic force and
- 1:01:11growth. And I do think the founder is,
- 1:01:14you know, has a lot of kind of narrative power around the.
- 1:01:17Company you know I lit them up on Twitter they had the most
- 1:01:20dark pattern unsubscribe no, I I don't and they're getting all
- 1:01:24this shit on Twitter about like whether their growth metrics
- 1:01:26make any sense I don't know no some red flags I was.
- 1:01:29Giving them some of that shit. So maybe it's, you know, maybe
- 1:01:31it's a dumb bet, but I do also think that like, probably there
- 1:01:34will be a consumer brand in like make a website with an LLM and
- 1:01:38maybe it will ultimately just be open AI and which point this is
- 1:01:41all irrelevant. But they feel like they have by
- 1:01:43far the strongest, like lead on being kind of the, you know,
- 1:01:47consumer brand for vibe coding, for lack of a better term.
- 1:01:51So yeah, I don't know, it's a bet.
- 1:01:53I'm not in love with it, but I've got to take some shots.
- 1:01:57And they feel like they have a very strong, you know, strong
- 1:01:59memetic force behind them, I guess I would say.
- 1:02:01And you're you're going to get a lot of picks That then is strap,
- 1:02:04I think because Eric and I are really picking.
- 1:02:06All right. I'm up and I can't give away all
- 1:02:08the companies that I think are good left.
- 1:02:10We're definitely deeper. It's less like, oh, just buy the
- 1:02:13highest valuation now I'm I'm not willing to go for a there.
- 1:02:17There's, you know, companies that I like sub 1 billion that
- 1:02:20I'm like, Oh, you could bet on them, but I it seems risky.
- 1:02:24I'm OK I'm embracing one of Max's themes, which is video is
- 1:02:28doing better than voice. And I also do think, you know,
- 1:02:32there it's more fun to this isn't like a fun business to
- 1:02:35pick, but it's like selling to businesses is a good idea.
- 1:02:39There's money to be made. That's sort of the glean case.
- 1:02:43And so, and I really like the CEO of this company, Victor.
- 1:02:48So I'm gonna pick Synthesia. Synthesia.
- 1:02:50Yeah, okay, yeah. And it's also got the, it's
- 1:02:52European undervalued sort of thing going.
- 1:02:55And, and so yeah, I, I think, I think they'll go far.
- 1:02:59I'm picking Synthesia. Well, they just leaked that they
- 1:03:02turned on an acquisition offer for $3 billion, right?
- 1:03:04I missed that. I didn't even know that.
- 1:03:06Oh yeah, yeah. So they're definitely gonna
- 1:03:07raise it like 5 or 6 pretty shortly.
- 1:03:09Some great smart pick, smart pick, good value in the lower
- 1:03:12part of the the draft if they're thinking about raising or you
- 1:03:16know, close to 5, right. James.
- 1:03:19James final pick. I'll pick.
- 1:03:22I'm gonna go out on a limb as well here deep, deep in the
- 1:03:27waivers pool right now, but in talks I believe to raise in the
- 1:03:332 billion range, Suno. Yeah, that was also pretty high
- 1:03:39up there. I don't think that's deep in the
- 1:03:40waivers is. It right, I mean right now if
- 1:03:42you. They're they're, they're
- 1:03:43obviously leaking all their ARR growth to close it around right
- 1:03:46now. Yeah, I'm I'm just saying like
- 1:03:48they their current valuation is sub a billion, you know, 500
- 1:03:52million or something. Sure.
- 1:03:54OK, so yeah, like yeah, there's been a lot of hype around them
- 1:03:58in the last like week and they yeah, they leaked like 150
- 1:04:01million or I don't know if they did, but they that leaked 150
- 1:04:06million. Arr lot of discussion, obviously
- 1:04:08a lot of potential IP risk, but every IP lawsuit these days
- 1:04:12seems to be going in the way of the way of the the models.
- 1:04:17So we'll see if that stands. But I think no matter what
- 1:04:21they're really interesting acquisition or.
- 1:04:25Spherical Valley alumni. Yes, I I I interviewed Michael
- 1:04:28at Mikey at New York CVAI last year.
- 1:04:32You looked into his eyes. And I was like, yeah, I want
- 1:04:35this guy on my team. He's a, he's going places.
- 1:04:40He was a yeah. OK, that's my pick.
- 1:04:44Great. Nice.
- 1:04:45All right, all. Right, James and I are done all.
- 1:04:47Right, you guys are done. It's time to try to grab some
- 1:04:49value off the. So how many do you have?
- 1:04:51Off the ground ground here. Well I have 3 so far but I get 3
- 1:04:54more. This, yeah.
- 1:04:57Yeah, yeah, this is. Exciting.
- 1:04:59There was some. This is a fun thing for you to
- 1:05:00do. There are some.
- 1:05:01This is my rebuilding opportunity here.
- 1:05:03I did like Suno and Synthesia. They both would have been would.
- 1:05:07Have been in the mix. Oh my gosh.
- 1:05:08Yeah, Yeah. I literally have them in my
- 1:05:10notepad ones, ones I wanted here for the first one.
- 1:05:15I'm just going to grab one with a good valuation that we use a
- 1:05:17lot at Volley because I think that I don't know, it's a solid
- 1:05:22product and I have no idea if the business model is any good
- 1:05:25or what the future holds here, but I'm going to I'm going to
- 1:05:29take fall AIFALAI. Well, I've been hearing about
- 1:05:32that one a lot. Interesting.
- 1:05:34But it's not. I don't really know much about
- 1:05:35it. They're essentially like a kind
- 1:05:38of a clearing house or single API provider that lets you call
- 1:05:44on any of the photo and video models.
- 1:05:45So it kind of plays into this photo and photo and video theme,
- 1:05:47which is like you're going to need to be able to switch
- 1:05:50between these models really easily and grab the new ones.
- 1:05:53And they take a little cut of that.
- 1:05:55So it's almost like a marketplace for photo and video
- 1:05:57models. And I don't know if that's
- 1:06:00ultimately where value will accrue in the stack.
- 1:06:02There's a lot of argument that their margins will be competed
- 1:06:04to nothing or whatever. But like today, we at Volley as
- 1:06:08a consumers company get a lot of value out of having like one
- 1:06:12place to sort of call into for any kind of photo or image
- 1:06:15generation. And so, yeah, I don't know,
- 1:06:19feels like a. Invest in what you know, yeah.
- 1:06:23Pretty highly valued at 4 billion as well.
- 1:06:25Again, the opposite logic of normal investing, which is Oh
- 1:06:27yeah, I want to take those high valuations.
- 1:06:30Yeah, exactly right. I mean, there's this whole world
- 1:06:32model startup theme that James and I are very familiar with,
- 1:06:36which is this idea of like video, AI generated video that
- 1:06:39you can walk around in and, you know, participate in as if it's
- 1:06:43a video game. One of our good friends and
- 1:06:46board member, Moritz buyer Lentz just raised $130 million for one
- 1:06:51of these companies. But in this space, I'm going to
- 1:06:56take, I think I'm gonna take World Labs, the Fei Fei Lee
- 1:07:00startup. I think that it's just too much
- 1:07:04name brand to avoid. I mean, I've seen D cart, I
- 1:07:11haven't seen general intuitions product.
- 1:07:13I ultimately think Demosus, Abbas and Jeannie will probably
- 1:07:15win this whole thing or whatever, but maybe this is a
- 1:07:18good acquisition or role up play at some.
- 1:07:20Point what? What did you think about
- 1:07:21Descartes? Why not pick Descartes over
- 1:07:23World Labs? Yeah, I literally am just doing
- 1:07:27this based on like name brand. And my assumption is that she
- 1:07:31can raise a shit load of money and that might ultimately lead
- 1:07:34to them winning the compute and the data war here.
- 1:07:37Yeah, but obviously that could be totally wrong.
- 1:07:40OK, last pick. I think the last one I'm gonna
- 1:07:45take, and this is, again, probably just biased on what I
- 1:07:49know, is I'm gonna grab Sesame, which is this glasses startup
- 1:07:56that's focused on realistic voice assistance built into
- 1:07:59smart glasses, which sounds terrible, terrible in a lot of
- 1:08:03ways. I.
- 1:08:03Would never pick that. They just raised $250 million
- 1:08:07from Sequoia Capital, not nothing.
- 1:08:11And I do think their TTS voice is quite good.
- 1:08:14And I think that ultimately, it's interesting to see people
- 1:08:18playing in the sort of voice and glasses space.
- 1:08:22And they might get rolled up by either Mark Zuckerberg or Apple
- 1:08:26or someone else like that at some point.
- 1:08:29So it feels like just a, a real, a real Yolo play here.
- 1:08:35And yeah, I don't know, it's just kind of fun and interesting
- 1:08:38and I think the voice product is pretty good, so I feel like why
- 1:08:41not invest in that? So all right, that's it.
- 1:08:44Nice. Do you guys want to do it can.
- 1:08:46We review. Quick review.
- 1:08:48Sure. Yes, I will kick it off my
- 1:08:51roster. Previously Databricks, Cohere,
- 1:08:54Sierra and that was it. And I added a bunch of new
- 1:08:57companies today. Cursor, Star of the Show,
- 1:09:00Reflection, AI, Lovable, Fall, World Labs, and Sesame.
- 1:09:06And Eric? I had Open AI, Glean, Mistral,
- 1:09:10Perplexity, Safe, Super Intelligence, and Harvey and I
- 1:09:14picked up Cognition, Versal, and Synthesia.
- 1:09:19And I had anthropic replit XAI runway 11 labs poolside.
- 1:09:25I just added Mercor, Thinking Machine Labs and SUNO.
- 1:09:30Good drafts. Yeah, fun.
- 1:09:31We have one more. Episode before St.
- 1:09:34Roll Valley on November 12th, we'll be making predictions or
- 1:09:38reacting to some AI predictions. And yeah, if you're in San
- 1:09:43Francisco and startup founder or investor, reach out about
- 1:09:48Cerebral Valley on November 12th.
- 1:09:50Thank you for tuning into this week's episode of the podcast.
- 1:09:53If you're new here, please like and subscribe.
- 1:09:55It really helps the channel. We're building a YouTube
- 1:09:57channel. I think you can tell we're
- 1:09:58investing a lot more in our production and we appreciate
- 1:10:01your support. And if you want the data insider
- 1:10:05takes real reporting, go to newcomer.co and subscribe to the
- 1:10:10sub stack as well. Thanks for following along.