Institute for Progress researcher: US export controls on H20 chips were overdue and enforcement gaps remain critical

remote workers are in shambles. Yeah, it's a disaster. I mean, I guess everyone else just uh taking the day off. I appreciate you guys have a backup plan. This is great. Yeah, it's great. Uh well, can you give me the high level on the report? It seems like you published right before there were some changes.

uh give me the high level and how things are developing. And in all uh half joking, half serious, but do you deserve uh some credit for uh the shift here? Yeah, it seems on the credit side, I don't think we can take much. It looks like there was a lot of things happening behind the scenes.

I will say the fix that we have now is just a short-term thing and there's a lot of long-term stuff that needs to happen. But to I guess back up as to what's actually going on here. So yeah, it's been a pretty whirlwind 24 hours for the AI chip export controls.

Um, so you guys might know in uh early 2024, Nvidia launched this new um AI chip, this AI GPU called the H20. Um, it was specifically designed for the Chinese market. So it was designed to stick within the uh computational limits that were defined um through export controls.

Um, but it turns out that the H20 actually had great performance for doing inference, so running AI models as opposed to training it. Um and this resulted in a situation where actually it was really good at inference and even better than some chips that are already banned.

And this all happened during this sort of 2024 trend of inference for AI models becoming way more important for capability growth through you know models like 01 and deepse using test time compute as well as other techniques like synthetic data generation RL.

Um so a bunch of these like sold like crazy u over a million were shipped in 2024. Uh and so this is like a really big deal. We had inference compute flowing freely from the US to China during this period where inference compute was becoming the key strategic input to AI development.

And people like myself were going, "Wow, Washington should really be reacting to this. " Uh, but they were really slow to react. Um, and uh, it became sort of a couple of things that we learned last week.

One, there was a meeting between uh, Trump and Jensen, the Nvidia CEO at Mara Lago, which is apparently a $1 million ahead dinner. Um, and after that dinner, the White House reportedly reversed course on any plans to um, uh, control the H20 chip.

Um, and then last week, we also learned that Chinese companies had placed massive new orders for these H20s, so over 1. 3 million um, so far this year. And this is like recently over just the last week to this a lot of calls including from us to stop these orders going through.

And yesterday, yeah, it became clear that BAS is actually BAS is the Bureau of Industry Security who does export controls has taken action on this. Um, I just threw a lot of info at you, but that's just sort of where we're at now.

Um, can talk more about sort of what we think should happen next, but yeah, it's a pretty crazy situation at the moment. Uh, I'm curious what what, um, what's your broad stance on on on Nvidia at the moment.

Jensen has obviously been in a tough space of, you know, being being an American, uh, yet at the same time needing to represent the interests of Nvidia and its shareholders. China is a massive market.

Um and uh you know do you do you imagine that uh everybody just needs to move on from from China and and focus on servicing demand in the US and and our and our allies or or how do you think uh it should be approached at this point? Yeah, it's a tough situation for Nvidia.

I think the line that they use is, you know, if you set the uh speed limit at 60 and we're going at 55, that should be fine, right?

like we're just trying to stick by the parameters that you set in these export controls and sell chips that are compliant and then you have the government sort of uh you know making these big changes to sort of after Nvidia has designed these chips and is like ready to start selling to the market making these big changes that like prevent their market access overall.

Uh so you know if I'm a big Nvidia shareholder I'm like pretty annoyed about this right okay let's uh let's have like better proactive policym around this stuff like I'm pretty sympathetic with that perspective um but yeah I think realistically uh you know Nvidia is a publicly traded company that their revenue they want to sell chips into China they're not sort of seeing the same national security concerns that a lot of people in Washington are seeing and to be frank a lot of people in the AI industry uh directly as well um whether yeah I think there's some question marks around whether they have been fully responsible with sort of doing their due diligence for uh shipments that are going into China and ensuring that like control chips aren't being smuggled uh and that they're sort of complying with the letter of the law.

Um yeah, overall I think we just need clearer policym ahead of time so it's easier for them to make these decisions. Do you have a sense of what the black market for GPUs looks like? Um I imagine you know you guys made a recommendation to add sort of like geoloccation to these high performance chips.

Um and I'm curious if you have any insight into how um chips are are flowing around the world uh in a way that that may not be aligned to your we've been tracking this for the past couple of years. So we put out a report in late 2023 that sort of looked at this question of okay we've got these chip export controls now.

uh what sort of scale of smuggling should we expect? And at the time based on sort of news reports mainly the sort of cases that were being surfaced were just sort of singledigit number of GPUs at the time. So at most sort of like you know a single server like eight GPUs being smuggled at once.

Um and now uh sort of like a bunch of reports came out last year that this is now just a huge industry. Um some estimates is there sort of around 100,000 GPUs went into China um last year control GPUs. Some people think it's much more.

there's individual cases that have been uncovered where there's over 20,000 GPUs being smuggled at once. So, this is some pretty serious shipment sizes like going into the hundreds of millions to billions of dollars. Um, and there's likely a lot we just don't know.

Um, especially for these large orders which are sort of happening through professional operations behind the scenes. Um, and in as much as it is happening, this, you know, shouldn't be a surprise. You know, China has a long history of smuggling controlled goods despite sanctions and export controls.

And you know, it's proven that they were instrumental in helping to smuggle ships to Russia over the last two years to support um the war in Ukraine despite sort of US sanctions. So, this would be like pretty on brand with the China playbook for this kind of thing.

Uh how far behind do you think that China's domestic chip uh chips really are? Right?

You know, we we just did an entire backstory on on Smick and Huawei and things like that for for our audience, but uh and we had Aaron get on the show yesterday talking about how the gap um the gap in capabilities is maybe not as significant as companies like Nvidia or TSMC would would um have people believe, but um I I' I'd love to get a sense um from you of how how wide the gap the capability gap is.

Yeah. So I think there's two dimensions to this. First is the quality of the chips themselves. Um so if we look at that you know how how many flops you were able to thing per second that are coming out now sort of three to four years behind the best stuff that's available from Nvidia.

Um but you know that has been a um move towards catching up over time. I think export controls are the one thing that could reverse that trend is these have only really just become effective over the last year or so.

Um the other and this is sort of like a really important factor is not just how good each chip is but how many they're able to produce.

So you know in in AI especially it's a it's a not so much a um matter of quality of chip that that is important it's also very much quantity like how many you able to produce and there it's where they're really constrained.

Um, so being able to produce, you know, high-end chips by the sort of millions like TSMC is able to do is a very serious manufacturing operation and China doesn't have that level of scalability and because we have export controls across the stack with tooling and fabs as well, that is one of their key bottlenecks.

Uh, what do you think Chinese state investment into semiconductors looks like today? It's obviously they've had 14 separate five-year plans, I believe, around Yeah. uh around the industry.

Uh so at this you know at this point uh I don't know how many hundreds of billions or you know who who knows even uh dollars have gone into the industry but I imagine this is you know one of the you know will just continue to be more and more important every single year. Yeah.

So the way this has been desal to revitalize the American domestic semiconductor manufacturing industry is being over time. Uh China is doing this this sort of equivalent amount of funding roughly every year um to support it domestic industry. Uh Tim, we we keep uh we keep losing you, by the way.

You're kind of cutting in and out. Um I maybe they're maybe the cyber attack has shifted onto your personal mobile device. They really don't they really don't want Tim talking right now. Uh because he's dropping uh dropping bombs. Um but could you repeat I think we have you now.

Could you repeat uh what what you said your potential estimate was chip and science act in the states uh this big landmark piece of legislation to re revitalize their semiconductor industry here that was about 52 billion as sort of a once-off package that's being distributed over a few years uh best sources I've had say China is distributing like this equivalent amount of money roughly every year to their own semiconductor industry um how do you think Huawei is uh you know reacting to the tariffs.

Uh they're a private company, so I feel like we have a lot less just information on on on a lot of things.

But, uh are they, you know, approaching some of these other companies, you know, nations that are facing tariffs, uh specifically with kind of like new OP, you know, basically trying to take advantage of the situation. Yeah. I would be doing in this situation.

So, similar to how Huawei managed to outmaneuver the United States with 5G, installing base stations all over the world and sort of surpassing the US in that technology.

Uh, a similar play that they could be executing now is trying to do the same thing on data centers, especially if US equivalents are becoming much more expensive. But now they don't have comparable technologies on the AI chip side, but they certainly do on the lower end stuff.

um you know, Alibaba Cloud, um like Huawei, they all have their own data center offerings that they're sort of pretty aggressively courting the global market with. Um makes a ton of sense.

What what else uh what else are you thinking about right now in terms of uh you know, now now that the news came out that the order is being stalled, uh what do you think the the admin should be focused on, you know, over the next uh weeks and and months? Yeah.

So I think the one big lesson from all of this is that we keep making the same mistakes in the sense we keep acting too late to really be effective. The thing with export controls is you want to act early because there's always the risk of things like stockpiling.

Like if there's a room that exports controls are coming and they don't happen for another six months. Uh it just gives a huge opportunity for companies to take preventative action and you can completely ruin the effectiveness of what you're trying to do. This has already happened with HBM memory.

So like high-end memory that's really crucial in AI chips. The rumor was leaked that uh controls for this stuff was coming and Huawei reportedly stockpiled around 2 million chips worth of this um before the new rules came into place.

Um we've also been caught blindsided, you know, by Deepseek finding out that they trained their models with US chips that were sold before the restrictions and policy makers being like, "Wait, what's going on here? " uh the sort of smick when they came out with their 7 nanometer uh uh breakthrough in 2023.

It turns out they were using secondhand DUV machines which we have since controlled but they already got their hands on them. There's been a bunch of cases like this. Uh and really yeah we've just been much way more reactive than we need to be.

I think if you just had a team within government who is technically competent and actually see and what's really important going to be really important for AI technology in six months you could actually be doing much better productive policym.

So, one thing we're advocating for is just setting up this kind of situational situational awareness capacity within government so we can get get it right next time. Yeah, that makes a ton of sense. Um, well, we should have you back on when we have uh video internet is not under attack.

Uh, apparently Google Meet is down as well. Uh, I don't know exactly what's going on, but story still developing. But I mean, thank you so much for taking the time. Uh we really I really do want to have you come back for uh a much deeper dive into all of this because it's it's fascinating on the calendar today.

Absolutely. We can do this properly. But thank you for uh making it work Tim and uh thank you for you know following the situation so closely and um trying to get us on the right path here. No worries guys. Catch you later. Cheers. Talk to you later. Bye. Awesome. Uh this is crazy. Everything is down.

Everything is down. uh down detector is showing that not only is Zoom down but uh uh Google Meet is down as well. I thought we could shift over. Um but uh we will have to figure it out. Let me uh message the rest of our guests. Uh zoom down. BTW uh Google Hangouts to working on a fix. Uh pretty pretty crazy times.

I I I don't know if there's anyone actually reporting on what's going on. All I'm seeing is that apparently Well, I have some reporting talks to buy winter. Yeah, I was about to say we should just talk about that.

Maybe we just do the timeline um and we and we and we get the we get the guests to come back uh tomorrow and just do a stacked meeting because uh it seems like we are absolutely under attack. Yeah. Uh and hopefully hopefully it resolves tomorrow. Um, but we had uh we had two interesting companies coming on.

Captions uh is launching some stuff today. That's the app that uh actually Ev Randall over at uh Kleiner uh invested. Absolutely. And uh I use that app all the time. Anytime you see me post a video that has some captions on the bottom. Always captions app.

I love uh and then Deck announced a big fund raise today, something like $12 million. uh and we were excited to talk to them, but instead we might just talk about the timeline because talk about the timeline's working. Um so yes, uh this is a scoop from uh Bloomberg.

Uh OpenAI is in talks to buy coding app Windsurf, formerly known as Kodium, for around $3 billion per sources familiar. It would be OpenAI's largest acquisition to date. They haven't been super inquisitive, but uh it makes sense.

It's like uh if they need a rapper in every single space, maybe they should just go and acquire them. Uh we've been hearing rumblings about like a code first model, something that's really really trained.

It seems like uh opening eye on the API side has become at least this is the vibe on X is like oh if you're going to use cursor you're going to use cloud, right? And and for most consumers there's still yeah chatt app I use for 40 or whatever.

Um but for the developers who really care about the the quality of the code, that little extra push is what's valuable.

And so um having both a model that's trained by OpenAI to be specifically good at programming and then also having something that's popular like uh you know the wrapper built on top of it, all the integrations that you get from Windsurf uh makes a ton of sense. So uh Zack Dit uh over at Wing VC I believe.

Yeah, partner at WingV. I clocked it. Uh good memory. Uh in he says incredible outcome in four years. Um they did a seed three million on 26 post by Green Oaks. Greens another billion dollars for Green Oaks. Another billion dollars for Green Oaks.

Uh yeah, quick plug for uh Patrick latest episode with Neil Meta over at Green Oaks. You can go find the episode finding future S&P 500 companies on Invest Like the Best. Uh go check it out. Yeah. Um uh then they did a series A 25 million on 215 post again by Green Oaks doubling down.

Then Kleiner comes in for the series B 65 on 510 million post and then $150 million series C at 1. 25 post by General Catalyst. So founders and the team likely own 55%. Uh so that's over a billion dollars in liquidity to the founders and the and the team.

And uh at 20% of 3 billion, Green Oaks will be just shy of that billion. So So you know a small size gong for I actually I would assume they own Do you think Oh, maybe they did like stuff. No, but I just assume they they bought uh you know, they continue to buy up. So I don't know. Um I mean 0. 25 of 30 what was it?

Uh three billion, right? It's 750 million. So something around there. Not bad. Not bad for for a $28 million investment. Yeah. To go from 28 million in to 750. Yeah. It's interesting because you often see Green Oaks doing some of these later stage rounds. Yep. They're known for like growth and being very analytical.

But but doing this one from seed. They got in the seed and the series A. You love very impressive. Your jacket is going viral right now. People love it. Yeah. We should just since RV is still up and all the other platforms. Yeah, just do some let's just do more ad reads for the rest of the hour. Back.

Yeah, the back is great. I love Lucy and Ramp. Uh we put we put our sponsors on here. We also just put companies we're affiliated with. We still have uh space. This is going to be uh you're that's legacy. So that's yeah that.

So yeah, I mean if you're in the audience and you got some DAO, maybe you want to get on the jacket. Now's the time to email Jordy. Get get pretty much out of inventory. Get out your ramp card and uh sponsor us. Uh we are 100% corporatebacked. By the way, just so you know, you'll never ask our audience for a dollar.

We're not audience captured. We do what's in the interest of the corporations that sponsor us. Uh independent media is dead. Corporate media, as I call it, is the future. That's right. Free media for all.

Anyway, speaking of corporate uh media, the media around Blue Origin, we should go back to this because we didn't get to finish the conspiracy theory. Can we play the video? Yeah, we can probably play the video if it's up. We'll have the we'll have the team try and pull that up.

But, uh, so the door opens to the Blue Origin capsule as it descends. Uh, then they say, "Hey, don't open the door. Close it. " And then finally they open it up and everyone comes out. And so this was seen as proof that the mission was faked. Um and uh there is a conspiracy theory here.

But my conspiracy is a little bit more anodine. Uh it's that they wanted all the cameras to be turned on before they did the big reveal of like getting out of the spaceship. And so they open the door. Yeah. You can see they open the door and then they close the door.

Not yet because you know NBC is here but CBS hasn't arrived yet with the truck because these capsules they come down you can't just be right there. They don't know exactly where they're going to land in the desert.

So if you watch the zoomed out feed at the very beginning I think at the very beginning of this you'll see um all the trucks are driving. They're they're they're scrambling to get to the landing site. Bezos was clearly just the first one in the chase vehicle to be like I'm going to get there first.

Make sure my wife's okay. You know check in on the girls. the girls the girls chat as they say. The girls group chat. Yeah, he sent the group chat to space. Um, sent the group chat to space. Sent the group chat to space. Uh, but Oh, Bezos falling down the hole. But he gets right back up. He gets right back up.

That is crazy. Oh my gosh. That is a big drop. I hope he's okay. Big drop. I hope he's okay. Um, but isn't that what Brian Johnson is deathly afraid of? Like a fall. Oh, really? I didn't know that. Yeah. I thought he does leg press. I think it's because he doesn't do that much leg press, so he's worried about it.

So, Andrew uh Coat Cotay says, "Can any space nerds help me understand how the Blue Origin capsule came back down without any re-entry burns? Was this a trajectory thing or some kind of superior material used Falcon crew capsule for comparison? " So, Scott Manley jumps in.

He says, "It only flew at Mach 3 fast enough to reach space for a moment. " And to be clear, Andrew is trolling. He knows what's going on here. He's baiting the timeline to make it be like, "Oh, it's a conspiracy.

" But he he everyone's like laughing like this is like S tier bait because like like he knows that when like this mission I mean they've done these missions like all the time and they never get burnt because they don't go that fast. And so that's what Scott's enumerating here. He's he's laying it out.

Uh Scott says it only flew at Mach 3. Fast enough to reach space for a moment, not fast enough to stay in space. For that you need Mach 25. As you can imagine, there's a huge difference in the kinetic energy between those speeds.

And so the real conspiracy theory here is that this was a, you know, a fun space tourism thing. It's it's marketing for Blue Origin. They did a great job of puzzling together what is the crew that would probably go the most viral. So, it's not just Lauren Sanchez because that's probably not newsworthy.

You got to put Gail King on there so CBS has bought in. But then also throw in Katy Perry because you throw in Katy Perry, you you create an even wider audience of, you know, she's a global celebrity. Everyone needs to know about Katy Perry going to space and it just seems like it's a lot of fun.

So I would definitely go on one of these. I think this seems like a really fun trip. It's only a quarter million dollars. So I was thinking they're basically given space trips away at point.

They are they are I was thinking like there might be a trend of like send the girls chat the group chat to do something extravagant like this. We might be seeing this with other tech people. Maybe you don't own a space company, but maybe you just uh Yeah, guys out there, if you want to have a boy's weekend, Yeah.

send wife and her friends to space. To space, but that then they're only uh up there for 12 minutes. Yeah. 12 minutes. I mean, still needs to be like uh how about uh you know, cross the ocean on a sailboat. Then the boys can go do the Nurburg Rand for a month. Get them a wander. I don't know. Um, this was funny.

Jeff Tang highlighted uh, so this this icon uh, Ken Ken Davidson on the phone uh, before on the show, he is a funny guy. He put out a competitor landing page uh, with a comparison page and it was pretty hilarious as as every com competitive uh, sort of comparison page should have.

Icon checks every single box, including 100% money back guarantee, script writing and creative briefs, audience research, video ads, static ads, creative storage, asset tagging and splitting, lip sync, UGC videos, creative analytics, competitor ad spying, ad manager upload, tier one investors, tier one engineering team, 7-day work week.

Um, and he did not give OpenAI credit for tier one investors or tier one engineering team, which is brutal because they have same investors. Well, and they're probably Icon's probably using various chat GBT. Oh, for sure. For sure. And also like Yeah, it's like Founders Fund is in open is in OpenAI and Icon.

And so it's just such bait to be like, yeah, we're the only one with tier one investors. Like no one else in the AI category has tier one investors. That whole AI thing like this is true in other categories. Like we talked about general matter.

Like there aren't a lot of Let's pull up the website too because I actually think it's icon. I think the icon. me website is like fascinating in general. Pull pull it up real quick.

So, so the story here is that is that obviously Jeff Tang is very much like trolling trolling Kenan being like this is a ridiculous comparison. Uh I love comparison pages, competitor comparison pages because this is hilarious. And then Kenan says, "What a lame comparison chart. This seems to be the website.

" And so he's he's like dunking on his own thing just to link to his own website because he just wants the traffic. He's like a total like just drive the bottom line. Doesn't matter if it's embarrassing. Just make the money. And uh Jeff Tang quote. Okay, so look at this website.

Oh my god, this is the most conversion optimized. This is hilarious. It's actually so It's actually so chaotic that I actually investors tier one team tier one engineers. I've never even heard the term tier one engineers before. Um and but look how optimized this is. Wow. Us versus them. There it is.

Hey, hey, they changed it. They updated it. They updated it. They put OpenAI tier one investors. Wow. They must have gotten some. And it's interesting. They're selling the product as you can become a founding advisor. What? To Icon. What? It's like it's like almost like MLM coded. It's like a pyramid scheme now.

I don't know. I don't know. But they also sell they they sell you just sign up for this, right? You can go and app like go on the go on the pricing page. Just click uh I'll send it in the chat. Yeah. This is the most chaotic SAS website I've ever seen. It's honestly funny. This guy wild.

Well, look at the pricing page on here. They basically built it like a They basically built it like an e-commerce. Yeah, it looks like checkout on e-commerce. Wow. Limited offer. Only four spots left. Ends 418. No, no, no. For a SAS product. And scroll down. They sell a premium banana, which is Wow.

I guess a banana you can just check out with Apple Pay. This is so hilarious. I mean, at this point, it seems like with him with him like retweeting these and engaging these, like he's in on the joke. And so, it's gotten way funnier. Like, if this was just, oh, he's really really serious about all this stuff.

Like, he really thinks that he has two invest OpenAI doesn't. They got Saquon Barkley in here, too. Did they? Who knows? That's the thing with like this comedy landing page. I don't know what's real anymore. Like, if Well, hopefully the product is real. Hopefully the product is real.

We I mean, we did talk to the Ridge guys. They said they were using it. It was pretty useful. I think obviously it's an early stage company. Uh Kenan is trying to get some eyeballs, get some attention, and just kind of break through the incredibly crowded market of like AI tools, right?

And so you got to make some noise to be able to just get get through. Um but uh we'll have to put that whole site in the truth zone, figure out if Saquon's really in Icon's uh icon. So far to our knowledge he's in and ramp. That's what he's a two company portfolio. Yeah.

But he could be ripping checks all over the place, all over the valley. We'll have to figure it out. But speaking of ramp, time is money. Save both. Easy to use corporate cards, bill payments, accounting, and a whole lot more all in one place. Go to ramp. com. Switch your business to ramp. com.

Switch your business to RAMP. Um, and uh Ryan Peterson had a good post. He said uh a 1% daily improvement compounds over a year to a 37x improvement over the entire year. But I think he's thinking too small. So I asked Chad GPT, what does a 100% daily improvement compound to over a year?

If you compound by 100% every day, you're doubling each day. That's exponential growth, baby. That's right. So if you start with one as a base and double it every day for a year, by the end you will be 1. 52 * 10 to the 109th power better than you were at the start of the year. That's one followed by 109 zeros or 1.

52 novem dicilion in the short scale doubling every year every day for a year gives you 1 golden retriever 1. 52 novem dilion times your original value. This is absurdly huge. A classic example of why compounding growth explodes with even modest periods. You won't understand.

So, if you're thinking about improving 1% daily, just do 100% daily and then you'll just be so much better off. So, risk. It's great. Soundboard is not broken. They haven't cacked the soundboard. Yeah, it's been our most chaotic show to date. It has it. We're facing cyber attacks. We have a new soundboard. Yep.

Um, Zoom is down. Google Hangout is down. Is down. I'm not in a suit. I think it's a more relaxed show. We're just having fun with it. It's okay. It's okay. The show must go on. That's the most important thing. Well, you know what service is not down, John? Getbbezzle. com. That's right.

And the bezel concier, they're working overnight. They're working overtime. Uh, your bezel concierge is available now to source any watch on the planet. Seriously, any watch. Uh, pick up a, you know, they're not going to they're they're not going to hack your GMT Master Batman.

They're not going to they're not going to hack your Submariner. Yeah. They're not going to hack your Paul Newman daytime. I'd like to see them try. I'd like to see them try.

It's gonna keep your your PC Philippe Cubitus is still going to tell time after the AI overlords have destroyed all of the data center infrastructure in this country. We'll see. Um should we pull up this post from uh Josh Kushner? Yeah, sure. He says uh so we talked about this yesterday. Yes. Yes.

Somebody asked, "Do VC funds invest in competitive companies? Kyle Harrison pulled up a chart that said evidence would point to yes, showing how many of the big venture capital firms have invested in multiple uh or at least a couple foundation model companies.

And uh Kushner over at Thrive Capital has uh I don't even think he's on the chart because he's only invested one um and he's invested in a very important one and he likes I think staying off of charts like this for the most part.

I think uh Benchmark or sorry um uh Jeff Lewis and the team have uh also just gone all in on OpenAI. So there's still quite a few.

Um but again I I think that this is an evolution of venture capital and it is a a natural evolution that firms will invest in competitive companies and I do think that firms generally do a great job of keeping firewalls up and not you know sharing information that would be you know damaging to any any of the individual companies.

Yeah, it's interesting.

I mean I I I think that obviously there are secrets in these foundation labs and there are um algorithms and just even just you know the path down the R&D tech tree like reasoning models came from Ilioskver working on um Qstar at OpenAI uh when he leaves to start SSI you know that there's going to be a reasoning component a reinforcement learning component on top of LLM there's going to be self-play play like this worked well at OpenAI.

He takes that and you can't patent that idea of like a reasoning model or applying reinforcement learning on top of LLMs.

And so not only can you not control that or patent that or keep that out of the hands of uh Grock or Thinking Machines or Mistral or any of these other companies, but um I don't think that the VCs are the vector for these types of leaks.

I think it's like the AI happy hours and the parties in San Francisco where people are talking and they're like, "Oh, did you see this paper that someone just published like or somebody, you know, reimplemented the open?

" I think this is founders in the early part of their career or I I've seen them be worried to pitch a firm because they were invested in something that was competitive or adjacent to what they're doing. And the reality is is your competitors are going to find out about what you're doing generally. Yes.

And that's not necessarily your mo. It's rarely remote. Like the Coca-Cola formula rarely applies.

But at the same time, like it is reasonable to say if I'm if I'm really partnering with a firm and they're going to take 20, 30, they're going to build a big position, they're going to be on my board, do I really want them to have this weird like incentive? Yeah.

This weird conflict where they they they find a great VP of engineering and they're like, "Well, I could send it to you or I could send it to your direct competitor that I'm also invested in. " uh like any sort of like value ad does get does kind of get spread thin.

Um but I guess it I guess it gets back to this idea of like how commoditized is the foundation model layer and how different will these companies look over over time.

I mean you know Andre here is in OpenAI and XAI those two companies seem at each other's throats like like OpenAI is launching a X clone XAI bought X and so they're like really direct competitors.

Um but at the same time like if OpenAI is is buying wind surf and developing more consumer products and then SSI is saying hey we're staying out of that entirely and then Mistral seems to be focused more on B2B and government implementation and anthropic is very much on like the the API side for example um you could imagine that the that foundation model just becomes like cloud and at that point like you wouldn't be that upset about investing in like a Salesforce course and a pager duty and a and a Twilio like these are very different companies over time um versus if they it's clear the end state is not just 10 yes you know chat apps yeah so so it's like the foundation model layer may commoditize but that doesn't mean that the companies are going to commoditize like and Uber are true true supplements to each other like they are they are not complimentary and in any way you're either spending money on lift or Uber they're very similar experiences now one might be way better than the other businesses For a long time it was capital because people would open both apps and see which is faster which is cheaper.

Yep. Yep. Yep. But but if if open AI winds up looking completely different than thinking machines or SSI or ML in a few years that like there could be winning companies that are very different but just built on the same you know foundational technology. Yeah.

Um uh and I and I assume that we'll see even some of the foundation model companies merge too. I mean we're yet to see that because they're still extremely wellunded. Yeah. Didn't was it Sequoia that did Intel and Nvidia? I mean they're like a decade apart but in theory they're both like chip companies, right?

But they operate in very different spaces and they have very different models now. Fabulous versus integrated. Um, but uh, you know, there are some funds that have stayed the course. Uh, Josh Kushner over at Thrive is a serial monogamist.

Delian also chimes in to say, "Sorry, can someone explain to me why we're on this if we only have one arrow? " Doesn't that mean we're not spraying? Managed to say off the list, founders fund was not as likely was clocked on there. Yeah.

But it is it is weird that like this is I mean I guess it's just I I guess this original uh this original diagram was just to show like what funds are in what uh are in what foundation model companies.

Um but then very quickly became you know oh they're investing in multiple but and then it makes less sense for for FF to be on there. Um anyway did you see Black Flag launched today? The founder sent me this. Uh it's an incubator for uh hard tech companies and they are split across a variety of uh different industries.

John, I'm actually launching an incubator for incubators. Oh yeah. So I'm going to reach out to these guys and see if they want to be a part of my incubator incubator. The meta incubator. I mean I think it's just being called the LP.

But uh Harpoon Ventures is the is the the main firm behind this and they're in collaboration with Shieldcap VC who we've talked to before and Inutel is an investor which is the CIA's venture fund. So that's uh pretty pretty fun and good news.

I mean, I think that there's a lot of a lot of folks that, you know, I I think you could say like YC is the incubator for defense tech and YC has been open to this, but at the same time, like there might be some benefits to having an incubator for like specific hard tech companies.

So, you know, you maybe advertise the cost of some like CNC machine or some like, you know, shared hard workspace or some capex stuff, but I don't know if they're planning to do that. No, I think it's awesome. I just want to see just more of this stuff is great, more competition.

Um, you know, companies are now going through multiple programs. I mean, we talked to uh Alli and it seemed like some of those companies had done uh had done NEO and then also done teal fellowships and then fellowship fellows and then they also raised YC. Exactly. It's like Yeah, there's an overlap of these.

There's just like at the end of the day it's just like No. for for Black Flag. I don't know anything about it other than this graphic.

But um what's cool um about uh specifically defense and aerospace and anything government related is that there's an entire they can provide a very clear benefit to founders that maybe don't have a defense aerospace background. Yeah.

but uh can sort of they can really accelerate that knowledge and and networks and things like that and allow people that maybe we're building in SAS to come in and and play in a new uh arena. So yeah.

Uh well, if you uh wind up with uh 20K from one of these incubators and it's no strings attached money, why not throw it in public? That's only NEO, by the way.

Neo is the one that uh I mean, we've heard stories about Teal Fellows investing their their money if they're profitable because it's it's it's kind of no strange not recommending venture dollars. These are grant basically grants. But uh I mean, why don't we have an accelerator for hedge funds?

Like you go in, they stake you. I know they exist, but like you know, you just the first like high frequency trading shop from YC, that's going to be a banger company. Come in. It's like your demo day slide is just like we started with 500K from YC. AUM's 5 million now. We 10x that. There we go.

We got a couple more 10xes. It's a billion dollar company. Get in now. There we go. Here's the LP agreement. We're not We're not doing safes. No safes here. That'd be funny. Uh anyway, uh you know about public. com investing for those who take it seriously.

They got multi-asset investing, industryleading yields, and they're trusted by millions, folks. Um anyway, I think we've been through most of the timeline, most of the great This was great. This was a chaotic show.

I'm excited to figure out why all of our, you know, internet services We need a poly market on this right now. We need a poly market. Why did Zoom and Google Hangouts Meet Video, whatever they're calling it these days, uh why did all the why did all the services go down? Was it a cyber attack? Was it not?

I want to know about it on Poly Market. That's our other sponsor. So, go to Poly Market and check it out. Uh we'll try and get a mark up ASAP. Uh and I just want to say I've I think I've gotten at least 30 individual messages about the TVPN jacket from people are wanting it. So, that's amazing.

Uh we weren't even planning to release uh to sell these, but um should we should we tell them? Uh it's uh it's $9. 99, but financing it will be available. You can buy now pay later. It's uh it's 12 payments of $99 if you want one. So, um this is kind of what the market will bear. Yeah.

And we're going to donate the profits to Allin, which doesn't run No, no, no. We're going to donate the profits to my next watch on bezel. That's right. Okay. Okay. Get bezel. com, folks. They got concieres. They'll find you a great watch. John loves watches so much. He's running two bezel ads on the same show.

But thank you for tuning in, folks. Um, I'm excited for tomorrow's show. We want to talk about who's coming on. We're coming back. We're going to bring on tons of people. We're going to stack it. Maybe we'll do a six hour stream to make up for the lost half hour.

I feel like if you lose a half hour, you owe the fans four hours. That's right. Right. Every half hour, it's four hours. No, I mean, we we we got a great show tomorrow. We got Delian coming on. We got Sham Sankar from Palunteer, CTO of Palanteer. CTO of Palunteer. Uh first Ford deployed engineer.

He's been the company for like 20 years. It's incredible. I I was hanging out with him. It was public company. And I was like, so what's next? Like you're a public company. Like what's your goal? And he was like, "Oh, we got to get into the Fortune 500. " And I was like, "That's an insane goal.

I've never heard anyone like frame it like that. " And then like a couple months later, I had to send him a congrats message because it was like, "They are now officially in the Fortune 500. " And I was like, "That's sick. Congrats.

" Uh, and then we're also going to have some robotics CEOs come on, uh, work on a couple different robotics companies and talk about those. Uh, and of course we'll we'll be bringing you to the news, talk about whatever's top story.

I think we covered the Nvidia stuff pretty well, but I'm sure there'll be more developers. I just realized we completely missed the major story. Figma. Oh yeah, the S1 dropped. Did an S1. Congratulations to everyone who's been working on Figma. Uh, amazing milestone.

It's I mean it is one of those rare companies where it is an overnight success. It's been like what 14 15 years but uh overnight success. Uh but congratulations to Dylan Field and all the folks over at Figma. Uh what a roller coaster ride years in stealth back and forth with Adobe.

Wasn't it like four years basically before they shipped a product or something like that and they were still raising money. It was sheer force of will. Classic overnight success. Classic overnight. love to see it. And we're going to be diving in deeper to Figma. We should tell the whole story.

We should do the whole a whole deep dive on the company. It's a fascinating story. And we'll be at Config. Yes, we will. Um we're going to be uh doing a postgame interviews from Config, interviewing some of the greatest designers in the world, some entrepreneurs, some investors, all sorts of folks in the design world.

That should be fun. Uh I'm I'm newer to the design world than you. Not a uh I'm not a Figma daily driver like you, but I'm excited to learn a lot about it. You're a creative athlete, though. Yeah, creative athlete. Yeah, a little bit. I mean, there's creative stuff all over. Design is in everything in this world.

Anyway, thank you for watching. All right, folks. We will see you tomorrow. Fantastic afternoon. Uh we are going to call up um Tim Cook. Yep. David, uh Daniel Elk from Spotify.

uh try to get these services back online and sundar because Google went down that's unacceptable I think our SLA says no downtime and who's a Zoom CEO again it's Eric Eric you on I just know him by Eric when I text him I just what the hell is going on group chat and we'll say let's get it together guys because let's get it together let's step it up now is the time you can't be slacking off can't be slacking off guys every minute of productivity counts it does um all right have a great afternoon we'll