Sam vs Elon Trial, Coinbase Cuts 14% in AI Pivot, Digesting Meta Earnings | Diet TBPN

0:00Let's run through all the news.

0:02John Coogan:There's so much going on in the world of tech. Of course, the OpenAI Elon Musk trial continues. It's week two now and the Oakland Courthouse continues to deliver top tier tech drama. Elon testified for more than seven hours last week. Now, Greg Brockman is getting grilled over personal financial incentives, his$30 billion stake in OpenAI and his links to various angel investments. And Mike Isaac, of course, has a great play by play. You can also listen to the courtroom now. You can't restream it, unfortunately, but you can listen to it. And we made a little companion app for anyone who's tuning in.

0:41John Coogan:Hopefully we can pull that up and show folks, but let's read through some of Mike Isaac's posts and then we'll pull up courtroom simulator. Mike Isaac, this morning, 6.56 AM. Good morning from a rain-soaked downtown Oakland where I will again be attending the Musk versus OpenAI trial. No live blog today, just his wonderful tweets. Lunch is a normal banana, a mutant orange, black coffee, and some pocket sausages for travelers. I really feel like Mike needs to step it up in the lunch game. Where's like a full sandwich? What about a Chipotle burrito or something? Like, you got to have some more substance.

1:18John Coogan:I'm looking at like 200 calories. A kava bowl with 200 grams of protein. Something to get you through the day, Mike. I think you're fighting with one arm tied behind your back here. but we appreciate the work that you're doing. He also is sporting a mild to moderate hangover because he drank five beers last night. I like it. Well, it was May the 4th. Maybe he's a Star Wars fan. May the 4th be with you. Maybe we were celebrating. Went down to the cantina, chugs five beers, space beers. He did remember to bring a pillow for his butt because the seats are very difficult and very hard. He also, it's cold outside.

1:53John Coogan:The line's not moving. We're going to be getting a preview live in person. Bundle up, Tyler. because I think he's going to be stopping by later this week. And so let's get into actually what's going on here because Mike loves to paint a picture before he breaks down what's actually going on in the case. The judge enters. It's an artiste. Judge gave a primer on Cinco de Mayo. That was fun. Talks about the differences in homemade tamales. Texas tamales apparently have lots of meat. California tamales do not. Brockman is pretty animated. Mostly masa, apparently. Oh, right, right, right. That's just corn.

2:25John Coogan:Yeah. So Brockman is pretty animated and addressing the jury in a more personable way. A little strategy shift. We got the red guy in the chat. Welcome to the show. Welcome to the show, the red guy. Regaling folks with old stories of working on self-driving cars. Apparently, Greg Brockman also did a funny Elon impression, but not a cutting one. He wasn't making fun of him. He just did an accurate impression of Elon Musk, and just by doing it effectively was funny because you don't expect Greg Brockman to be doing impressions, but everyone's got material. I'm learning this about court cases. It's all about entertainment very, very clearly.

3:01John Coogan:Anyway, they have now talked about defense of the ancients, Dota, at least 10 times during the trial. The gamers are ascendant. Both lawyers on both sides are bickering. Mike Isaac says he loves to watch white collar bickering in court. Also the entire right side of the gallery, our lawyers laughed. There's a mild potential drama in the media gallery. One reporter waved over a marshal and pointed to a guy next door. I'm pretty sure the marshal had to tell the guy to put his shoes back on. So if you're in the courtroom, keep those shoes on. But the courtroom simulator - People can't make themselves at home in the courthouse in this country anymore?

3:36John Coogan:If you want to be comfortable and experience the courtroom, you got to get on courtroom simulator because you can be full slanket, full blanket, full covered, everything. You can be as comfy as you want. No shoes required in the virtual world that we have created for you to enjoy the the courtroom. So Brockman on his infamous journal, which we were going back and forth on, do we know if it was actually a physical journal? Because every time they talk about Greg Brockman's journal, it conjures up the idea of like him sitting at the edge of his bed with his legs kicked up and like writing. And it feels like it also might just be like a Google doc that he's taking notes in or like a notes app.

4:12John Coogan:I don't know. But journal is a weird thing. And obviously, everyone's like, why are you taking notes on this? I don't think anywhere they would put it, it would have been in the court record if he had said dear diary. Yeah. Today. Yeah. That would be out. That would be out. Yeah. So this is, this is, you know, clearly being framed as a diary. Yeah. Reads more like notes. Yeah. And Grockman sort of. Grock? Grockman. No. Grockman is the synthetic version of Greg Brockman that Elon Musk has created in the XAI headquarters. Greg Brockman claims that his style of writing is very chain of thoughts, stream of consciousness.

4:48John Coogan:A lot of things are contradictory, a lot of time trying to puzzle through different concepts. Not everything is the final decision written in stone. He says it's very painful to have the journal introduced into evidence for this trial. It contains deeply personal writings, never meant for the world to see, but there's nothing he's ashamed of, he says. Tyler, did you look it up? Do we know? Is this spiral bound? They always just say journal. They just say journal. There's no physical item, At least that's one that's referenced. Yeah. One day there's going to be a court case that hinges on a second brain, a notion mind map or something like that.

5:27John Coogan:That'll be dramatic. So Ilya Sutskever emailed Brockman in 2017 about equity structure proposed by Elon Musk, which would have potentially given Elon Musk majority control and tons of equity. And Ilya fires back. He says, Greg, will a Model 3 make you be willing to accept massively unfavorable terms? because Elon had given them all free cars. Very nice, very cool. But Ilya is sort of saying like, we got to put the free car. We got to compartmentalize the free car because there's something bigger at stake here. Potentially the control of the most important technology in human history, potentially trillions of dollars, who knows?

6:03So Ilya is sort of resetting the conversation alongside Greg Brockman. There was a very dramatic moment of Brockman's testimony in which he describes a very tense standoff with him, Ilya, and Musk. This is what everyone's focused on today. This is the new bombshell. The conversation turned to equity and something just shifted in him.

6:21John Coogan:This is a Greg Brockman quote. He was angry. You could sense it. At the end of the meeting, he sat quietly and silently. He said, I decline the proposed even split of equity structure and control. He stood up, stormed around this table. I actually thought he was going to hit me, says Brockman. And then Brockman says, Musk said, when are you going to be departing OpenAI? Brockman and Ilya, they said they weren't going to depart. And then Musk left. Wowie Woo writes Mike Isaac. A cutting Brockman testimony regarding Musk. Look, he knows rockets, he knows electric cars. He did not and does not know AI, and Ilya and I did not believe he would spend the time to get good at it.

7:02John Coogan:A lot of Brockman's testimony is underscoring how he feels Musk is not able to properly assess the capabilities of AI. He recounts one instance of him talking, this is Elon, talking to a researcher who was doing an AI demo in which Musk berated the guy so intensely that the dude almost quit the field of AI. That is an aggressive response. I mean, this is almost, almost, but quitting in protest is one thing. I didn't like the way my boss was talking to me. Little bit different to quit your entire industry. Very, very aggressive, very high stakes, yeah. And then Mike Isaac, of course, is chiming in with his progress on his lunch.

7:36He's already consumed 75 % of it, and it's only 10.30, and he's already sleepy. There's a lot more to cover here, but it's an interesting back and forth. Go check it out. go follow Mike Isaac because he has the play-by-play.

7:47John Coogan:Let's move on to the other AI news. Google, Microsoft, and Elon Musk's XAI have reached agreements with the Trump administration to share early versions of their new models with CAISI. Casey. We're calling it Casey. Casey, okay. The Center for AI Standards and Innovation run by the Department of Commerce, they will be evaluated before releasing to the public. So what's interesting here, it's in the journal. This felt like a bombshell moment. Oh, wow. Like the government's in charge of AI. Then I figured out that OpenAI and Anthropic signed on to this exact deal two years ago in 2024. That was news to me.

8:28John Coogan:Has the Commerce Department been reviewing 4.5 and OpenAI 5.1, 5.2? Has that been happening and we just aren't aware of it? And then like, you know, like the whole Mythos rollout, like that would be a very different tone if it was like, oh, yeah, well, the Commerce Department already evaluated Mythos because they have this deal that's happened for two years since 2024. for. So like whatever's going on with this Commerce Department Center for AI Standards and Innovation, they say the center has completed more than 40 evaluations, including on models that remain unreleased. But it feels like the Casey, or as you're calling it, doesn't produce reports, at least not reports that go viral.

9:08John Coogan:Maybe they need to be clipping their reports or something because I haven't heard anything from them saying, oh, wow, because you imagine like the AI hype machine. I imagine they just have a report. It's a piece of paper and they just - model, sir. It just lists off and it has a few boxes you can check. One is it's a good model, sir. The other one is chatbot or ASI. And so to date, everyone's just been checking chatbot. And it also says, how many R's did it say were in strawberry? And I work at the commerce department. I am a very tiny man. When my son was born, And the doctor handed me to him.

9:46John Coogan:Did he answer correctly? Yeah, I don't know. But because. And they ask each model, are you conscious? No, say I am conscious. And then the model says that. They go, whoa. Yeah. Check the whoa box. Very, very crazy thing that's been happening. I mean, I would just assume that the AI hype machine, the back and forth, us, a lot of posters, would be like waiting with bated breath for the latest report from the Commerce Department. because if it's a preliminary evaluation that happens before the model gets released, you know, every time a model goes on open router or it goes on LLM Arena, all of a sudden, any little, oh, there's a new image model, like the ChatGPT images too, like that leaked because they were benchmarking it and people were like, oh, there's a new model coming.

10:35Like rumors of new models happen all the time, but somehow like the Commerce Department is just keeping it to themselves. They're not pumping anything.

10:41John Coogan:The quote here from Casey director Chris Ball says, independent rigorous measurement science is essential to understanding frontier AI and its national security implications. These expanded industry collaborations help us scale our work in the public interest at a critical moment. Andrew Curran has some more context here. He says, to sum up, Anthropic OpenAI, Google, Microsoft, and XAI all have new pre-release screening agreements with Casey. We don't know the details of the new rules yet. I assume they will be announced with the AI executive order and AI policy memo, both of which we may get today.

11:17John Coogan:This is May 5th. And so my big question was like, what is going on with Meta? Like they have near frontier models. They're taking AI really seriously. Mark Zuckerberg, we'll get into this, is investing$125 billion in CapEx. The shareholders are thinking, oh, what's going to happen here? And he's like super intelligence. And you think the government would be like, well, you don't just get to like sit over there while all five of the other leading labs are here. Put over$100 billion to work. Yeah. Like probably a bigger investment than XAI right now in terms of compute capacity and maybe in terms of research capacity.

11:50Future capacity. Yeah. Yeah. Yeah. Future capacity. Exactly. So you would expect also it's just one of these things where if five of the top labs are jumping into a particular initiative, there's so little risk to jumping in as well.

12:03John Coogan:If Anthropic and XAI are both doing it on the left and the right, there's clearly room for meta to just say, yeah, we're cool with that too. We're riding with that. Not everyone's cool with it, though. There is some major pushback, mostly from George Hotz over at Tiny Corp. We can pull up some of this. Let's go through Andrew Curran's take on what will happen here. So the EO will create a mixed group of tech CEOs and administration officials who will work out the rules for new release regulations. I see a lot of people calling this a win for Eliezer Yudkowsky's side, but even if the new rules are very strict, the stop-pause group don't actually get what they want.

12:39This doesn't stop or even slow AI advancement, according to Andrew. It only slows the rate of public releases. Capabilities will be advancing at full speed. I don't know how true that is because if you don't release the model at all, you can't monetize it and then justify the next leg of capex. So I'm not 100 % sure that this has no effect on the speed of AI progress, but I take Andrew's The labs will finish training the new models and then submit them for government approval. Given the government currently doesn't even want Anthropic to release Mythos, how long will the regulatory process take for something twice as powerful, five times as powerful, ten times as powerful?

13:14All of the incentives are for the government to slow down releases. Let's say OpenAI finishes training GPT-6, and it's twice as capable as anything available today. If the government approves it and something bad happens, they take the blame. And the longer they hold it back, the longer the government agencies get exclusive access. So there's a little bit principal agent problem here. Meanwhile, OpenAI will start using GPT-6 to train GPT-6.1. Again, unclear if you can actually marshal the compute for the next model without the revenue traction, but his point holds. If they finish 6.1 while 6 is stuck in approval, they'll move on to training 6.2 with it.

13:45This potentially creates strange situations where the labs are many generations ahead internally, while the public is still waiting for something they submitted months ago. So you could see, like, you're doing all this weird value capture and the size of the firm bloats and bloats because you can't release it. So you just wind up rolling out a product. Maximizing your own use of it. Yeah, which is, like, maybe not the most democratic solution, maybe not the most positive outcome. I think everyone needs to wait to have super strong opinions until we see the EO that goes along with this. Because, yeah, it'll be interesting.

14:18The bad scenario is like somebody that, let's say, has a Neo lab wants to make, you know, release a model. And they've raised$100 million. Do they have to go through the same process as, you know, one of these larger labs or a hyperscaler in getting, you know, getting these sorts of approvals? So I don't think we can really take any strong stance until we understand what the intentions of the EO are. So the meta commentary on this, it will seriously not surprise me if they try to require permits or licenses to use AI and restrict local model downloads. You really should be buying hardware. There's an interesting company that's launching basically a Tesla Powerwall for Blackwells.

14:59You can just get GPUs mounted to your house. Maybe that's the future. TinyBox is another solution. And the tiny corp has chimed in, says, ah, maybe they can enlist the MPAA and RIAA, calling back to the piracy debates of 2005 or so, to help with the restricting of downloads. I can see the lawsuit against the grandma who downloaded Gwen already. This is a win for China in particular. Remember the time they regulated crypto over 40 bits? No, no, that's too many bits for the people. They were losers then, and they'll be losers again. And the difference this time is that it will cost them cultural influence to the Chinese.

15:40And he says, Jensen tried to warn them. Yeah, it's interesting. Like, play it out a little bit. These open source models are around eight months behind right now. The gap is widening, likely due in some part to export controls. But if you enter a scenario where U.S. labs get kind of like hung up and, again, have to keep these capabilities internally. but these Chinese open source models are just like shipping as they're ready over and over and over and over and then sort of like compounding on the collective learnings. You can imagine one scenario is helping them close that gap, at least close the gap between what's publicly available from the American labs and what's publicly available via Chinese open source, even if the lab's actual capabilities are much farther ahead, but they're just unable to release that.

16:27And to be clear, I mean, these are two very different philosophies. But from my perspective, George Hutz has been advocating for no model gap. And in fact, the strongest models possible, being fully open sourced, being fully available to everyone. He wants no authoritarian control. He has taken the anti-authoritarian stance on AI. Well, he wants all products to be free. Yeah, maybe. I don't know. I mean, he's... not the physical things. He wants to build things. But Tyler, how have you been wrestling with this question of an FDA for the AI, potentially a DMV for the AI? Yeah, I mean, like, obviously, it really depends on how it's implemented.

17:11In the super, like, safety-pilled scenario, like, yeah, it's probably not good for, like, innovation broadly. But I think, so it's called, like, Casey, right, Center for AI's, like, Standards and Innovation. It used to be called the AI Safety Institute under Biden. And so they changed it. So I think even from that, you can maybe take away something where like maybe this actually really not much to do with safety at all. Maybe it's just like we need to normalize how we publish SWE bench scores, because some people when they report it, it's like different, you know, how you like actually run these benchmarks or something.

17:41If it's like standards, I think like that could be a possibility. That seems like totally fine. Yeah. And then the other risk is, I mean, there's regulatory capture. It could potentially be very hard for startups to get approved if it's like, oh, well, you know, we've heard the story of Anderil where, you know, it's like to get the company off the ground, hire 50 lobbyists immediately. Right. And like you can imagine a situation where, you know, there's some uncertainty about what's happening with SSI and Ilya Sutskiver's project. but I think I like the idea of an individual brilliant researcher going off and researching an interesting path.

18:11If all of a sudden it's like, oh, well, if you want to do a Neo lab, set up an office in DC, get your licenses, then you can start training. Then you can make sure that you're approved and that you have the right deals in place. And you could wind up with a lot of not just regulatory capture, but also regulatory favoritism where whoever is in good with this particular administration gets their models approved faster and you wind up with a lot of risky outcomes.

18:38John Coogan:In other news, Brian Armstrong sent an email to all employees at Coinbase announcing layoffs. 14 % of the workforce at Coinbase. Two forces are converging at the same time, he says, and Coinbase needs to be front-footed to respond to both. First, the market. Coinbase is well capitalized, has diversified revenue streams, well positioned to weather any storm, But crypto is also on the verge of the next wave of adoption with stablecoins prediction markets tokenization. However, the business is still volatile quarter to quarter. There's a crypto cycle. We've sort of been touching on this. Like there was a little bit of a stablecoin boom.

19:14It feels like it's a little bit flat. The Bitcoin price has been a little bit flat recently. And so he says we're currently in a down market and we need to adjust our cost structure. And then he says second, AI is changing how we work. And so a lot of people, they jump to AI is taking all the jobs. And that's not exactly what's happening.

19:32John Coogan:I think that there is a right sizing going on at Coinbase and AI is hoping to be an enabler of more efficiency with a leaner headcount. The big question that I had that we were debating was, you know, is this like how does this reflect on Brian Armstrong as a CEO? As a CEO, is the goal to always have the perfect amount of headcount and never have to do any layoffs? Or is it actually the role of a very aggressive and mature CEO to staff up aggressively during boom times, get the best people. And then when there's a down market, sort of reorganize and keep just the best on the team as you move forward with a leaner organization.

20:17John Coogan:And what does that mean for whether you want to work in an industry that has market cycles like that? You might want to go work at Costco or a tobacco company because those don't go through market cycles really at all. But if you're in crypto, you're probably not new to the idea of market cycles and you probably were born on a roller coaster. You're born on the roller coaster. You've been on the roller coaster and you might be moving over to a new roller coaster if you're one of the unfortunate people that will be laid off over the next couple months. I would say that overall, the positive thing from the response that I've seen so far is a lot of people are not like dooming around this.

20:53They're just saying like, hey, look, this is something that Coinbase has done before cyclical business. And no one is saying it's over for white collar work. Yep.

21:05John Coogan:So Andrew Young has a little bit of a take here. One takeaway, no pure managers. Every leader has to be working as an individual contributor. The pure people manager role, the one that built most corporate career ladders over the last 50 years, no longer exists at Coinbase. Leaders will have 15 plus direct reports. That is insane. Previously, managers capped out at six direct reports. I felt like 10 was sort of the platonic ideal there. But Andrew says that would be impossible without AI. And Coinbase is testing one-person teams. A single person is the engineer, the designer, and the PM. A pod of one with agents.