Leopold Stays in the Game, Big Tech Earnings, OpenAI Slashes GPT-5.6 Prices | Diet TBPN

31 Jul 2026 · 27 min · 11 chapters

Ask about this episode

Ask anything about it. ChatGPT or Claude reads this page and answers with the times it was said.

Connect VO and ask about every podcast you hear, including the moments you saved. Add to ChatGPT · Add to Claude

In short

A guest co-hosted discussion about (1) Situational Awareness LP’s July drawdown and its response—selling the public equity book to Citadel to remove leverage and close shorts—plus debate over whether this was a “liquidation/bailout,” (2) Big Tech earnings and AI CapEx signals from Microsoft, Apple, Amazon, Meta, and Alphabet, and (3) OpenAI cutting GPT-5.6 prices and improving performance on Arc-AGI benchmarks via harness/API settings.

Guests/backgrounds

Tyler (guest co-host) in the “Ultradome.” Leopold Aschenbrenner is the central subject via an LP letter; Joe Weisenthal and Tracy are referenced via live-tweeting.

Key claims

Fund wasn’t shut down or transformed into private-only; it became fully paid-for public book with no margin liquidity risk. July net month-to-date performance was -67% but YTD remained +80%. Big Tech beat earnings; stock moves hinge on AI CapEx guidance. OpenAI reduced GPT-5.6 costs and tripled Arc-AGI V3 scores by enabling specific API settings.

Notable examples

Block sale of public equity to Citadel; “bank run” dynamics from adverse trading and liquidity drying up; Microsoft Azure +43% YoY; Meta capex $31.1B; Alphabet Google Cloud +82% YoY; Arc-AGI V3 harness/API change.

Written by AI. May contain mistakes. Listen to the episode to check what was said.

Chapters

Tap a time to open that second in VO

Leopold's Recent Challenges

0:45 to 2:10

Discussion of Leopold Aschenbrenner's struggles with his fund and market performance.

“was underperforming that month but what does it mean is the fund gone forever is he going to work at McDonald's as some people were trying to make it seem like it was happening obviously that's not going to happen.”

Analyzing Leopold's Letter

2:10 to 4:50

In-depth review of the letter Leopold sent to investors, detailing fund performance and strategy.

“So he says, over the past two years, we have delivered outstanding results despite occasional sharp pullbacks.”

Wall Street Dynamics and Market Behavior

4:50 to 6:10

Exploration of market behavior and dynamics impacting Leopold's fund during volatile periods.

“To be clear, this should rightly have been a very painful month in terms of the performance of our fund.”

Lessons Learned and Future Outlook

6:10 to 8:10

Leopold's commitment to learning from mistakes and the optimistic outlook for future investment opportunities.

“AI may continue to intensify market volatility for years to come.”

Comparisons and Industry Commentary

8:10 to 10:00

Comparisons of Leopold to other financial figures and the dynamics of investment banking.

“They're still up 80%, which is better than any investment fund ever.”

Leopold's Media Presence and Public Relations

10:00 to 13:50

Discussion about Leopold's media strategy and the recent emergence of new photos.

“Brand new rare Leopolds from Wall Street Journal.”

Industry Insights on Risk Management

13:50 to 14:00

Insights on risk management strategies and the importance of diversification in investment.

“Roy Driscoll says, there's nothing to learn from the situational awareness situation about the AI trade.”

Market Trends and Risk Management in Tech

14:00 to 18:45

Explores the current market trends, risk management strategies, and perspectives on leverage in trading.

“And based on the Amazon results, he's still right today.”

Debate Over Wall Street's Blow-Up Culture

18:45 to 20:12

Discusses the perspectives on the necessity of experiencing financial setbacks in trading careers.

“And the bids came in above liquidation level.”

Earnings Season: Big Tech's Financial Results

20:12 to 23:36

Analyzes the recent earnings reports from major tech companies and their impact on the market.

“But like an in the room, fly on the wall, minute by minute account of this would be interesting.”
Show all 11 chapters

OpenAI's Price Cuts and Model Efficiency

23:36 to 27:12

Covers OpenAI's recent reductions in model costs and its implications for AI efficiency and performance.

“You can see on the Pareto curve, this is actually much cheaper than a lot of open source models.”
Hear the part that matters, and keep it.Open this episode in VO. Double tap your headphones to save a moment as you listen.
Get VO free

Transcript

Automatic transcript. May contain errors.

0:01John Coogan:We have a guest co-host today. Introduce yourself for those who don't know. Jordi's out again. I'm back. Yeah, we got Tyler in the Ultradome in the hot seat. You're back. You know who else is back? Leopold Aschenbrenner's back. He says you're going to have to drag me out of SALP, out of situational awareness LP, because he's down but not out. A little beat up, but he shared a letter that's making the rounds, thanks to some intrepid reporters on the DBPN team that posted this. he sent an LP letter that clarifies a lot of the questions yesterday I mean even internally we were going back and forth on like okay he sold a bunch of the portfolio to Ken Griffin to Citadel does this count as a liquidation yeah does this count as blowing up and these are like sort of vague terms like what what does it mean to blow up there was definitely a drawdown the fund definitely was underperforming that month but what does it mean is the fund gone forever is he going to work at McDonald's as some people were trying to make it seem like it was happening obviously that's not going to happen.

0:57John Coogan:He's going to have a long career. Lots of people are rooting for him. I'm certainly rooting for him. There are some facts in this letter that we should read through. So he writes, this is Leopold Oshenbrenner to the LPs of Situational Awareness LP. We let you down this month. We came closer to permanent capital impairment than is acceptable to us. While we ultimately found a solution that protected the fund and you as investors, that was the sale of the public equity book to Citadel. There was some other structure going on to get liquidity He said, we ultimately found a solution that protected the fund and you as investors.

1:31John Coogan:Our intention in running the fund is to never find ourselves in such a position in the first place. Volatility is the price of long-term investment returns. Over the past two years, we have delivered outstanding results. That's 100 % true. It was up, what, 1 ,000 % at one point or something like that? Yeah, something like that. I mean, it got up to, what, 45 is the number, right? You have 45 billion AUM from an original raise less than two years ago, I believe, of$250 million, which seemed crazy at the time. People were like, he's a young first-time hedge fund manager. He's got$250 million. That's crazy.

2:04John Coogan:Then pretty soon it was like, oh, he's got a couple billion. That's crazy. Then it was like, he's got tens of billions. Then he's got half a centibillion. So he says, over the past two years, we have delivered outstanding results despite occasional sharp pullbacks. Probably not the first time. There's been other pullbacks in the market, and those have probably been amplified, but never gotten to this level of actually distressing the fund in this way. He said, but our fund must always be structured such that we can take a loss and fight another day. And that's a recurring theme in this. The writing in this letter is really good.

2:40John Coogan:Very clear, very direct, not being dodgy, very upfront. I love the way it's written. Almost kind of like there's the PG advice to write very clearly. Yeah. I think it was very kind of in that line. There's a lot of that in here, yeah. So he says, I will make it my mission to ensure that we learn the necessary lessons from this experience. Here's where things stand. One, the portfolio experienced a significant drawdown over the course of July, which was exacerbated by extreme moves in core positions over the past week. Many AI names drew down by half or more, while our positive long-short spread reversed violently.

3:14John Coogan:While we could say much more about how unusual the month was, we hold ourselves to a higher standard irrespective of market conditions. Two, as these moves proceeded, we started to see increasingly adverse trading in names publicly associated with us. So this is the rumor that Martin Screlly was talking about yesterday, this idea that there's blood in the water. Yeah, you can kind of sniff out if someone's hurting. Exactly. And then short sell those positions, sell those names, put some pressure on those, downward pressure, to actually intentionally hurt that fund. It's a knockout, dragout fight there on Wall Street, clearly.

3:52John Coogan:But that's the game you're playing. That's why you get paid the big bucks, if you can pull it off. So these dynamics are essentially similar to a bank run. Crazy to put that word in there. A lot of people would be dodging that, but very, very direct. I love it. Vulnerability begetting more vulnerability. We worked to keep the portfolio within our risk parameters, but gradually this became more difficult as positions rapidly moved against us and market liquidity dried up. On Wednesday night, Thursday morning, we took decisive action to protect LP Capital. We traded a portion of our public portfolio in a block transaction to remove all leverage from the fund and prevent further losses.

4:27John Coogan:All shorts were closed and reliance on portfolio financing removed. We currently manage a fully paid for public book, long stock and long fully paid for options with no margin liquidity risk. This restored stability and allowed us to preserve our private positions. So this feels like down but not out for sure. And he says, I take full responsibility for these events. That's just the full paragraph. He just says, I take responsibility. No equivocating. It's great. To be clear, this should rightly have been a very painful month in terms of the performance of our fund. When AI stocks draw down dramatically, while AI technical business fundamentals are improving, you should expect our fund to be down a lot.

5:07John Coogan:We embrace volatility, but it should never jeopardize the fund. The fund was not shut down. It was not liquidated or transformed into a private-only fund. This was something that a lot of people were speculating on, was, is this going to be private-only? Are they only going to have their private book? Is it just going to be the anthropic position that's going to be riding? Or is it just going to be liquidated and they're just going to return capital LPs and just say, hey, we're going to start completely fresh, do something completely different? Even like an acquihire, like the situational awareness becomes like a desk at another fund.

5:39John Coogan:None of that's happening. He's very clear about this. situational awareness is not shutting down. It's not liquidating and it's not transforming into a private only fund. He says we are continuing to operate as a hybrid public private fund as before. However, we will manage our public book on a fully paid for basis while we draw the lessons from these developments. Most importantly, we took the steps that were necessary to fight another day. I love it. A rallying cry to both the LPs and the employees, I'm sure. In the coming weeks, I will focus on putting in motion the necessary changes across a portfolio management risk team and vigilance applied across the board to ensure a higher level of resilience going forward.

6:16John Coogan:AI may continue to intensify market volatility for years to come. And that is something that is so clear outside of the situational awareness, bottleneck trade, long tail, low market cap, high volatility stocks. I have never seen the MAG7 trading like this where across earnings, we're going to get into this with recapping, Meta, Apple, Amazon, Microsoft. Yeah. I think the stat was Microsoft had the biggest day ever of any public company. The biggest move in value ever. Yeah. So you're seeing trillion-dollar companies move by 10%, 9%, 15%. It's insane that anything can happen at that scale. And so clearly there is going to be a lot of volatility.

7:00John Coogan:And I think he's right to point out that it is based on the AI trade. There's so much uncertainty about one little number about how the CapEx is going to trade back. You know, the investors in these large companies, let alone the small ones, are moving the stocks significantly. And that makes his job all the harder. He says, these were very expensive scars, but I am dedicated to ensuring they will be invaluable lessons for our organization and for myself as we move forward. My core promise to you is that we will not waste the opportunity to learn from these events on the portfolio itself. We are very optimistic about the current investment opportunities.

7:37John Coogan:Of course, I mean, the thesis still holds. The underlying fundamentals are accelerating at the very same time that prices have declined significantly. Thank you for your patience and your partnership. I'm fully invested alongside you. Virtually all of my capital is in the fund, and I intend to work relentlessly to demonstrate that the events of this month have made me a wiser and stronger investor. He says he's available for calls. But he also says that as an interim update, the current unaudited estimate of net month-to-date performance, this is for all of July, basically, negative 67%. Sounds atrocious until you realize that net year-to-date, They're still up 80%, which is better than any investment fund ever.

8:19John Coogan:So people are definitely maybe down but not out. There's going to be a second act here, which I think everyone's very excited for. A lot of people were praying for his downfall. It's very unfortunate to see. I think this was a really good letter. I mean, this instills so much faith. Yeah, he's completely level-headed. He's not freaking out. You're calling it another billion-dollar PDF. Yeah, this could be the second billion-dollar PDF. It might be. I mean, Schulte agrees. Schulte had a great take. Had a great position. What did he say? He said, prediction. Situation awareness will be bigger than Citadel by the end of the decade.

8:54Leopold has predicted the last two years better than anyone else. Now that he can combine that with very expensive lessons in risk, he will be unstoppable. He has my full confidence.

9:02John Coogan:This is such a wild post. Kane Griffin sitting there being like, You got to ride with your boys. Schulte, you're going to take a shot at me like that, bro. Really? Really? You're going to come for me? that because I will die before I am not the biggest hedge fund manager in the world. Pull up a live view from the Citadel trading floor because we got some leaked video.

9:28John Coogan:This is not the vibe that Martin was articulating Ken Griffin's wanting to be framed as like the savior, the life of last resort, positive force. This is the guy you want to call. I guess. Dune, so good. Is this from Dune 1 or Dune 2? I think this might be Dune 1. I want to say Dune 1. Yeah. Beautiful. It's very heavy metal. So I think one of the big stories of the last few days is we've seen all these new Leopold photos. We've never seen these before. Yes. Brand new rare Leopolds from Wall Street Journal. This might be the biggest story of them all. Because, I mean, for a while, the only image of Leopold was basically there was like one headshot and then it was just stills from Dworkhash podcast.

10:15Yeah. And now we're just seeing all these new ones. Where did these come from?

10:18John Coogan:There was one and then there was a photo that was done, I think, for The Wall Street Journal. But then The New York Times writes up the whole story of the situational awareness deal with Citadel. And they just drop a banger new photo that they just had in the archive that they could have leaked. Let's pull it up. It's here. It's Leopold looking very pensive behind a glass wall. This one's in the Wall Street Journal today. This one's new, too. Everyone's been clamoring for this. Because the one that goes viral is him in that green suit. This is the one. I'm pretty sure that's AI. That's AI. But this one is not.

10:51John Coogan:This is from the New York Times. They went and shot this and then never published anything. Like, the first time Leopold was mentioned in the New York Times was yesterday. And they used this photo. And so you have to wonder if they were, like, working on a profile. Yeah, they just been sitting on it. But Leopold's been so quiet with his public relations strategy. He's not talking to media, doing photo shoots, doing profiles constantly. He certainly could be doing more in Bloomberg and Forbes and Fortune. He could be doing a lot. But he's had a very narrow strategy, and I think it's worked very well for him.

11:21John Coogan:But it's funny that somehow all the mainstream media just has secret Leopold photos. They've been dropping on the timeline. John Arnold is chiming in. And he says, my philosophy when I used to hire traders was that the optimal number of past blowups was one. He's not saying zero. Yeah. He says you've got to learn your lesson. The question is, how does the FTX future fund count? I don't think that counts at all. It definitely doesn't count as a full blowup. He wasn't a fund manager of it, right? It wasn't like for donations and then FTX was just the one that was funding it. Yeah, it was like plans free.

11:52John Coogan:He was doing plans free stuff. That seems completely separate. Is the wedding photo AI or is this real? And is he carrying an American flag? Let's pull up this image. Yeah, this one I've never seen before until yesterday as well. I mean, if it's the wedding, like the wedding's happening right now, this wouldn't exist. But I'm wondering if this leaked onto the timeline from someone who was there. Also, this photo hit Wednesday, and I think the wedding would be over the weekend. But it's cool that he's just rocking, carrying an American flag international. Yeah, the real lesson here is never travel internationally because he takes one day off, one weekend off to go to Europe, and everything blows up now, of course.

12:29John Coogan:This one I was laughing at before the show. Rambo says, comparing Leopold Aschenbrenner to Bill Wang. Wang is the goat. Bill Wang is from Ark of Ghost. Wang is the goat of degenerates, and Leopold is a sheep compared to him. Did you know that Wang turned$200 million into$36 billion, and it was all personal capital? The guy literally led prayer circles in the conference room before trading day started. He had$160 billion of stock exposure on just$36 billion of capital. It's like five or six X levered. His blow up happened in two days and he literally caused the collapse of one of the most prestigious investment banks.

13:08John Coogan:Banks lost a total of$10 billion combined because of his collapse. Leopold is nothing compared to Bill. It's so. Gets her numbers up. Yeah. No. Yeah. I mean, that's the interesting thing here is that like it is this sort of like dramatic unwind. But at the end of the day, it is just like an over-the-counter transaction with Citadel for a block of trades and a block of equity positions. Yeah, and the fund is still around. I mean, they still seem to be like probably going to be doing very well. Yeah, they'll be okay. Citadel will be okay. They're going to be bigger than Citadel pretty soon. Any day now.

13:41John Coogan:And importantly, all of the prime brokers, the big banks, like they were not affected. There was not like a liquidity crisis that – a contagion effect did not take root. Roy Driscoll says, there's nothing to learn from the situational awareness situation about the AI trade. Leo was right in 2024. And based on the Amazon results, he's still right today. Hyperscale CapEx continues unabated. There's obviously something to learn about risk management. Forex leverage with high beta stocks is a mistake in trading stocks. Half the battle is getting the trend right. But the other half is nailing the portfolio construction.

14:15Yeah, I mean, this is what Martin was saying yesterday, right? Yeah. Like, the underlying completely makes sense. but like you get into these crazy psychology things where it's just like, yeah,

14:24John Coogan:everyone's focused on the leverage. It does also seem like there were like every time the 13 F would drop, it would be like 12 names, which is like not a lot of diversification. So I wonder like right now, the message from the letter is we're not using leverage right now. We're going to be learning the lesson. Maybe the lesson is, Hey, two X leverage or three X or something like that, or four in certain scenarios with smaller trades, not portfolio wide or something like that. But it will be interesting to see if there's a difference in, uh, If the lesson that's learned, when the next 13F drops in a couple quarters, we see, oh, wow, he has like 100 names.

14:58John Coogan:Or there's, you know, he's using more options or less options or, you know, whatever, however it changes, that will be interesting to see for sure. So Leopold still has Anthropic, Maddox, and FluidStack, tier one private companies. He can probably raise two to three billion more. It's not over for him by any means as Zephyr. There's been this vibe of like, it's too good to be true. He's too young. Tall puppy syndrome. You know, he's the AI. Can't keep getting away with it. He's the wonder kid. Yeah. Yeah, you know. Yeah, people hate to see a young... A young goat. A young hedge fund manager run it up crazy.

15:31John Coogan:Lexi Guzzi says, people keep making fun of Leopold on the timeline, but everyone needs to get margin called once in their life. Is this true? Delian had a similar take, right? He said, basically, all the goats on Wall Street have had some sort of blow up earlier in the career. part of the game, being a live player on the field. Yeah, is that true? I don't think that's actually true. I don't think Warren Buffett ever blew up. I don't even think Ken Griffin ever really blew up. I think he had a really bad year in 2008 during the housing crisis, the financial crisis. But early on, I think he got his start sort of post.com and was doing convertible debt trading and never really, like the entity has always been Citadel.

16:16John Coogan:There was no precursor to that. but it's a fair take that clearly people can build back up after there's a... Yeah, I mean, there's a lot of comparisons to PT, right? Yeah. And then they're saying, oh, this is also kind of a loose take, right? Going to VC, then you can kind of do the long only thing. The real hack would be to just raise the smallest hedge fund ever,$10 ,000, lever it, blow it up, and be like, wow, I'm post-fall. Oh, yeah, he's post-fall now. He's post-fall now. But if you do it with like such a small amount of capital, but you can still be like, oh, man, I learned so much. That was really crazy.

16:54John Coogan:Those were crazy, crazy times. I lost$500. Ready for the real fund now? No. Leo still made incredible returns. His fund will do incredibly well in the long term. Lots of people coming out in support. One person that's not in support, Joe Weisenthal, is going back and forth with Tracy. This is hilarious. So Joe has been live tweeting this. He's been making a bunch of great points and just illuminating the deeper level of like what's going on with prime brokerages and all these different aspects of what's going on. But Joe started by sharing the Wall Street Journal article that said that Citadel buys situational awareness stock portfolio after big losses in AI.

17:33John Coogan:And Tracy says, why does he have to get bailed out at all? And this is another question. Like, is this a liquidation? Is this a blow up? Is this a bailout? it would be it would have been a very different conversation if this had been like a government bailout of situational areas that's not what happened but Tracy says why can't we just let the speculators fail Joe Weisenthal says who says he's getting bailed out he entered into a transaction with a willing counterpart and Tracy says isn't that a bailout why not just keep managing the fund why not be Kathy Woods and have a bad day and live to tell another tale except there were probably too many redemptions so it was spiraling Joe says he got margin called.

18:10John Coogan:And Tracy says, so it is a bailout. Just let it fail. But maybe it was too big and could see the contagion. Joe says, I don't get what you're saying. Someone gets margin called and they have to pay the broker. And the way they pay back the broker is selling off shares to some other counterparty. How is that a bailout? Like he's just selling. And people associate every sale with a bailout now, I guess. But that's not what this is. This was not the government stepping in. Yeah, yeah. He was not too big to fail. No, not at all. I mean, some people are saying that he could have been too big to fail going in, but.

18:39John Coogan:Yeah, it doesn't seem like that's what happened. It seemed like there were significant losses and then they ran an auction and there were three parties bidding. And the bids came in above liquidation level. So the fund is not liquidated and it remains. And so, Joe, after fighting back and forth for several posts, he says, I think we might have a different definition of the term here. And I think you do. I think you do. Very fun This is an interesting scoop from Berber Gin Over at the Wall Street Journal related to this Situational Awareness tried to sell A$3.5 billion stake in Anthropoc To a group of investors led by Green Oaks And Sequoia Obviously there's a lot of demand for the stock The parties reached a deal late Wednesday But then Situational Awareness pulled out Thursday morning They turned it down They turned it down They were going off for like$3.5 billion Something like that And they turned it down just to grind for the public equity book.

19:35John Coogan:They sold that to Citadel. That obviously cleared a lot of the risk out. And they said, hey, let's keep this position. We're extremely excited about this. We're bullish. And so, I don't know. I think, will this be the subject of a book? Will this be the subject of an actual movie? Is it drama enough? Have we gotten the FTX movie yet? Because that's way more dramatic. And I don't think that ever happened. Yeah, I haven't seen it. And then there was Infinity Machine. Not Infinity Machine. There was a Going Infinite. Is that the one? That was the Michael Lewis book. Michael Lewis book. But that was written like before the blow-ups.

20:08John Coogan:And so it was like sort of, it didn't really tell the story like day by day. But like an in the room, fly on the wall, minute by minute account of this would be interesting. But it's not that dramatic because it doesn't end with an explosion. It ends with like a, okay, we're back in the fight, which is cool. I mean, it's maybe a more positive outcome. Big Tech's AI spending is continuing to produce blockbuster financial results, even as investors have become increasingly selective about which companies they're willing to reward. Over the past two weeks, Microsoft, Apple, Amazon, Meta, Alphabet, all reported quarterly earnings that largely exceeded Wall Street expectations.

20:44John Coogan:It's very boring when you pull the, did they beat on top line? Did they beat on bottom line? It's like everyone beats, and then the stock goes down 10 % or up 10 % based on CapEx forecasts and also just messaging around AI diffusion and AI uptake. Microsoft led the group with shares surging after reporting fiscal fourth quarter revenue of$90 billion, up 18 % year over year, and ahead of the$87.4 billion that analysts were expecting. That was the consensus estimate. EPS came in at$474 versus expectations of$421. So they beat top line, beat bottom line. Azure revenue accelerated 43 % year over year.

21:23John Coogan:Yeah. So they gained$450 billion in one day, 16%. $450 billion in one day. Yeah. That's for Microsoft. Biggest one-day market cap gain for any U.S. company. Look at that. Thank God. God. That's really, really impressive. It's up 25 % over the month. Very impressive. Apple also beat expectations, reporting$109.4 billion in quarterly revenue, earnings per share of$202. Stock briefly pushed the company market cap above the$5 trillion mark, but it has been absolutely tanking today down, what, 10 % today or something? that? Let's see. Down 9.47%. Last, we'll go through two more. Amazon has also impressed investors with revenue climbing 20 % to$200 billion,$200.6 billion.

22:10John Coogan:AWS growing 37 % to$42.4 billion, sending shares sharply higher in after hours trading. Here's Amazon. We can pull that up as well. The market is up 13.76%. And the day is looking pretty good too, up 15 % today. The market's reaction wasn't universally positive. Meta posted stronger than expected revenue of$60.8 billion, up 28 % year over year. But earnings per share fell short of the$7.22 analysts had expected. Investors focused on the company's$31.1 billion in quarterly capex, along with$3.6 billion in one-time legal and severance costs, sending the stock sharply lower. Let's see what Meta's doing down just a bit.

22:58John Coogan:Alphabet, meanwhile, reported revenue of$119.8 billion, while earnings per share of$9.11, comfortably beating expectations while Google Cloud revenue surged 82 % year over year to nearly$24.8 billion. Even so, investors remained focused on the escalating cost of AI infrastructure as hyperscalers continue pouring hundreds of billions of dollars into new compute capacity. And here's Google. So we can dig into this more. There's a whole bunch of deeper questions about what is the actual efficacy of meta spending on AI. How much are they spending on tokens? How much are they spending on headcount?

Read the full transcript

23:33John Coogan:All these things matter. But we'll dig into it another time. Take us through what's going on with OpenAI pushing the model frontier access across efficiency. What happened? They dropped the cost of Luna? Yeah. So there's Luna, TerraSoul. This is the cheapest model. Yes. Massively reduced cost. You can see on the Pareto curve, this is actually much cheaper than a lot of open source models. We've been talking about this recently. There's cost per task, not just like, can it do it and how much do it? It depends a lot how token efficient the model is. Yeah, because you can measure it on cost per token.

24:07John Coogan:But if a certain model takes 10 times the amount of tokens and it's only half the cost, you'll wind up spending more. Why is the Pareto frontier in this graph flipped? I feel like the Pareto frontier used to be this direction. Am I hallucinating that? Has it always been this way? You always want to be on the left side? I thought you wanted to be on the right side or something like that. Well, it depends on where you are on the Pareto frontier. I suppose. I think I see what you're saying. I suppose. Anyway, we also, I don't think we touched on this, but ArcGIV 3, the leading labs have been going back and forth.

24:39John Coogan:Opus 5 put up a very, very impressive number. Then OpenAI fired back with 5.6 SOL. used to solve open problems in mathematics. So why was it struggling with Arc AGI V3, which you at one point were in the top 10, right? Yeah, I was globally ranked Arc AGI V3. I don't think it's still up, but yeah, I was. Ranked Arc AGI V3 player. That's up there. You were like, you were pro-am. Yeah, yeah, I would say. You didn't go pro, you turned it down. I turned it down, yeah. You had the opportunity to be at Arc AGI V3. They were going to give me like 10 more tasks, five more tasks, something like that. something like that yeah uh but apparently open ai was able to investigate uh the low score of 5.6 soul on rkgiv3 and the harness was not letting it remember what it had learned we found that enabling two api settings tripled our scores with 6x fewer output tokens so very interesting uh to watch yeah this is fascinating i mean we've seen this a lot over the past like i don't know year and a half almost where the harness like really matters a lot yeah and if you have the wrong harness or it's like limiting the model in some way i mean it can have like massive like effects on the downstream task yeah people were not expecting this it was definitely like the the the the model the god model will be just one model and you'll just ask it to predict the next token it'll just do it perfectly yeah there's a lot more that goes into the integration here i still think arc agi v3 i mean fantastic benchmark love the team obviously mike's been on the show multiple times but uh it's also just a great way to actually illustrate ai progress to someone that maybe just doesn't want to build software or hasn't built software before and doesn't really, can't really feel that viscerally.

26:21John Coogan:I can't say viscerally. I don't know. Especially, you know, the famous like time horizon task doubling. Yeah, me either. Like that's basically like we can't actually measure the high end now. It's like too hard. We don't have enough tasks to like measure it efficiently. Yeah. And a lot of people are just like, what's a task that takes me 12 hours? Like, what is that? I don't even know. Yeah, it is a bit hard. It's hard to think of that off the top of your head. Like, what does that mean? Like building a whole report or something? Or like, a lot of people work in like various ways. Like, yeah, 12 hours of meetings.

26:49John Coogan:Is that one task? I don't know. But if you show someone the Arc AGI V1 puzzle and it's very easy and V2 is very, very, very easy. And then you walk them through the story of how AI has progressed on this and how hidden the answers are. You can pretty easily help someone feel the AGI, which is very, very, very, very fun. And we'll see you on Monday. See ya. Bye.

From the publisher

Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after.


Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella.


TBPN is made possible by:

Ramp - https://ramp.com

Public - https://public.com

Cisco - https://www.cisco.com

Console - https://www.console.com

CrowdStrike - https://www.crowdstrike.com

Figma - https://www.figma.com

MongoDB - https://www.mongodb.com

NYSE - https://www.nyse.com

Railway - https://railway.com

Shopify - https://www.shopify.com/


Follow TBPN: 

https://TBPN.com

https://x.com/tbpn

https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231

https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235

https://www.youtube.com/@TBPNLive

More from TBPN

All 686 episodes
Leopold Stays in the Game, Big Tech Earnings, OpenAI Slashes GPT-5.6 PricesTBPN · 27 min
Listen in VO