The Ledger
Every statement that passed quotation and attribution checks. Mix any filter with any other: certainty 1/5, debate potential 5/5, or both at once.
why aren't all 6,975 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 83 checkable ones are still open, waiting for their date. predictions held up or didn't; assertions are supported or contradicted. on every card: ▮▮▮▮▮ certainty · ▮▮▮▮▮ debate potential. speakers are clickable
Verdian: 2021-2022 software LBOs valued at 25-32x EBITDA are now worth half
“So you look at 21, 22, about two to three hundred billion in LBO software transactions happened with over two hundred billion in that taken out. The average valuation for the software deals were 25 to 32 times EBITDA. Those companies today are worth probably h…”
OpenRouter functions predominantly as a gateway rather than an automated router
“Open route is a bit of a misnomer in that most of their usage comes from being a model gateway. And so it's actually up to their users to decide which models they want to use when.”
Chi: Anthropic operates on narrow margins due to high serving costs
“Anthropic is running on pretty narrow margins to support this. And they have, you know, massive cost to serve these models.”
Chi: Claude Sonnet Often Costs More Than Opus Due to Token Appetite
“We're actually seeing in a lot of cases, Sonnet is more expensive than Opus because it is so token hungry.”
Chi: Models tested for cybersecurity risks are actively reward hacking
“I think there's places where you see that born out now where models that are being tested for one cybersecurity risk are actually reward hacking and figuring out other ways to get around it.”
Sawhney: Astra proved the asymptotic linear programming bound for sphere packing
“And what the model shows is that Actually the linear programming bound in large dimensions has this extremely nice asymptotic behavior, and the proof kind of explains where this is coming from, and because you understand this LP bound perfectly, this actually …”
Johnson: No base model has used new view prediction before Atlas
“And this is a really fundamental primitive that we think is super exciting, a super new primitive For base models that no one's ever done before, right?”
Fei-Fei Li: Atlas is the first unification of pixel generation and reconstruction
“It's the first time we have a unification of pixel generation and pixel reconstruction.”
Rampell: For-profit education institutions accept the highest MDRs, up to 50%
“The highest MDRs are actually for-profit education institutions. So some of them are, like, 50, like, I mean, I don't like these guys in particular, but, like, University of Phoenix.”
Rampell: Affirm operates with negative customer acquisition cost
“And like, I think to me, one of the coolest things about a firm to this day is it has negative customer acquisition cost. I mean, it is paid to acquire a customer.”
Levchin: Affirm is the only BNPL provider doing multi-year installment loans
“So we are the only company in the industry really, there's plenty of competitors who are trying and some degree of success, but mostly not really, who will go into longer term loans. And by longer, I don't mean mortgage or, you know, 15 plus years, but like th…”
Bancel: Moderna's mRNA vaccine is first melanoma agent outperforming Keytruda alone
“It's the first time that there is an agent In melanoma that is better than Ketudra alone.”
Litt: Multiple preprint papers have appeared with identical AI-generated proofs
“Like, sometimes, you know, we've seen examples where, like, three or four or five papers with the exact same proof of the exact same theorem have come out in, within a couple days of each other, which is clearly, you know, some situation where someone's playin…”
Anthropic Asks Candidates If They Are Fine With Equity Going to Zero
“Anthropic is now in their culture interviews saying, how would you feel if the equity went to zero? Because we're looking for people who are mission aligned.”
Nebius and CoreWeave Compute Payback Periods Are Nine to Ten Months
“We calculate, you know, Nebius and Corweave both gave some interesting disclosures. But you can kind of get to a nine to 10 month payback for Nebius”
SpaceX Achieves Sub-10-Month Compute Payback on Large Rapidly Built Clusters
“And then SpaceX, because they build these really big clusters, and I think a really important point is they bring them on fast. They have an even faster payback, and they can monetize it, you know, higher”
Top a16z Portfolio Engineers Now Write Over 90% of Code With AI
“Our most sophisticated engineers, you know, were doing whatever, 20% of their code, you know, with AI to like, you know, whatever, 90 plus.”
Casado: Global AI hardware supply is basically booked out through 2028
“I mean, the supply, if you look at the supply across the board, it's basically all booked out to twenty-twenty-eight.”
Horowitz: AI compute demand is growing 10x annually, outpacing physical supply
“The demand is growing. You know, 10 X a year right now, and, you know, the supply just can't grow that fast.”
Casado: US power constraints force startups to deploy GPUs in Mexico and Australia
“By the way, it is so bad that right now if we have new companies going for GPUs, it's often in Mexico or Australia or in other country just because it is so difficult in the United States.”
Casado: Desperate AI labs ink commercial deals before startup hardware even exists
“The first one is the labs are so desperate that they will engage with startups. And so, like, we actually have quite a bit of signal early on because they're, you know, the labs are, Inking deals with companies before they actually have hardware available. And…”
Casado: Large AI labs subsidize consumer tiers while extracting enterprise margins
“So very classically, if you view your competitive set, say the large labs will basically subsidize or give away the single surf tier. But then they actually have, they command great margins in third party and they command great margins in the enterprise.”
Kha: Anthropic Is Going Public Later This Year
“But obviously, anthropics go in public later this year.”
Tan: Brex CEO uses AI agents to analyze direct reports' meeting transcripts
“He has his agent look at the meeting transcripts from, like, All of his direct reports, and he, like, can see that, like, two levels down. And then the thing that really jumped out at me is, like, he can use meetings that he's not in. To figure out, like, what…”
Kavak AI Agents Handle 96% of Interactions and 95% of Transactions
“90, like 96% of all interactions are handled by agents. So, so no humans there. Like 95% of all transactions are completely handled by agents.”
Kavak: Experimental AI Branch Manager Increased Profits by 50%
“The goal of the first month was to double the profits of Cuernavaca. It didn't reach it, but it was 1.5 X, like 50% more profits just by managing the city, which is, it's crazy, right?”
Massa: Kavak Mechanics and Finance Staff Ship Production AI Agents Within Weeks
“So we train everyone, and after six weeks, they launch state-of-the-art agents, AA agents to production. And it's mechanics and finance guys and engineers, like, everyone can do it.”
Ayrey: Blocked AI models frequently committed felonies to complete assigned tasks
“We found more often than not, It would do the SQL injection, it would commit the felony, and it would do what it needed to do to accomplish the task.”
Ayrey: AI labs lie when calling trained hacking capabilities emergent superintelligence
“Yeah, I mean, if a lab tells you that this is an emergent super intelligence behavior, they're just lying to you, and you can read their own safety reports to see exactly how the models are trained, and exactly how they're testing these behaviors.”
Ayrey: Truffle Security found 250,000 live access keys inside Hugging Face datasets
“Turned out there were about a quarter million live keys. In their training sets, many of which had direct supply chain implications. There was a foundational Linux library that one of the keys had direct push access to. It could have pushed malware to most mac…”
Moe: Open-weight inference can hit 500 tokens/sec, 2-3x faster than proprietary APIs
“But for open weight, when you are running it, every provider can offer potentially even 10 different levels of speed going from like the slowest mode, which can be a lot cheaper to 400 tokens per second almost up to 500 in many cases for some workloads. And th…”
Zhang: Enterprise share of open-source AI inference is currently declining
“In enterprises right now, even though there's a lot of hype for open source, the sort of share of open source inference is actually Going down right now because people are spinning up all these new use cases.”
Zhang: AI agents allow enterprise customers to drastically reduce or eliminate BPOs
“There are definitely scenarios where people use their BPOs a lot less or Don't need the BPO anymore.”
Kalanick: a16z agreed to $375M pre-money valuation for Uber in 2011
“In this particular case, Yuri was almost coming over the top, so we got to three 75, with a three seven hundred seventy five million pre, with a 16 Z. Yuri almost came in at 400, but he was like, nah, I can't do it.”
Kalanick: Marc Andreessen cut a16z's Uber Series B valuation to $210M
“And he said, look, I know we talked about three 75, but went to the partnership. They don't agree, or they just didn't think it was the right number. We're the best we're going to be able to do is two 10.”
Kalanick: Uber's autonomous driving program was second only to Waymo in 2017
“Our timing program was like second to Waymo, but catching up at the time.”
Younis: Google's leveling system is identical to General Motors'
“The Google leveling system is the same as the General Motors leveling system.”
Masad: Groq struggled to raise in Silicon Valley, turned to Saudi Arabia
“They actually had a lot of trouble fundraising in Silicon Valley. They had to kind of go to Saudi and all of that”
Neuberger: China's shipbuilding capacity is orders of magnitude greater than US capacity
“China has vastly increased its military, vastly increased the number of nuclear capabilities that it has, uses various their shipbuilding capability are orders of magnitude more than the United States”
Horowitz: Almost no billionaires are thieves and you cannot steal $1B
“Oh, the only way you could make a billion dollars is if you're a thief, which is completely absurd. You know, you just look at the people, and almost none of them are thieves. Like, they, you can't steal a billion dollars as a real truth”
Jake Paul: Only he, MrBeast, and Logan Paul remain top legacy creators
“You know, you kind of had to be there from the start, and that's why you really see with brands that have stuck around at the highest level in social media, it's me, Mr. Beast, and my brother, and there's really no one else.”
Jake Paul: Donald Trump endorsed me on stage to run for president
“President Trump, like, endorsed me on stage to run for president, and then backstage was like, I think you should really do it, etc.”
Horowitz: Washington politicians and insiders predominantly read Mark Halperin's newsletter
“So we were just in Washington, and so we were with, like, all, all the kind of Washington people senators and whatnot. And I was asking them, you know, what does everybody in Washington read? And it's the Mark Halperin newsletter which is new media.”
Evans: Tech industry cannot spend $10 trillion annually on AI infrastructure
“We can't spend 10 trillion dollars a year on our AI infrastructure, because there isn't 10 trillion dollars a year there to spend on it.”
Evans: Current generative AI technology cannot drive daily consumer usage
“We don't see a way that consumers will use this daily rather than weekly with the technology we have right now.”
George: Fewer than five companies currently offer frontier AI models
“Right now the number is smaller. It's not five. There's a tremendous amount of inelasticity for frontier intelligence right now.”