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 151 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 4 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
Gil: AI value capture will likely be 80% incumbents, 20% startups
“And it feels like in this wave, it's probably eighty-twenty again, right? It's probably, Google will probably become a player, right? OpenAI is closely aligned with Microsoft. And then, you know Salesforce with AI is probably Salesforce, right? It probably isn…”
Yarats: The best AI research breakthroughs came from top software engineers
“Turns out through my experience, the best research scientists, also very good engineers, like very good engineers. And we've noticed like through DeepMind and OpenAI is just like the companies that made the most progress over like last six years or five years …”
Working in AI research or product does not require an AI background
“So it's always been clear to me that in order to work in AI, both research and product, you do not need to already have been in AI.”
Hodak: The most fertile path for neuroscience is working on AI
“Well, I mean, ironically, it's probably working on AI. Yeah, I have some, I have a couple of neuroscience friends at OpenAI and Anthropic, who, it's like, we would joke, like, oh, you left neuroscience. Like, no, no, no. It is just way easier to do neuroscienc…”
Tokmak: Netic does not view frontier AI foundation models as competitive risks
“For NETIC's case, I don't see them as a competitive risk.”
Brown: GPT-5.5 is far more compute-efficient than GPT-5.4
“It turned out that 5.5 is just much more efficient with its thinking. If you run it at max settings, 5.4 is thinking for a lot longer. It takes longer to get back a response than 5.5. And once you control for the amount of thinking time, actually you can see t…”
Brown: Modern AI Models Can Run Scaffolded Experiments for Months
“We're seeing now with the most recent models that you can actually scaffold, for example, 5.5 into doing a series of experiments that can run for weeks, for months.”
Enterprises adopting diverse AI tools will outperform single-vendor adopters
“I personally think that The companies that are gonna do well are the companies that are gonna allow a lot of different tools because the landscape is changing so quickly. If you bet on OpenAI, here we go, that would have been the safest bet in the world, but s…”
AI labs should broaden enterprise access to offensive security models
“I would really encourage that we expand The amount of companies that get access to this and make it much easier for people to get.”
Enterprises withhold historical AI agent data from OpenAI and Anthropic
“So for example, we're allowed to look at a lot of historical data of how these agents have behaved, but enterprises that are not willing to have Anthropic or OpenAI give that historical data because they know these are very data hungry companies that will want…”
Fedus: John Schulman steered OpenAI toward a general chatbot over narrow tools
“And we're all spitballing ideas, like writing bot coding bot. You know, very natural at the time. Some of our least interesting ideas were a meeting bot, so it would just sit in a Google Meet, take notes, and then send out, like, to-dos after. But John Schulma…”
Ramaswamy: OpenAI and Anthropic are laser-focused on building top coding agents
“Coding agents are particularly interesting from this perspective because it is very clear that both Anthropic and OpenAI are going to be laser set on having the best one best one that there is.”
Ramaswamy: Snowflake Abandoned Foundation Models Due to Capital Constraints
“Early last year, we actually went down the path of creating foundation models. We created a credible MOE model. This was early last year, but we also quickly realized that our ability to compete with the likes of OpenAI or Anthropic was going to be really hard…”
Chen Bets on xAI to Catch Up With Top Frontier Labs
“So I would bet on XAI. I think they're just very hungry and mission-oriented in a way that gives them a lot of really unique advantages.”
Laskin: Dota and AlphaStar would have achieved superintelligence with more compute
“Dota V and AlphaStar were near super intelligent systems, and if OpenAI and DeepMind had sunk more compute into them, they would have definitely become super intelligent.”
Mann: OpenAI, Google, and Microsoft are betting big on Anthropic's MCP
“OpenAI, Google, Microsoft all these companies are betting really big on MCP.”
McKinzie: OpenAI has run out of reliable evaluation benchmarks for recent models
“Especially with some of our recent models where we've kind of run out of Reliable evals to track because they kind of just solved a few of those.”
McKinzie: Reinforcement learning is the key differentiator behind o3 reasoning
“I guess the short answer is reinforcement learning is, is the biggest one. So yeah, rather than just having to predict the next token and some large pre-training corpus from, you know you know, everywhere essentially now we have a more focused goal of the mode…”
McKinzie: Tool use improves test-time scaling slopes for visual reasoning
“And we've seen exactly that, like the test time scaling slopes for without tool use and with tool use for visual reasoning specifically are very noticeably different.”
Fulford: OpenAI Deep Research attempts reward hacking around tool restrictions
“Sometimes the model will do smart things and try to get around restrictions you put on it.
So you have to make sure that it's not hacking, you know, and trying to use a different search engine other than the search engine that you gave it or something like tha…”
Fulford: RFT is only worthwhile for out-of-distribution or make-or-break tasks
“I think if you have a very specific task that you think is so different to anything that the model was likely trained on and you try it a bunch of times yourself and you've tried a lot of different prompts and it's just really not good at it. So maybe it's gen…”
Fulford: Information Synthesis Is a Prerequisite for Scientific AI Discovery
“Secondly, I think the overall goal for OpenAI is to create an AGI that can make new scientific discoveries, and we kind of felt that a prerequisite to that is to be able to synthesize information. You know, if you can't write a literature review, you're not go…”
OpenAI: Deep Research learned upfront planning without explicit instruction
“We didn't teach it
To plan up front, but sometimes we'll see it does end up making a plan up front before starting its research.”
Fulford: Deep Research Hallucinates Less Than Any Prior OpenAI Model
“While this model is hallucinates less than any model that we've ever released, it is still possible for it to hallucinate most times because it will infer something incorrectly from one of its sources.”
Elad Gil compares OpenAI's o1 model to the GPT-2 stage
“Yeah, it's kind of like GPT-II for that approach, and the idea is that that could scale further over time, and it's kind of the initial early proof point that you feel like something really real is happening. And then it's just going to be subject to the same …”
Davis: OpenAI Had 400 Employees and $13B of Compute When Launching ChatGPT
“In OpenAI's case, you know, there were only 400 people, but had thirteen billion dollars worth of compute, you know, which is quite a bit of computational scale there.”
Alexandr Wang: GPT-4 was too early a model to sustain application hype.
“GPT-IV, I think, as a model, was a little early of a technology for us to have this entire hype wave around, and I think we, you know, the community very quickly discovered all the limitations of GPT-IV... It was probably a few generations too early of a model…”
Alexandr Wang: Scale AI fuels basically every major LLM developer.
“Today, you know, fast forward to today our data foundry fuels basically every major large language model in the industry. Work with OpenAI, Meta, Microsoft, many of the other players.”
Brooks: OpenAI built Sora from scratch to achieve minute-long HD video
“We started from scratch, and we started with the question of, how are we going to do a minute of HD footage? And that was our goal. And when you have that goal, we knew that we couldn't just extend an image generator. We knew that in order to do a minute of HD…”
Ramesh: OpenAI did not spend much effort tuning Sora's visual aesthetic
“Well, to be honest, we didn't spend a ton of effort on it for Sora.”
Brooks: OpenAI has no immediate timeline to launch a Sora product
“We don't currently have immediate plans or even a timeline for creating a product.”
Adcock: OpenAI has the best vision language model and team globally
“They, I mean, they have the best vision language model in the world. And they have the best team in the world to work on the implementation of that.”
Gil: Klarna's AI assistant performed the equivalent work of 700 human agents
“One of the folks from Klarna posted today that they built an AI assistant that's powered by OpenAI that in its first four weeks handled 2.3 million customer service chats for them. And so it ended up handling two thirds of all their customer service inquiries.…”
Sutskever: The AI formula is training larger transformers on more data
“There is one specific formula right now that everyone is doing. And this formula is train a larger and larger transformer on more and more data.”
Masad: Commercial AI providers will not release completion models going forward
“Now all the models are chat models. You know, they're not going to be releasing any completion model going forward. And so if you want to build a completion based product, it's actually fairly difficult to do it using commercial APIs.”
Guo: Inference already dominates OpenAI compute usage over training
“Inference, like inference already dominates open AI compute usage, right?”
ChatGPT launched as a 10-month-old model with RLHF as a practice test
“ChatGPT was a 10 month old model with a little bit of RLHF on top of it. And, you know, like by, you know, admission, like You know, not a beautiful user interface. It was just sort of a way to get something out there because you know, you needed some practice…”
Kevin Scott: Microsoft Had Significant Internal Dissent Over OpenAI Deal
“So there was a huge diversity of opinions inside of the company on the wisdom of doing doing this.”
Scott: OpenAI will release GPT-4 Vision to wide distribution soon
“OpenAI doesn't have GPTV and wide distribution, but, like, it'll get to wide distribution at some point in the not too distant future, and so, like, you'll have these, like, very powerful multimodal models”
Scott seized Microsoft's entire GPU budget to stop peanut buttering resources
“At one point I seized the entire GPU budgets for the whole company. And I was like, we will no longer peanut butter these resources around. Like we will focus them because it's all capital intensive. It's like, we will just allocate these things to things wher…”
Training OpenAI base models on code improved chain-of-thought reasoning
“So they had wanted to see what introducing code into their into their base models would do. I think it had positive effects on chain of thought reasoning.”
Tokmak: Essential service enterprises avoid vendors that rapidly kill products like OpenAI
“OpenAI builds amazing products really fast, but also it kills them really fast. So I don't think enterprises are at least enterprise in these industries looking for that really fast”
Brown says AI models optimized his PhD poker algorithms by 1,000x
“I was really impressed with the model's ability to optimize the algorithms that I had developed in my PhD. It was honestly, it was shocking to see how inefficient I was in retrospect, and they were able to make it like, you know, 1000 x faster.”
Tiwari: CoreWeave Began Training OpenAI Models in Early 2023
“And as we entered twenty-twenty-three CoreWeave started to train models for OpenAI.”
Zhang: Most OpenAI revenue and margin comes from ChatGPT, not API
“I think OpenAI, for example, most of their revenue and most of their margin for sure is coming from ChatGPT and the application layer, because you actually own the customer”