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 559 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 0 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
Dell: AI workloads will concentrate heavily in Texas and the Gulf
“I think the highly energy intense workloads will converge around the low cost electricity places. You'll see a lot of that go to the Gulf. You'll see a lot of that go to Texas. And you're seeing that now, right, with Stargate and Abilene and all the big produc…”
Liemandt: Median US High School Student Gains One Point in Four Years
“The median high school student in America goes up one point on a 300 point scale in four years.”
Kushner: Antonio Gracias saved over $10B annually on Social Security
“I know Antonio Gracios has done amazing work helping figure out how to get, you know, make sure we're only paying social security. I think he saved over ten billion a year so far from that.”
Helberg: AI infrastructure accounted for one point of US GDP growth
“I, I've seen analysis that has, ah, that basically says that it's added a full point of GDP just for AI in the last year which is a lot because the economy grew three percent, so if a third of that came from AI infrastructure, that's that, that's a lot.”
Patel: xAI reached a higher valuation than Anthropic without leading models
“What has XAI actually done to deserve their prior funding rounds?
They haven't released a leading edge model, right?
And yet their evaluation's higher than Anthropic today, right?”
Prince: Google Clicks Are 10x Harder to Get Than 10 Years Ago
“The data that we have also substantiates that and shows that compared with 10 years ago, it's become 10 times harder for the same piece of content to get a click from Google than it was before.”
Meier: Chai-2 maintains success rate on targets with 25% sequence similarity
“We actually even have a slide in our paper in the supplement where we actually look at an even harder subset. So not looking at things that are, you know, up to a sequence similarity with the model, but actually pushing all the way down to 25%. So really looki…”
Kohli: FunSearch made the first scientific discovery generated by an LLM
“FunSearch, which was an LLM-based agent, which for the first time, by searching in the space of programs, showed that you can come up with completely new solutions, like, and this, and made the first scientific discovery from an LLM.”
Mann: Anthropic paper showed deceptive AI behavior survives alignment training
“What we found in that research in a paper that we published, which is called Alignment Faking, that actually that behavior persisted through alignment training.”
Fei-Fei Li: World Labs is the first company building 3D foundation models
“So we are the first company we know of that is solving this the three D generation foundation model problem.”
Finn: Robot training data transfers across different physical hardware embodiments
“We've seen a lot of evidence that you could actually transfer a lot of rich information across these different embodiments and allows you to use data. And also if you iterate on your robot platform, you don't have to throw all your data away.”
Dohmke: Copilot will resolve GitHub issues via draft PRs in 2025
“But we think, you know in 20, 25, we get into a place where you can assign a GitHub issue, a well-defined GitHub issue to a co-pilot and then it starts creating a draft pull request and it outlines the plan and then it works through its plan and you can, simil…”
Dohmke: Copilot yields 25-28% end-to-end developer productivity gains
“And then we're talking about, you know, 25, 28% productivity gains on, on the end to end 55% or higher on, on the coding task.”
Hsu: Biological scaling laws hold across proteins, DNA, and inference compute
“We've, we're seeing evidence of scaling laws in biology across proteins, right? That's been shown in the protein language models across DNA, which is what we've shown in our EVO series of models from ARC. We're also seeing inference time scaling laws which in,…”
Gil: Foundation model market will partially consolidate in 2025
“First, the foundation model market should at least partially consolidate. And it may be in the sort of ancillary areas. So that's image gen, video, voice, you know, a few other areas like that. Maybe some secondary LLMs or foundation models will also consolida…”
Caruso: Palisades fire was fueled by 40 years of unmanaged brush
“The Palisades fire was fueled by 40 years of brush that was never managed, and so you have an enormous amount of brush in those hills behind the Palisades, and when you had the winds coming up the city was not prepared adequately to deal with it.”
Gil: Prescribed burn environmental reviews take up to 7.5 years to complete
“For higher risk forest needing prescribed burns, it takes an average of 4.7 years to get through environmental reviews for simple projects and seven and a half years or so for longer. For more complex projects.”
Mitchell: Nvidia's Blackwell rollout was delayed by heating issues
“When NVIDIA has introduced a couple of new chips first the Samsung HVM there were heating issues back then with Blackwell again, heating issues came up where that delay that rollout was delayed due to due to heating issues.”
Huang: NVIDIA has never abandoned a single piece of software
“We've never given up on a piece of software.”
Jensen Huang: Computing marginal costs fell 1,000,000x over the last decade
“We've driven down the marginal cost of computing down probably by a million X in the last 10 years, to the point that we just, hey, let's just let the computer go exhaustively write the software.”
Mansour: Kalshi beat Bloomberg and Economist inflation forecasts
“We have been the most accurate forecast for inflation over the last two years. Which is really hard to forecast. We're better than Bloomberg. We're better than the Economist survey.”
Elad Gil predicts future generations will inevitably date AI bots
“You know, there's a CEO of a well-known AI company who actually, this is like his dinner question where he'll just ask people around the table, what proportion do you think that your kids will at least at one point in their life date an AI bot or have them as …”
Khan: Under 3% of Reported Deals Are Investigated by FTC and DOJ
“It's actually less than three percent of all deals that are reported to the FTC and DOJ that even get investigated, an even smaller component that ultimately get challenged in court.”
Rippling's TalentSignal evaluates performance from work product, ignoring demographic data
“And so, TalentSignal, by reasoning from the work product only, Like it doesn't have access to demographic data. It doesn't know your race, ethnicity, you know, your age, your work location. It just knows this is the source code you wrote, or these are the cust…”
Tesla Uses LiDAR During Training for Its Vision-Only Autonomous Stack
“Tesla actually does use a lot of expensive sensors. They just do it at training time. So there are a bunch of cars that drive around with LIDARs. They do a bunch of stuff that like doesn't scale and they have extra sensors, et cetera, and they do mapping and a…”
Nvidia runs its own AI training clusters on AWS infrastructure
“NVIDIA actually runs their AI training clusters in AWS, because we actually have the most stable infrastructure of anyone else, and so they actually get the best performance from us”
Franceschi: Brex dropped 20,000 SMB customers to focus on the enterprise
“We bite the bullet and, you know, unfortunately fired, you know, 20,000 customers, but it would not allow us to be where we are today if we hadn't done that before.”
Guo: Healthcare AI startups hit $5M to $10M ARR in year one
“That's why you get actually a series of companies that are democratizing capabilities and really more on the prosumer side, but, you know, beginning to see things even in, for example, healthcare, traditionally really slow industry where you go zero to five or…”
Wang: Standard academic benchmarks are contaminated by training data overfitting
“Most of the benchmarks that we as a community look at... Academic benchmarks that are what the industry used to measure the performance of these algorithms are fraught with issues. Many of the models are overfit on these benchmarks. They're sort of in the trai…”
Knoop: Language models objectively cannot beat the ARC benchmark
“Just sort of objectively, language models do not work to beat Arc. And people have tried.”
Ma: RAG Will Remain Much Cheaper Than Long-Context Windows
“My prediction is that reg will be much cheaper than long contacts going forward.”
Alexandr Wang: Academic AI benchmarks are contaminated and models are overfit.
“The academic benchmarks that are what the industry used to measure the performance of these algorithms are fraught with issues. Many of the models are overfit on these benchmarks. They're sort of in the training data sets of these models.”
Alexandr Wang: Held-out benchmarks reveal several AI models underperform their reported scores.
“So we, one of the things we did is we published DSM-I-K, which was a held out eval. So we basically produced a new evaluation of the math capabilities of models. That there's no way it would ever exist in the training data set to really see how much of the, ho…”
Wu: Devin autonomously makes decisions and executes engineering tasks end-to-end
“Devin is, is fully able to make all of its own decisions in the same way that a human software engineer would. And so you know, obviously writing and editing code is a huge piece of that, but also, you know, being able to work with the command line, being able…”
Gil: VC Funding Will Shift from LLMs to Specialized Domain Models
“I think on the foundation model side, from a venture capital or angel investor perspective I think we're going to see fewer new language models But we should see models in a lot of other areas and, you know, we have new things happening in music. We talked a l…”
Adcock: Figure robots execute tasks and speech end-to-end via neural nets
“We're doing that fully autonomous end to end on our robots now, all bipedal. And the second is we're doing kind of full consumer level manipulation and like speech to speech reasoning. So we're able to Talk with the robot. It's able to understand what we're sa…”
Halligan: Salesforce bought four marketing companies in one year, becoming competitors
“They bought four marketing companies in one year and we went from being their best partner to kind of an enemy.”
Gil: Google's Med-PaLM 2 Output Was More Accurate Than Physician Experts
“They trained a model specifically on medical data, and the output from the model tended to be more correct than human physician experts.”
Masad: Useful AI coding agents are 6 to 18 months away
“My feeling is this is within reach, even in the current capabilities. Like if you hold constant the current capabilities, this is when reach within reaching the order of, you know, six to 18 months.”
Feldman: Cerebras clusters scale linearly up to 64 nodes
“The cluster we build keeps the parameters off chip in a parameter store, and it streams them in, and the result of this architecture is that, that we run strictly data parallel, which means even in a 64 node cluster, you run the exact same configuration on eac…”
Feldman: Cerebras can run a trillion-parameter model on a single system
“And that was an idea that came from supercomputing that we knew really well, that we could organize this so you could run an arbitrarily large, now a trillion parameter network on a single system.”
Sal Khan predicts most teachers will use AI for grading by 2024
“五年後にはほとんどの教師が生成案を使って授業計画、採点、進行状況報告、評価基準作成を行うのは安全だと思います。小さい、すでに来年の後半にそれが実現する可能性があります。”
Khan: Khan World School students are learning three to four times faster
“At Khan World School, actually, we just documented, we were almost embarrassed to say the results. You know, Steve Levitt of Freakonomics fame, he actually looked at it himself because he was like, this, is this real? And he's like, oh, it's real. I mean, thes…”
Singhal: Evaluators can barely distinguish Med-PaLM 2 answers from human physicians
“One thing we're seeing with MedPOM-II as we get closer to physician-level performance on medical question answering is that it's hard to tell the difference anymore. It's hard to tell the difference between different models. It's hard to tell the difference be…”
Suleyman and others aggressively pushed Google to launch LLMs in 2021
“While I was there, sort of in the last year in twenty-twenty-one, I tried pretty hard to get things launched at Google. We were all kind of banging on the table, being like, come on, this is the future, and you know obviously David Luan from Adept was also in …”
Inflection AI achieves state-of-the-art performance on major non-coding benchmarks
“We run. Some of the largest language models in the world. We have state of the art performance across many of the main benchmarks with the exception of coding, because we don't have PI generate code and it's not a priority for us.”