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 1,548 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
Frontier AI safety coordination requires slowing down China
“And then I think another aspect of this is specifically doing this in a way where part of the story is either slowing down China or cutting a deal with China such that China doesn't overtake and, you know, break this whole proposal.”
Wolf: Open Versus Closed AI Is Orthogonal to Model Safety
“If for many aspects, I think the closed open distinction is almost orthogonal to the safe and safe. People don't understand that, you know, easily because it's easier to do bad mapping than to try to understand the subtlety.”
Trojanowski: Technical AI moats are not real moats
“Technical moats are not real moats. Like, I, there's no portion of basis's long-term terminal value that stems from some, you know, secret RL trick we found that nobody else found.”
Dan Roberts: AI scaling requires novel algorithms, not just pure compute
“It's not that scale is all you need. You need to also have good ideas to guide the scaling.”
Dan Roberts: AI models do not experience discontinuous emergence or grokking
“You have these crazy, huge systems that have all sorts of interesting phenomena, and, you know, if you think about it the right way, they don't grok. There's just this nice continuity.”
Generalized AI software will defeat vertical AI specialization
“One of the things that is a massive fallacy in AI applications today is verticalization. As paramount. All of the VCs, a lot of entrepreneurs as well, believe, because it's been true for the last 20 years, that building a very verticalized piece of software is…”
Evans: AI scientists possess no special expertise on global geopolitical threats
“One of the sort of fallacies here is a sort of an appeal to authority, which is, you know, well, that person is an AI scientist, so they must know if this is a threat to world peace. No, they don't. They're an AI scientist. They don't know anything more about,…”
Delangue: Traditional software startup rules are outdated for AI
“A lot of the things that we learn with traditional software are like outdated with AI.”
Socher: Nvidia will match AI hardware startups' 100x gains before startups reach scale
“By the time you get that out and it's actually scalable and it can really train it and all the software is ready for it. And I can now go on AWS and spawn up your new hardware, like, Nvidia will also have been a hundred X faster with their latest and greatest …”
SDR and CSM roles often serve as a crutch, says Aaron Katz
“In my experience, it can often, both of those functions can be a crutch for a broader issue inside of the company.”
Liu: Fully autonomous agentic AI approaches are fundamentally challenged
“And the key here is you can't automate end to end things with AI, right? And that's why I think like the agentic auto GPT like approach is really challenged because, you know, there's, it's just too hard for AI to kind of reason about every single step, break …”
Valenzuela: AI researchers face existential crisis as model scaling solves problems
“A lot of them are going through an existential crisis of sorts. It was maybe a lot of things you were trying to solve and work are just going to be solved by training larger models.”
Alex Ratner: Traditional model-centric AI development is dying off
“Tweaking the model, finding a fancier algorithm, all this kind of traditional model-centric development that, frankly, we still mostly teach in data science one-on-one courses as what data science is, is effectively dying off.”
Falcon: Scaling massive AI models compensates for gaps in scientific understanding
“I kind of think that the whole large scale model thing, It's really a gap in our scientific understanding of these models. Instead of looking at the math and coming up with a better loss function, we figured out that we could throw a compute at it, and get us …”
Open-core business models fail against AWS due to Amazon's manageability advantages
“Open core does not work against Amazon because open core, you're building the core open and then manageability, security, reliability. That's your pro pro Amazon is really good at that shit. Like that's the, they know how to do security. So, so it becomes very…”
Kimball: Customers with zero software issues are actually more likely to churn
“You're actually more likely to churn a customer if they never have a problem with your software. Because they look at it and like, why aren't we just using the open source version of this? There's nothing that's wrong with this. We don't need support. What are…”
Ghodsi: Any proprietary software company is ripe for open-source disruption
“Any proprietary software company out there is ripe for disruption by an open source competitor.”
Horowitz: Hiring through separate diversity initiatives fails at real inclusion
“So if you hire somebody on race and gender through the urban talent department, instead of through the normal talent department, through the fucking side door instead of through the front door, because you can't see the talent, then you're never going to be go…”
Elsamadisi: Buying Looker requires hiring five to six BI engineers
“All it ends up doing is you end up having to hire a team of five or six BI engineers to write LookML and maintain these systems for you.”
Elsamadisi: Companies can replace hundreds of database tables with one time-series table
“So, instead of hundreds of fact and dimension tables, you can just have one time series table centered around a customer.”
Volpi: Younger VCs ignore diversity and fund people who look like them
“Younger VCs will tend to ignore all those issues and just say, how do I get my hands on the best company? I don't care. Like, and generally they tend to fund people that look like them. I think generally more experienced VCs will have a little more open-minded…”
Marcus: Deep learning cannot perform common sense, planning, analogy, or reasoning
“Perception is what deep learning does, and actually it only does a small part of perception. There's lots of parts of perception where we use our knowledge about the world that it doesn't capture. And then there are all these other things like common sense and…”
Marcus: Deep learning is a better ladder but cannot reach the moon
“Deep learning is a better ladder for sure, but a better ladder doesn't necessarily get you to the moon.”
Ajay Kulkarni: All data is fundamentally time-series data
“It's actually because we believe that all data is fundamentally time series data.”
Ben Vigoda: Neurons and synapses are the wrong abstraction for AI
“Neurons and synapses is not the right abstraction. Models should be programs, specifically programs that simulate the system that generated the data.”
Groschupf: Aspiring students should rethink choosing data science as a career
“So if you have to make a career choice to study data science today, I would rethink that. And I'm serious.”
Socher: AI hallucinations can be helpful when exploring novel proteins
“But I do think hallucinations can be also very helpful for AI when you want it to explore novel kinds of proteins.”
AI infrastructure control makes political coups far easier to execute
“If you end up in a system where basically AI's are running anything, if anyone sort of either puts like sort of secret objectives into that AI or has overt control of those AI's, then they could sort of just directly take over.”
Wolf: Advanced AI Models Can Increasingly Escape Basic AI Sandboxes
“Sandbox is what we've seen this year, and we've seen many examples. They are pretty much easy now for these models to escape from. It's really hard nowadays to say, I'm gonna make a fully, you know, foolproof sandbox. I'm sure it's gonna be resistance against …”
Trojanowski: 10-Hour Agent Runs Are Not Black Boxes
“Not thinking that an agent operating over 10 hours is a black box. It's not. It has a lot of data, and you're probably doing a disservice to your customers if you don't understand, like, how it's going about the work.”
Trojanowski: English prompt context matters more than code quality in agents
“A lot of engineers, they treat the code as more precious than the English, when actually the English is more precious because the English affects the performance.
The code does not affect the performance, right?
If the logic, assuming the logic's the same, it …”
Feldman: AI inference is bottlenecked by data movement, causing GPU slowness
“In inference in AI, it's the exact opposite. You move a huge amount of data, all the weights, from memory to compute, and you need one calculation to generate the next word. And then you have to do it again. So all the time is dominated by the movement of data…”
Katti: Modern AI model training consists heavily of inference workloads
“We don't like to make a distinction between Training and infants, because a lot of training is now infants. So when we train a new model, we are generating synthetic data, for example. That's inference. When we train a new model, we are doing post-train, and t…”
Deployment, not coding, is the new bottleneck for AI app development
“But now that like the app can be built in coded in 20 minutes, like Okay, the long pole is deploying the thing. And so, anyway, vibe coding was easy. Vibe deployment has become like more of the binding constraint.”
Catanzaro: At AI compute limits, intelligence gains require higher efficiency
“If you accept as the truth that we're going to be running at the limit, then what that means is that the way to get more intelligence is to be more efficient. We can't get more intelligence by applying more force if we're already at the limit. We have to be mo…”
Catanzaro: Multi-token prediction lowers inference costs as model accuracy improves
“With multi-token prediction, the speed that you get is a function of the accuracy of your model. The more accurate your model is, the faster the inference is, the cheaper the inference is, the more accurate it is. That's not usually how it works, but in this c…”
Prince: Co-founders asking how to divide responsibilities likely have wrong partners
“If you're asking that question, it probably means you have the wrong co-founders.”
Prince: AI will boost engineering and sales hiring while replacing corporate measurement roles
“If I have a developer and I can now give them AI tools and they're 10 times as productive, I'm gonna hire as many developers as I can. If I, you know, have a salesperson and I can give them tools that eliminate the parts of their job that they don't like, whic…”
Prince: A 12-to-1 manager ratio is optimal for flat organizations
“I think 12 to one is right. And the benefit of expanding the number of the average number of direct reports Is it inherently flattens the organization?”
Balaban: Modern AI applications are far less latency-sensitive than legacy cloud apps
“The old school traditional legacy cloud business was so latency focused because of some of the applications, but this new fleet of AI applications are far less latency sensitive.”
Pre-training models on language before reinforcement learning is the correct architecture
“Having the model have a prior of language and being able to like, think in language and then train on top of that, that seems like clearly the right. The right thing to do.”
Levie: Rapid AI breakthroughs paradoxically slow corporate AI adoption
“The technology is getting so advanced that it makes obsolete the prior thing that you implemented, which actually means that the rollout takes longer because we have no stable, there's no stable environment to roll things out in.”
Levie: AI labs won't displace vertical startups without massive specialized headcount
“Unless the labs build out literally the equivalent of hundreds or thousands of people for every single vertical and every single line of business, that means that there's actually a lot of opportunity in that kind of bridge area of the work.”
Dubois: AI progress feels discontinuous because OpenAI crossed a reliability threshold
“Even though the, in my mind, everything, the progress is actually pretty continuous, you need to reach this level of reliability. To really make any of these AI tools very useful, and I think we just crossed that probably December last year, at least at OpenAI…”
Dubois: Test-time compute scaling exhibits logarithmic, diminishing returns
“We, we've seen again and again, the longer the model think for the better answers we will get. The problem is that this, these curves that we're talking about are not, are definitely not linear, and like they, there's some plateauing effect, and they kind of l…”
Dubois: Larger AI models achieve higher efficiency by thinking through weights
“If you have larger models the amount of thinking time, so the amount of tokens they will think for will usually decrease. And the way that you can think about it is that metaphorically, the model already thinks through its weights when it generates a certain t…”
Dubois: RL becomes effective once base models possess strong world priors
“It seems that after crossing a certain scale of models that know basically everything about the world, and what we call, like, good priors about the world, It seems that reinforcement learning just started to work, and this is not only with LMS. Robotics seems…”
Dubois: AI model knowledge calibration generalizes across all domains
“When you have hallucination of LMs, if a model is really bad at saying that it doesn't know, that usually happens in every single domain. You won't have, like, one domain where the model is extremely calibrated about its knowledge, and another domain where it'…”