Quick Takes

Thinking of drafting a post on war crimes, trying to answer the following puzzles: 1. Why do we have a notion of war crimes at all, given how bad war itself is? 2. Why are some things war crimes and not others? 3. Why do precursor notions to war crimes appear, independently, in essentially every culture that has fought wars at scale? 4. Given that essentially every culture has also broken these norms, sometimes spectacularly, why does the norm always come back, and often come back stronger? Common answers to these questions seem profoundly misguided. The naive answer, that war crimes are simply the most horrible things that we all agree is collectively wrong, does not survive even five minutes of scrutiny. More sophisticated versions of that argument also do not survive scrutiny: Just War theory is similarly flawed and question-begging on the descriptivist front, and the Schelling point - shaped argument that war crimes can’t limit all of war’s badness, but are aimed at curbing the worst excesses, does not explain why mass bombings and medieval sieges are/were considered acceptable, but false surrender is not. The “cynical” answers are (differently) flawed. eg some people think war crimes are completely fake and anything other than total war is just modern virtue signaling, ignoring the thousands of years of documented history we have on precursors to war crime (Xerxes in 400s BC: “The Spartans, when they do such things overthrow all law and justice among men.“). If anything, the modern version of “total war” is much newer than the idea of war crimes. Similarly, a naive “power analysis” that war crimes are simply defined by the powerful to limit the options of the powerless ignores that powerful people are often themselves constrained by these norms, sometimes hugely. Instead, my core answer here is surprisingly simple: A “war crime” is, in its oldest and clearest form, the category of acts that destroy the means by which wars can be ended early. The prohibi

So... what's the general take on the hantavirus outbreak?

Yay!!! https://www.wired.com/story/google-deepmind-workers-vote-to-unionize-over-military-ai-deals/

one medium term future that still seems possible is that models continue to be bad at generalization, and so a huge fraction of the economy is AI data labelling for various extremely niche or brand new areas. a world where new problems are solved once by humans and the solution reused for near-free forever via AI. ofc, once generalization is cracked then it's all over. but in the meantime, this could persist for some duration.

There's a meme that "nothing ever happens" that's popular among prediction market traders, with the idea being that the status quo changes less frequently and in ways less according-to-specific-priors than traders first-order expect. I think there's a similar principle that applies to reasoning about AI development and takeoff speeds: nothing ever happens (prior to the development of superintelligence). In non-meme form, I think people closely following AI development tend to systematically overestimate the likelihood and impactfulness of any particular event or change actually happening prior to the development of superintelligence, and this has some interesting implications. For one, semiconductor, energy, and tech stocks are way up, and capital markets more broadly are roaring around AI, despite some geopolitical chaos. But the actual wider economic and societal impacts of AI so far seem surprisingly small, given how smart and easily accessible SoTA models and harnesses are. If you showed a demo of Claude Code or Codex to someone in 2021 and mentioned that it was available to any business or individual for purchase at non-exorbitant rates[1], I think a lot of people would be surprised at how little impact / transformation there has been given (a) how smart the models are and (b) how readily accessible / deploy-able they are. This seems like a win for Eliezer's world model vs. Paul's, and a reason for pessimism about some iterative-deployment takes and plans are more broadly. It now seems more plausible that things will look and feel pretty normal ("nothing ever happens") for the vast majority of people, until they very suddenly don't. For two, I think there's also a related story about a narrower impact on research specifically that goes something like: "we'll learn a bunch from iteration on earlier models, and also the world will look importantly different in ways that we'll be able to leverage to do or automate the most critical safety work before crunch ti

I read the ARC-AGI-3 paper entirely, and I'm unimpressed. The "100% human-solvable, <1% AI solved" is basically p-hacking. They cook their metrics to guarantee high human scores and punish any sub-human score. They also prevent measurement of super-human performance, so in practice it's close to a binary metric of "matches best human or not". There are also a number of incoherences in the stated methodology, but they're non-central. Their metric is: Environment must be solved by at least 2/10 humans. Among the successes, pick the median (¤) of actions taken, that's the baseline (per level of the environment), call it b. Humans are defined as 100% for being the baseline (no analysis of how many humans solve the environment, or whether the average score is 100% or any deeper analysis of human performance). An environment has n levels. Levels are attempted sequentially, in increasing order of difficulty; solving one unlocks the next one. The environment is solved if all levels are completed. If a model doesn't solve a level, it scores 0 on that level (and subsequent ones). If it does solve it in m steps, it receives (b/m)² score. (*) Then take the weighted average of its scores over levels, where level k is weighted k. (*) If the model is better than human (m < b), its level score is clamped at 1.15, but tbh it doesn't really matter. Also, environment score is clamped at 1 for some reason. (¤) They say "upper-median best", which doesn't make sense, and their example is the median of people who solve the environment, so I'm going with that interpretation. There are two problems with this metric: - Human variance. The baseline might be ultra-optimized, close to optimal, depending on the environment; it might also not. In their empirical evaluation of optimal score (probably from human performance not-first-run?), it's clear that the baseline is very noisy. - The way it's calculated punishes sub-human performance quadratically for no reason, and upweighs the hardes

AI drones now a reality and a gamechanger on the Ukraine battlefield. Apparently, the first AI drones were likely being trialled in late 2025 but it likely took until a few months ago to scale up production and iron out problems. This was likely the reason that Ukraine has been able to regain territoriy for the first time in years [plus slthe shutdown of Russian access to starlink] The killzone has now increased from 5 km to ~20-25km and with the new Hornet AI drones could extend up to 150km from the frontline.