There’s a lot of good insight in the Liang Wengfeng interview, most exciting to me that they don’t use OKRs! I should’ve probably listened to it Chinese, need to keep those neurons firing. Dario on DeepSeek and export control, which whew! Says a lot! The notes about AI progress are really worth understanding, but it’s silly that people are just now learning about the PRC. The Paris AI Summit was not a hit, says Anthropic. Decent analysis on the virality of DeepSeek.

There is finally a new pass at understanding what ASI misalignment or misuse looks like in detail. It’s probably extremely hard to make predictions around “this is how it’ll go”, but I think there’s a lot of clear predictions you can make in the shape of “given this plausible outcome of a particular variable, this is how it’ll go from there”. The Anthropic Economic Index serves a similar goal in providing bits that prepare us for different futures. I’m way less worried about how ASI will fit into our economy than our governance though. Anthropic jailbreaking paper — it works! You can just stop jailbreaking!

Good time to learn about the original Thinking Machines which seemed utterly incredible. The MiniMax attention paper. Another great eval (Humanity’s Last Exam), though the name is far too ambitiously great. I’m pleased with calibration reporting, though I asked for a bit more. OpenAI comes forth with computer use. The frontier math drama was, in my opinion, not a huge deal in of itself (especially after considering all the nuances) but is pretty strong foreshadowing about what’s to come. A simple sama post, with an extremely cute last sentence — but I’m still waiting to someone come forth with a strong pitch for neo-georgism in an ASI regime and it seems like sama might miss that train despite all the setup he’s done. OpenAI releases a lot of good content about evaluating on contest programming. Maybe now’s a good time to share that dream I had two+ years ago where the only way to defeat the unaligned AGI was to beat it at Codeforces and so I started a camp in the Canadian forest where we had to be awake at weird times to do the Russian contests?

I’ve always noticed a ton of evidence that people these days in many way aren’t as competent, agentic or even as emotionally healthy as they used to be. I’ve never attributed it to the way we raise children, but this essay was poignant on the matter. Weird that an American political faction can win being anti-child sexual exploitation by digging up huge news from a decade ago, from across the world. The Rotherham child sexual exploitation was uniquely horrific though. On a brighter note of things Indo-Aryans did that we didn’t pay enough attention to is Kumbh Mela 2025. At least 400 million attendees over 45 days, with an estimated 7M present on any given day. I think it’s an insane, sordid and beautiful human feat. It’s a super special place, with drone shows, an android app, 40,000 police, 150,000 toilets, 2,300 cameras, and underwater drones? Even though I think religious events likely have much lower rates of crowdrush, they definitely cover up deaths with only 30 reported. Lauren Powell Jobs went but did not bathe for health reasons (though supposedly, the water Does Not contain feces). I rant about it here because of how underreported it is.

Jane Street wrote too little on Dune, their build system! It used to be my dream to work on build systems and performance of devtools, and is still sort of beautiful to think about. I will definitely be rewriting Bazel recreationally after the singularity. An analysis on strategy and decisiveness in organisations.

Why our butts jut out. I learn about why I never want to try to buy a house. Reminder that UA Flight 93 happened and should be a huge part of the terrorism-zeitgeist. Year-old Matt Levine has a really good bit about AI researchers that aged well. Really impressed with how accurately Matty can clock the way our industry works from the outside. The Hiroshima Maidens happened? They just sent disfigured women to America for plastic surgery and press (not in that order), with the end results being pretty good and one accusation of doing human experiments (haha). I learned that one of the campaigns into China was called Operation Ichi-Go which is not militarily interesting it’s just funny that it sounds like “strawberry”. EGG PAPER! Which for some reason is in the Communications Engineering section of Nature! 🥚