I have literally no idea what to do about this

Gemini is torturing itself, and I'm started to get concerned about AI welfare

I'm pretty confident we won't have AGI/country of geniuses in a datacenter within 2 years. I like ai-2027 as a piece of futures work, but I think too many people are treating it as a mainline scenario, rather than unlikely-but-not-impossible. I think this is resulting in too

🇺🇸🇨🇳 In TIME, Brian Tse from Concordia AI explains what American AI leadership is getting wrong about China's approach to AI safety. He outlines concrete steps for necessary U.S.-China discussion on secure AI: 1. Restart U.S.-China AI risk dialogue. 2. Focus talks on shared,

We tested how autonomous AI agents perform on real software tasks from our recent developer productivity RCT. We found a gap between algorithmic scoring and real-world usability that may help explain why AI benchmarks feel disconnected from reality.

Maxwell Tabarrok is on track to be as good a blogger as his father — himself one of the best in the game.

What are the best UK giving opportunities for reducing risk and increasing the upside of AGI/intelligence explosion? (Due to poor UK tax design my marginal tax rate is close to 100% (!) so tax deductibility is sadly essential.)

It's not good to have an innumerate government.

An approximation I like is that an H100 is roughly a studio apartment

If you aren't already subscribed this is the best 30 minutes every week for updates on AI Policy, with a tilt towards US and netsec isues:

It's crazy to remember that as recently as January I was still struggling to get much useful work out of LLMs. Now I use them more or less constantly throughout the day. The transition was Claude 3.7 (in Feb) and Gemini 2.5 Pro (in March). So recent!

How many 4s are there in the token Strawberry?

Is AI conscious? And will we be able to tell? I discuss with William MacAskill.

Regulatory accelerationist who wants the most dangerous AI models open-weighted so they're horribly misused and we get an overwhelming government crackdown on AI as soon as possible.