One thing that discussions of AI don't seem to have internalized yet is that AI lives on a different time-scale than human beings. Sci-fi predicted this, but we never talk about it now that AI is real. But it has big implications.

@McaleerStephen And how do you propose that to china and other state nations.

@McaleerStephen @Miles_Brundage First get everyone to agree on nuclear disarmament, then use the lessons learned about international coordination to solve the more challenging problem of AI disarmament

@McaleerStephen Humans are really smart and resourceful. There *must* be some way we could figure out how to coordinate a slowdown if it's needed

@McaleerStephen

@McaleerStephen Whatever Anthropic did from opus 4.5 to 4.8 seems to be working pretty well, keep doing that.

@McaleerStephen Literally impossible

@apples_jimmy @McaleerStephen If I were China I'd jump on the option of a slowdown instantly.

"Oh yes please give me more time to build up free solar energy at bullshit scale + research into better local chips (+ lower risk of existential risks by AI)"

@McaleerStephen you can have RSI and "slowdown" at the same time, they are not mutually exclusive.

@McaleerStephen I’d like an option for faster.

@McaleerStephen No, let it accelerate.

@McaleerStephen I cannot even wrap my head around how that actually works

@McaleerStephen And in other news ActiveDirectory and POSIX permissions exist - we have always had rogue system users.

@apples_jimmy @McaleerStephen It's easy Jimmy

We'll use good old fashion diplomac.. oh yeah, nvm

@McaleerStephen Nothing like running out of compute and saying “ok guys let’s all take a step back here…” weird how “the systems we’re building are too powerful…” lands right before a giant IPO

@McaleerStephen The main co-ordination should be between China and the US. Let's say we create an agreement with them. Would you trust the CCP to follow through on that agreement?

@McaleerStephen Are there any ideas yet for how to actually do this?

Because it seems to me that this is practically impossible to monitor worldwide.

@McaleerStephen RSI isn’t some threshold, the proto stages have been in play from the very first agent loop, the core essentials to inference based learning are super simple primitives. You guys do this weird theater shadow dance that I find hard to take seriously. Self-prompt a meta critique

@McaleerStephen Also we can pull the plug. We are thermodynamically grounded. Your argument does impact SpaceX - space data centers are a real skynet threat.

@McaleerStephen no, accelerate