OpenAI joins Anthropic in thinking pausing may be needed 👀

"there should be an international organization that helps [...] make it possible for the world to take coordinated action, including slowing frontier development when needed"

Both Anthropic and OpenAI mention the possibilities of slowing AI development in their latest "what comes next" in AI posts, but say they need to be an action coordinated across the entire world using as-yet-unidentified methods.

1. Pausing until we had a much firmer idea of how to make AI safe—and well-aligned with human values—might well be a good idea. but 2. I trust neither of these companies, but fear that both will say “what about China”, if push comes to shove, and never actually agree to pause.

So it comes off to me as pre-IPO marketing, in effort to address public backlash.

I'm glad to see OpenAI also being public about the importance of preparing for a coordinated slowdown.

OpenAI and Anthropic want to pause global frontier model development

China would never agree If they did they would forever be behind these closed model providers 🤷‍♀️

Sam Altman and Jakub Pachocki wrote today in a blogpost that "[they] have long believed there should ultimately be an international organization" whose goals would include "tak[ing] coordinated action, including slowing frontier development when needed"

@peterwildeford Huh. That's more enthusiasm from AI companies for a global pause than I median-expected to see this year.

@emollick @dwallacewells Let me translate…

“We have an IPO, and we have no moat, so we are asking the world to agree to make us trillions by kneecapping our competitors…”

Are you freaking kidding me?

Are we goi f to pretend this is okay.

WTF.

Liars. Through and through. •

The coordination problem is the actual crux — it's structurally similar to arms control but harder, because there's no clear 'weapon' to monitor and no clean verification layer. Nuclear treaties worked partly because detonations are observable events. AI capability jumps aren't. The 'unidentified methods' may end up requiring compute governance at the hardware/data center layer — the only chokepoint that's currently observable.

@peterwildeford Until I see actual action in this direction to make it happen (they have the resources), then I’ll file this under “wishful thinking but won’t actually do it” at best. We’ve heard this all before…

Also I wouldn’t say “slowing frontier development” is equal to a pause.

@Chaos2Cured @emollick @dwallacewells If they think China will go along with that, they are even dumber than I thought.

@bindureddy The question: Why?

@bindureddy I'm worried about the possibility that the US and China agree to a pause but China fails to enforce the pause against its own companies. China has a history of not enforcing laws when outsiders would be the primary beneficiaries from said enforcement.

@peterwildeford If they really want coordinated action, they should provide the public with the best AI they can. People + AI will be the best coordination there can be. How governments coordinate - you can go look in Ukraine.

@peterwildeford Yes, they want government to swoop in and enshrine their oligopoly in law.

@peterwildeford They are saying this because they know a pause is untenable. Unplug it.

@emollick This is also what I would be saying if I thought I was about to have a potential slowdown in model improvement.

@emollick 'We'll slow if coordinated globally' = mutual insurance. Both competitors have cover: 'We tried.' Coordination never comes. They accelerate indefinitely.

hard to have an international organization when Chinese scientists are starting to be put on travel bans. You’d end up with the apparatchiks wasting your time

@emollick Isn't this the same as regulatory capture? They try to slow up others to keep their share prices high. If competitors catch up share prices drop.