Alex Amadori

We explore how a coalition of middle powers may prevent development of ASI by any actor, including superpowers. We design an international agreement that may enable middle powers to achieve this goal, without assuming initial cooperation by superpowers.

ControlAI has done a lot in 2025, very proud to be a part of the team!

We've just published our 2025 Impact Report! At a glance: ~1 in 2 UK lawmakers we briefed supported our campaign, for a total of 110+ supporters 2 House of Lords debates on superintelligence & extinction risk A series of hearings at the Canadian Parliament (+ more in thread)

@TheZvi It does mean that, provided said human can think many times faster than normal humans and can be cloned an arbitrary number of times given enough hardware

We can go a step further. The Anthropic constitution is a set of principles for how Claude should act and what its character should be. Does this mean that no human whose parents instructed them to have values other than the literal US Constitution should be in the supply chain?

@slimer48484 Just FYI, the title is clickbait, Max Harms is a promoter of corrigibility. It would be a stretch to call corrigibility "the exact opposite of making AIs good".

The nominative determinism too powerful

Very good post. Anthropic is not acting like you'd expect a responsible org to act. If you take their theory of change at face value, you should expect them not to fight regulation, or at the very least to loudly and consistently make public declarations that bluntly describe the situation with respect to extinction risk. "Help! We're trapped in a death race!" Or even "We've changed our mind, alignment is easy, we think we can navigate an intelligence explosion" But they make no such statement. If you think Anthropic is serious about avoiding extinction risk, you should be very confused. Anthropic's behavior is what you'd expect from an organization that is almost entirely dedicated to building ASI as fast as possible. So they won't seriously advocate for any measure that slows them down. At the same time, they can't just pivot their PR toward dismissing doom in the foreseeable future, since they've built a reputation of being safety-minded that gives them easy access to the smartest researchers. The minor differences between them and other major AI companies are routinely used raw thought-stopper material for employees and shills to justify, to themselves, what they're doing with their time and social capital. However, these differences can only persist on the condition that they are at most a minor inconvenience toward building ASI as fast as possible. The more time passes, the harder things will get. If Anthropic is dropping the ball when we're playing in easy mode, they will do much worse later on.

@svpino x.com/SamuelAlbanie/…

Is it just me, or has ChatGPT become dumber? It might just be that I'm using it more and asking more complex questions, but my feeling over the last couple of weeks hasn't been good. It has become more verbose, repetitive, and it constantly flip-flops to the point where I need to fact-check every single thing it says because I can't trust it. I'm considering moving to Gemini as my daily driver and see what's up.

We built a coalition of 100+ UK lawmakers who are taking a stance against the extinction risk from superintelligent AI and back regulating the most powerful AIs! From the former AI Minister to the former Defence Secretary, cross-party support is crystal clear. Time to act!

@robertwiblin Hey Rob, I recently published a paper on this, would love to chat with you about this! x.com/testdrivenzen/…

Who's the best guest to speak to what middle powers — UK, Australia, Japan, Germany, Korea, Netherlands, Canada — should be doing in light of AI advances and the possible radical impact of AGI or recursively self-improving AI?

@Gabe_cc @andreamiotti @_ebehrens_ And thanks @olysourbut and Mitchell Howe (MIRI) for their very helpful feedback on the paper!

This is a paper by me, @Gabe_cc, @andreamiotti, and @_ebehrens_ Check it out here: asi-prevention.com

We explore how a coalition of middle powers may prevent development of ASI by any actor, including superpowers. We design an international agreement that may enable middle powers to achieve this goal, without assuming initial cooperation by superpowers.

We at the MIRI Technical Governance Team just put out a report describing an example international agreement to prevent the creation of superintelligence. 🧵

Marc Andreessen (@pmarca)'s recent essay, “Why AI Will Save the World”, didn't meet the standards of discourse. ♦️ Claiming AI will be safe & net positive is his right, but the way he’s gone about making that claim has been undermining conversation quality. 🧵 Here's the proof:

the idea of having a chat interface is cute, but it makes it so that we can't see the messages until they have been fully typed which is annoying (and i guess unnecessary unless the AI is backtracking?) the blue bubble color also makes it impossible to select text in the user's messages should have a UI + hotkey way to switch between auren and seren instead of having to ask I can see that it's using web search (or lying about having used web search), but i'd like to see the sources it consulted from 5 minutes trial i like the personalities. the mean one gave useful advice on how not to get stuck on a doomed approach without being patronizing.

we let our AIs tell you you're absolutely wrong please say hi to them, now IN BROWSER! link: chat.auren.app if you'd like a free month pls respond here and i'll DM you!