AI companies are racing to reach recursively self-improving superinteligence. We dont know when they will reach point of no return - it could be in a few weeks, few months, maybe in a year. If everything was like a month ago + Astra, we would be doomed.
But there is hope. A tweet from a now former Anthropic researcher has reached 160 million views. In UK, over 70 MPs and peers are urging the goverment to speedrun the ban on superintelligent AI, just two days after it was introduced on 9th september. Thats 3 good news in 3 days. Bernie Sanders is going to propose simmilar legislature in USA, also consulted with ControlAI.
I see it myself in the czech PauseAI. The growth of the AI-safety movement is hyperbolic. In last few months we got MPs and senators to talk with us, volunteer numbers are rising rapidly, and major media outlets covered the tweet. Something is shifting in society, its not just that having more people and proper division of work allows us to do more, although thats also important.
These events dropped my personal subjective, largely vibe based p(doom) from its highs after Hugging face incident and Astra release from possibly >80% (10% self-aligment + 10% pause) down to 60% (10% self-aligment + 30% pause, extinction as default path, as proven by hugging face). Doubling my p(survival).
We are entering the final battle. Next few months will decide the fate of humanity, and all possible life in reachable universe. While we can no longer prevent AI from creating massive economic and social disruptions, we likely can still prevent extinction.
Politics happens slowly, until it happens very fast. A good example of that was the fall of USSR and communist regimes in Europe. When conditions were right, the revolutions snowballed very quickly. Wast majority of the population hates AI, and does not want to die, overtoon window is moving fast. AI is still constrained to human operated infrastructure. The goverments have power to force AI companies to stop. We just need to direct the overwhelming but covered public hate and growing fear towards real AI-safety legislature and enforcable global treaties.
I am not promising we will be there in 5 years. We may all die in 3 months. But I see a path to survival opening again. We were unlucky for years with AI developing faster then nearly anyone expected, and aligment being unsolvable in time we have. Why cant we be lucky for once?
I ask everyone who hasnt started yet to act quickly (and those who are acting to act even more)- write about what is happeing on social media, repost explainers why ban on ASI is crucial, send them to friends. If your country has AI-safety advocacy groups (PauseAI, ControlAI etc.), join them, unless you already work for AI-safety advocacy. If there are none, coordinate people around you, contact your parliaments and goverments, speedrun protests if you wish, shift the public discourse. Do work you are good at, and outsource work you are bad at to others. I am not the right person to tell you exactly what to do, but there are plenty who can. Overtoon window has moved, some politicans have seen that tweet, some of them will listen, and some will introduce the neccessary legislature to your parliaments.
There is still a hope that no one will build it and not everyone will die. Lets make sure that misaligned ASI never happens.
