Happys Diwali, @tszzl 🪔 AI vs nuclear safety debate could use better framing, imho. True, regulating AI is way harder than regulating nuclear weapons; but AI on a toaster is not going to cause mushroom clouds and vaporize a million people in a second. So not the same thing. The most dangerous AI - by definition - is going to be at big labs with practically infinite compute. If they fail to align it, then we shall have issues. The smaller models at a fraction of compute - even if they try to take over the world - will be no match to intelligence of the big models, who shall crush them. Unregulated AI has its own issues - which can all be very serious - but, even the worst cyber incidents are not the worst ones of history, and the digital infrastructure is being readied. Thanks to OpenAI, Ant and other labs are enabling defenders. If you want to imagine AI will take over power infra, airplanes, missiles and weapons etc. etc. it needs discontinuity in one’s logic. The world will evolve to live with AI, the way it evolved to live with fire, electricity and hydrocarbon fuels. Yes, it won’t happen automatically and we have a role to play and lot to do. Thank OpenAi for “Patch the plane”. Also, lot of great Thought Leadership by Ant. Also, thanks to likes of Moonshot for showing responsible behavior towards cyber safety.