🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Internal deployment was always the main threat model IMO. IIRC I tried to get the Preparedness Framework to cover internal deployment too, but was overruled. It's one of the ways in which this whole evals thing could end up being just safetywashing. (In the AI 2027 scenario, the

The AGI companies are getting more and more secretive about progress. I'm worried that we won't realize that the intelligence explosion is under way until its too late.

A common argument against fast AI takeoff is that bottlenecks like regulation will slow things down. But my view is that the bottlenecks will only slow early AGIs, leading to relatively small effects on the world until ASI steamrolls the government.

New Unsupervised Learning with AI Futures Project Researchers Daniel Kokotajlo and Thomas Larsen on: - Unpacking AI 2027 - Current alignment research and what's working / not working + challenges ahead - The path to AI dramatically accelerating AI research - The race between model

JD Vance told Ross Douthat that he read the AI 2027 paper

A clarification about what I meant by "scenario": - events attached to a timeline - descriptions of what the most important things going on in AI at each point - all the way from now until superintelligence I genuinely would love to see the Gary Marcus scenario + think it would

Obviously correct take. "but bottlenecks" is just the latest in a long line of cope.