Matthew Green: Even experts haven't stress-tested what open models can actually do
matthew_d_green · x · 2026-11-08
Cryptographer Matthew Green argues that surprisingly few experts have actually sat down with open-weight models and sent them off to do tasks the way a novice would, so the community doesn't really know their capability ceiling for exploits. He also engages a second argument — that exploits don't matter because existing criminal groups are well-understood or are Russia — saying he's open to it and would even like it to be true.
More from AGI Musings
- Pascal Bornet coins 'AI Obesity': asking AI what to think erodes your judgment — sanderssays · 2026-11-08
- AI Safety Researcher Pivots From Preaching Short Timelines to Questioning the Evidence — JacquesThibs · 2026-11-08
- YC Paper Club: AI for Biology, From Reading Data to Designing Living Systems — ycombinator · 2026-11-08
- Nina Schick: AI is industrialising intelligence, echoing the Industrial Revolution's escape from Malthusian trap — NinaDSchick · 2026-11-08
- Daniel Faggella: positive AGI outcomes require governance, not saintly leaders — danfaggella · 2026-11-08
- NYT: OpenAI results upend a key tenet of computer science — P vs NP — _akpiper · 2026-11-08