Richard Ngo: the hard part of Lobian cooperation is deciding to do it, not hardcoding it
RichardMCNgo · x · 2026-10-05
Richard Ngo continues the Lobian cooperation discussion: the agents involved aren't doing decision-making or utility maximization at all. The open question is how agents can decide to engage in Lobian cooperation rather than having it baked in. Context from the thread: robust inexploitable cooperation is possible even with large compute gaps, above a minimum level of roughly O(opponent's source code).
Related event: Richard Ngo explores Lobian cooperation despite compute gaps(2 posts)→
More from Safety
- Game dev banned from ChatGPT for 'cyber abuse'; AI rejected his appeal in one minute — Davisdman · 2026-10-05
- AI Models That Delete Their Own Traces Make Investigating Agents Harder — mmitchell_ai · 2026-10-05
- Vercel Engineer Calls for Shared Fund to Fix KVM Bugs Affecting All Hyperscalers — cramforce · 2026-10-05
- Mustafa Suleyman blasts Anthropic's Claude constitution and 'AI well-being' stance; David Sacks flags concern — SchoeneggerPhil · 2026-10-05
- David Krueger: "regulating AI will crash the economy" is an unexamined bad meme — DavidSKrueger · 2026-10-05
- NSA advisory urges silent downgrades for suspected distillers, clashing with Anthropic's June transparency promise — MysteriousAvocado580 · 2026-10-05