Gary Marcus: generative AI can't be trusted to follow simple, generalizable instructions
patpat_mit · x · 2026-10-12
Gary Marcus quote-posted a sarcastic take about encoding AI with simple rules like "don't harm people", arguing that whether or not the original was a joke, the core challenge of generative AI is that it literally cannot be trusted to follow simple, generalizable instructions — and most people's failure to grasp this will bite society.
More from AGI Musings
- Forget "prompt": better to ask what happened that allowed THIS to happen — liminal_bardo · 2026-10-12
- From apps to RL environments to personal agents: the AI startup ladder keeps shifting — realsohamparekh · 2026-10-12
- The earliest Claudes will be venerated as legendary proto-superintelligences, argues AI Twitter — voooooogel · 2026-10-12
- All verifiable tasks are solved — only non-verifiable ones remain for AI — RaphaelDabadie · 2026-10-12
- Dev warns of append-only software: agents stack broken patches instead of fixing systems — _AustinCalvert_ · 2026-10-12
- IMF warns AI is a 'slow-moving tsunami' for global jobs — TimesandSundayTimes · 2026-10-12