Alignment is a continual learning problem: AI models must be 'raised' with feedback
apsarathchandar · x · 2026-09-22
Quoting Thomas Dietterich: decades of automation research show autonomous systems never anticipate all failure modes and require continual monitoring and human oversight. The sharer adds that safety and alignment are continual learning problems—AI models must be 'raised' with constant feedback that lets them correct mistakes, like children.
Related event: Automation Veteran: AI Safety Requires Continuous Human Oversight(3 posts)→
More from AGI Musings
- Togelius: saving peer review from AI slop requires ending anonymous submissions — TuhinChakr · 2026-09-22
- Toby Ord's Swarm Scaling: how do AI agent swarm capabilities scale with more agents? — tobyordoxford · 2026-09-22
- Researcher pushes back as million-follower academic says "let AI wipe out mathematicians" — thegautamkamath · 2026-09-22
- Nature feature: AI co-scientists now generate hypotheses and design experiments, humans still judge — burny_tech · 2026-09-22
- AI slop is everywhere: market banners, business emails, even 70% of a journal paper — analisereal · 2026-09-22
- OpenAI's Will DePue on Why AI Chat Feels Flat: No Interruptions, No Edited Messages, and Models Aren't Funny — willdepue · 2026-09-22