Lila Sciences open-endedness team releases 4 preprints on the creative paradox
ChengleiSi · x · 2026-11-11
Kenneth Stanley's open-endedness team at Lila Sciences announces its first batch of open-sourced fundamental AI research: a vision-focused blog plus 4 preprints that critically examine ML status quo with counter-intuitive findings.
Highlights:
- A newly identified "creative paradox": directly reinforcing creative outputs can limit a model's full creative potential; the team proposes a novel training approach to sidestep it.
- An investigation into the advantages of "fresh eyes" — training an LLM intentionally naïve in certain domains to gain exploration advantages.
More from AGI Musings
- Hinton Lectures 2026 kicks off with 'AI alignment: what's working and what's not' — DKokotajlo · 2026-11-11
- Why AI Slop Is So Frustrating to Read, per Zonder Labs — floguo · 2026-11-11
- Engineer: I used to write code all day, now I think about compute and AI economics — shakoistsLog · 2026-11-11
- "AI nerfs the based and nukes the mids": taste is the only moat left — dosco · 2026-11-11
- OpenAI board member Paul Christiano sees 'material risk' of AI takeover in 6-12 months — connoraxiotes · 2026-11-11
- AI 2027 Team Revises Timelines: METR Report Points to Intelligence Explosion in 2027 — DKokotajlo · 2026-11-11