LLMs can't invent jazz, but agents interacting with humans might
dh7net · x · 2026-11-19
Responding to the claim that "an LLM fed all music up to 1900 would never invent jazz," the author argues that while a standalone LLM is bounded by its training data, agents powered by LLMs that interact with humans and other agents will likely invent new art forms. The exchange touches on a core debate about generative AI's creative ceiling: novelty may emerge from social interaction structures rather than the model itself.
More from AGI Musings
- Inaugural YC Debate: Ex-Anthropic Researcher vs Economist GMU Robin Hanson on Slowing Frontier AI — ycombinator · 2026-11-19
- DeepMind Institute Publishes Essay Mapping the Path From AGI to ASI — ShaneLegg · 2026-11-19
- Daniel Kokotajlo: preventing power concentration is pointless if we lose control to superintelligence — DKokotajlo · 2026-11-19
- Skeptic pits METR's AI extrapolation against the 1990s CPU clock-speed wall — SporksInjected · 2026-11-19
- MIRI's So8res publishes 'Anthropic is not living up to its ideals' on IPO day — trevposts · 2026-11-19
- Ray Bradbury's grandfather eulogy goes viral in AI debate over human worth — CatAstro_Piyush · 2026-11-19