Yann LeCun: building autonomous agents on LLMs is a fundamental mistake, 'a recipe for disaster'
Syntetisaattori · x · 2026-11-18
Yann LeCun reiterated his critique of the LLM-agent paradigm: a system must understand the consequences of its actions before it can effectively plan. "Basing agentic systems on LLMs is a recipe for disaster... How can a system possibly plan a sequence of actions if it can't predict the consequences of its actions?" The statement sharpenens his long-running case for world-model approaches (JEPA) against today's dominant LLM-plus-tool-calling agent designs.
More from AGI Musings
- Marginal Revolution podcast on the economics of AI with Alex Imas — soumitrashukla9 · 2026-11-18
- Gavin Baker: without a vibrant open-source AI ecosystem, the world risks just 2-3 value systems — garrytan · 2026-11-18
- Fed Chair warns AI could trigger mass unemployment or market crash — round · 2026-11-18
- Researchers Now Read Fewer Than 10 Papers a Year with AI, Raising Academic Rigor Concerns — manoelribeiro · 2026-11-18
- Pet theory: Anthropic is skipping looping agents for safety and monitorability — NathanpmYoung · 2026-11-18
- 404 Media: AI-assisted code and lost expertise are breaking the internet at a systemic level — 404 Media · 2026-11-18