LLMs and Agent Systems
François discusses the limitations of current LLMs, emphasizing their unreliability when used in agent workflows. He points out that while LLMs can provide useful guesses, the compounding of these guesses in agentic systems often leads to inaccuracies. The potential for LLM-based software assistants is acknowledged, but true reliability and autonomy in agent systems remain an empirical question yet to be resolved.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Francois Chollet - ARC reflections - NeurIPS 2024
Related Questions