How François Chollet Is Building A New Path To AGI

Francois Chollet, founder of the ARC prize and Ndea, discusses the current state and future directions of AI research. He introduces Ndea's focus on program synthesis, aiming to create concise, symbolic models as an alternative to deep learning's parametric curves. Chollet argues that while current LLM-based approaches are achieving automation in verifiable domains like coding, true AGI requires human-level learning efficiency across arbitrary tasks. The conversation explores the role of benchmarks like ARC AGI in measuring AI progress, highlighting how each version targets emerging capabilities such as reasoning and agentic coding. Chollet predicts AGI by 2030, emphasizing the need for AI development to trend towards optimality and for researchers to explore diverse approaches beyond the current LLM stack, potentially drawing inspiration from earlier AI research.