François Chollet
Co-Founder, Ndea
- AGI
- Deep Learning
- Benchmarks
Editorial bio
François Chollet is a French software engineer and AI researcher. Diplôme d'Ingénieur from ENSTA Paris. He created Keras in 2015, which became the high-level API for TensorFlow and the on-ramp for an entire generation of practitioners; also developed Xception and other contributions to the deep-learning stack. Senior Staff Engineer at Google from 2015 to November 2024, author of Deep Learning with Python (a widely used textbook).
In 2019 he introduced the Abstraction and Reasoning Corpus (ARC-AGI) – a benchmark designed to measure genuine generalisation and intelligence rather than memorisation, and still unbeaten by frontier LLMs. In 2024 he co-founded the ARC Prize Foundation and Ndea with Mike Knoop, aimed at pursuing AGI via program synthesis and better measurement of intelligence, and left Google to focus on these efforts. Named to TIME100 AI; a leading public voice on the limits of pure scaling.
Worth following if you want the AGI argument from the person whose benchmark is actively humbling the frontier labs.
Key facts
- based in
- United States
- current
- Co-Founder, Ndea · Co-Founder, ARC Prize Foundation
- prior
- Senior Staff Engineer, Google (2015–November 2024) – creator of Keras
- education
- Diplôme d'Ingénieur, ENSTA Paris
- author
- Deep Learning with Python (widely used textbook)
- recognition
- TIME100 AI
- focus
- AGI definition & benchmarking · program synthesis · ARC-AGI · neurosymbolic approaches
Featured work
Keras (Creator, 2015–present)
High-level API for TensorFlow; one of the two most widely used deep-learning frameworks worldwide.
ARC-AGI benchmark
Reasoning benchmark designed to measure generalisation; still unbeaten by frontier LLMs.
ARC Prize Foundation (Co-Founder, 2024, with Mike Knoop)
Public competition pushing progress on ARC-AGI.
Ndea (Co-Founder, 2024, with Mike Knoop)
AI lab pursuing AGI via program synthesis and better intelligence measurement.
Deep Learning with Python
Standard textbook on applied deep learning.
Xception
Deep CNN architecture; influential in computer vision.