·
Researchers & Thinkers
Francois Chollet
Co-founder |Ndea & ARC Prize
Creator of Keras, author of 'Deep Learning with Python', and architect of the ARC-AGI benchmark for measuring general intelligence.
3papers
GitHubfcholletResearch Papers
3 papersBiography
Deep Research
Francois Chollet is a French software engineer and AI researcher, born October 20, 1989. He graduated with a Diplome d'Ingenieur (Master of Engineering) from ENSTA Paris, part of the Polytechnic Institute of Paris, in 2012. He is the creator of Keras, one of the most widely used deep learning libraries with over two million users, and the author of 'Deep Learning with Python' (Manning, 2017; 2nd ed. 2021), which sold over 100,000 copies and was translated into 10+ languages. Chollet joined Google in 2015, shortly after releasing Keras, and rose to Senior Staff Engineer over a nine-year tenure focused on computer vision, formal reasoning, and abstraction. In 2019, he published 'On the Measure of Intelligence' (arXiv:1911.01547), introducing the Abstraction and Reasoning Corpus (ARC) — a benchmark measuring general fluid intelligence via novel reasoning problems that humans find easy but AI struggles with. In 2024 he co-launched ARC Prize, a $1M competition that raised the ARC-AGI state-of-the-art from 33% to 55.5%. He departed Google in November 2024 and in January 2025 co-founded Ndea with Zapier co-founder Mike Knoop — an AI lab pursuing AGI through deep learning-guided program synthesis. The ARC Prize Foundation expanded into a non-profit in early 2025, with ARC-AGI-2 released in March 2025 and ARC-AGI-3 (an interactive reasoning benchmark) launching March 25, 2026. On December 1, 2021, Chollet won the Global Swiss AI Award for breakthroughs in AI.
Artificial General Intelligence (AGI)Abstraction and ReasoningIntelligence MeasurementNeural Architecture ResearchProgram SynthesisKeras and Deep Learning FrameworksComputer VisionNeural Architecture DesignAI BenchmarkingCognitive Science and Developmental Psychology
Timeline
22 Research22 total
2026
2025
2024
2024-06Research
Appeared on Dwarkesh Podcast with Mike Knoop: 'LLMs won't lead to AGI — $1,000,000 Prize to find true solution'. Announced ARC Prize.2024-06Research
Appeared on Sean Carroll's Mindscape Podcast #280: 'Deep Learning and the Meaning of Intelligence'.2023
2021
2020
2019
2017
2016
2015
Key Contributions
Keras
Created Keras (2015), the most popular high-level deep learning API with 2M+ users. Originally multi-backend, it became TensorFlow's official high-level API, then returned to multi-backend support (TF, JAX, PyTorch) with the Keras 3.0 rewrite in 2023. Keras democratized deep learning by making neural network experimentation accessible to non-specialists.
ARC-AGI Benchmark
Designed the Abstraction and Reasoning Corpus (ARC-AGI), introduced in the 2019 paper 'On the Measure of Intelligence'. ARC measures general fluid intelligence via novel visual reasoning tasks that humans find easy but AI struggles with. ARC-AGI-2 (2025) and ARC-AGI-3 (2026) progressively raise the bar.
ARC Prize Foundation
Co-founded with Mike Knoop, a non-profit offering $1M+ in prizes to advance AGI research. The 2024 competition drove state-of-the-art from 33% to 55.5%. The 2025 competition attracted 1,455 teams. Notable donors include xAI, Google, and Tyler Cowen.
'On the Measure of Intelligence' (2019)
Seminal paper proposing a formal framework for measuring intelligence as skill-acquisition efficiency, grounded in Algorithmic Information Theory. Introduced concepts of scope, generalization difficulty, priors, and experience — foundational for ARC and the broader debate on LLMs vs. true intelligence.
Xception Architecture
Introduced depthwise separable convolutions as the basis for the Xception network (CVPR 2017), an influential architecture that improved efficiency and accuracy in computer vision. The concept became standard in mobile and efficient network design (MobileNet, EfficientNet).
'Deep Learning with Python'
Bestselling book (Manning, 2017; 2nd ed. 2021) that sold 100k+ copies and was translated into 10+ languages. Widely regarded as the definitive practical introduction to deep learning, co-authored 'Deep Learning with R' (2018) with J.J. Allaire.
Ndea
AI lab co-founded in January 2025 with Mike Knoop (Zapier co-founder), pursuing AGI through deep learning-guided program synthesis. The name derives from Greek 'ennoia' (intuitive understanding) and 'dianoia' (logical reasoning).
Notable Quotes
“
Skill is not intelligence. And displaying skill at any number of tasks does not show intelligence.
“
The next frontier is abstraction and reasoning, which will enable extreme generalization and decent sample efficiency.
“
Forget about scale, forget about benchmarks. Using human exams to evaluate AI models is a terrible idea.
“
I'm joining forces with Mike Knoop to start Ndea, a new AI lab. Our focus: deep learning-guided program synthesis. We're betting on a different path to build AI capable of true invention, adaptation, and innovation.
15 sources(click to expand)
Francois Chollet — Personal PageFrancois Chollet — WikipediaAI pioneer Francois Chollet leaves Google — TechCrunchAI researcher Francois Chollet founds a new AI lab focused on AGI — TechCrunchARC Prize FoundationARC Prize 2024: Technical Report — arXivARC Prize 2025 Results and AnalysisOn the Measure of Intelligence — arXivDwarkesh Podcast: LLMs won't lead to AGILex Fridman Podcast #120: Measures of IntelligenceLex Fridman Podcast #38: Keras, Deep Learning, and the Progress of AISean Carroll Mindscape #280: Deep Learning and the Meaning of IntelligenceNdea — AboutDeep Learning with Python — ManningKeras GitHub Repository
Research generated March 19, 2026
Researchers & Thinkers/Francois Chollet
All Profiles