François Chollet, AI researcher and influential developer of the Keras AI framework, is leaving Google after a decade to start his own company.
With more than two million users, his Keras framework has become a cornerstone of AI development and powers numerous critical applications.
The technology runs in recommendation systems for YouTube, Netflix, and Spotify, fraud detection for credit card transactions, and Waymo's self-driving cars. Chollet will maintain close ties to the open-source Keras project after his departure and continue to shape its development from the outside.
Jeff Carpenter, whom Chollet himself describes as an ideal successor, will take over leadership of the Keras team at Google. In a blog post, Google reaffirmed its commitment to the framework, particularly focusing on developing the new Keras 3 version, which is compatible with JAX, TensorFlow, and PyTorch. With the recently launched Keras Hub, Google also aims to further simplify access to AI tools.
Looking beyond language models
During his time at Google, Chollet says he witnessed deep learning evolve from an academic niche into an industry employing millions. He plans to announce details about his new company soon. The AI researcher is also behind the challenging ARC benchmark and its associated ARC prize, which currently offers $1 million to redirect AI development away from pure language models and back toward general AI development. The ARC benchmark tasks are easy for humans to solve and require no special knowledge. However, modern AI systems struggle with these tasks because they're designed to resist solutions through memorization.
Similar to deep learning critic Gary Marcus and Meta's AI chief Yann LeCun, Chollet sees current LLMs as only part of the solution. Recently, some methods have approached average human performance on the ARC prize, but the gap to top human performance remains significant.
