Francois Chollet, a leading AI researcher at Google and the creator of Keras, the deep-learning framework, announced his exit from the company. Chollet took to X to share that he would continue to advise Google.
Chollet’s departure from Google marks a significant development. Other notable exits this year include Julian Schrittwieser, a leading AI researcher who left Google DeepMind to join Anthropic. Additionally, Shailesh Prakash, vice president of Google News, stepped down over growing AI and news traffic disputes. Last year, Geoffrey Hinton, often referred to as the godfather of AI, also parted ways with Google.
“Today, we’re announcing that Francois Chollet, the creator of Keras and a leading figure in the AI world, is embarking on a new chapter in his career outside of Google. Keras is more than just a tool—it’s a community driving AI innovation,” Google posted on its official blog. “As Francais Chollet begins a new chapter, our commitment to Keras and open-source AI remains strong.”
How Keras has touched millions of lives
Google noted that Keras has over two million users and has become essential in AI development. It simplifies workflows and makes advanced technology widely accessible. It supports applications like Waymo, YouTube, Netflix, and Spotify.
Chollet was dedicated to Keras, ensuring its support for JAX, TensorFlow, and PyTorch, while continuing to guide its development. Google’s commitment to Keras 3 highlights its support for major ML frameworks, offering developers flexibility. The launch of Keras Hub further democratises access to AI tools, speeding up the creation of innovative multimodal applications.
Chollet’s Arc Evaluation
Chollet introduced the ARC-AGI evaluation in 2019. His research paper ‘On the Measure of Intelligence’, aimed to fairly measure human-like intelligence in AI systems. His Abstraction and Reasoning Corpus (ARC-AGI) remains the only formal benchmark for AGI. He hypothesised that it would be difficult to beat, and it still remains unbeaten.
Recently on X, Sam Altman hinted that they might have achieved this benchmark internally, however, Chollet disregarded this claim as premature. Earlier this year, he introduced the ARC Prize 2024, offering a $1.1M prize pool to advance AI innovation.
In the first ARC-AGI competition in 2020, the winning team achieved only a 21% success rate. In 2022, Chollet and Lab42 launched ARCathon 2022, with 118 teams competing globally. In ARCathon 2023, 265+ teams from 65 countries competed, with two teams tying at a 30% success rate.
