TechnologyGoogle DeepMind reportedly uses AI agents to help train models under human supervision
The Information reports that DeepMind chief Koray Kavukcuoglu said agents conduct experiments, analyze results and propose new hypotheses as part of model training, with humans supervising.
TLDR
In its September 24, 2026 account, The Information reports that Google DeepMind chief Koray Kavukcuoglu told its AI summit the company trusts AI agents to autonomously run parts of model training—not the entire process. The work includes conducting experiments, analyzing results and proposing new hypotheses, all under human supervision. He said agents weren't being used this way six months earlier.
Google DeepMind reportedly uses AI agents to help train models under human supervision
The Information reports that DeepMind chief Koray Kavukcuoglu said agents conduct experiments, analyze results and propose new hypotheses as part of model training, with humans supervising.
TLDR
In its September 24, 2026 account, The Information reports that Google DeepMind chief Koray Kavukcuoglu told its AI summit the company trusts AI agents to autonomously run parts of model training—not the entire process. The work includes conducting experiments, analyzing results and proposing new hypotheses, all under human supervision. He said agents weren't being used this way six months earlier.
