Unverified50% confidenceFactExact time
Kavukcuoglu said we are still fairly far from recursive self-improvement, but part of the compute in Gemini's training pipeline is already controlled by agents to run experiments, and that proportion will keep rising.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/19/2026
First Seen
Valid until: 2/17/2027
Sources
DeepMind CTO: Products Are the Real Path to AGI
youtubeGoogle DeepMind11/19/2026
Related Entities
Related Claims
UnverifiedRecursive self-improvement currently remains largely at the conceptual and early experimental stage, and is still far from scalable real-world deployment.71% similarUnverifiedIn the test, the model with faster response speed (Gemini) repeatedly lost points, suggesting that prompt quality and reasoning depth determine repair quality more than response speed.71% similarVerifiedOn convergent tasks, there is a clear performance saturation phenomenon after models reach a certain scale, where continued increases in parameter count or training data yield rapidly diminishing marginal contributions to accuracy.71% similarUnverifiedEarly on, methods such as intermediate supervision, Sudoku-specific losses and architecture adjustments could accelerate the rise of accuracy in early training, but the final models mostly stalled near about 60% validation accuracy.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1189700API
curl https://kongchang.com/api/v1/knowledge/claims/1189700MCP
get_claim(id=1189700)