Jeff Dean Exits Google with Top Researchers to Found Discovery Loop, an AI-for-Science Startup Targeting Recursive Self-Improvement

Skynet Chance (+0.08%): A well-funded venture led by some of the field's most capable researchers explicitly naming recursive self-improvement as a goal — cutting 'human iteration out of the loop entirely' — directly targets the feedback dynamic most associated with loss-of-control scenarios. The announcement mentions no corresponding alignment or oversight commitments beyond the public benefit corporation structure.

Skynet Date (-1 days): Concentrating elite talent, Alphabet-backed capital, and massive compute on automated experimentation loops plausibly compresses the timeline on which self-improving systems could emerge. Fragmentation of Google's research leadership also reduces the coordination advantage of a single large lab governing such work.

AGI Progress (+0.04%): Automating the full experimental loop and applying AI to AI research attacks the human-iteration bottleneck that currently paces capability gains, a plausible route to compounding progress toward general capability. The founding team's track record (Google Brain, Gemini multimodal, DeepMind) makes execution credible rather than aspirational.

AGI Date (-1 days): Substantial multi-firm funding plus a founding team with direct experience scaling frontier models suggests near-term acceleration in AI-driven research throughput. If even partial automation of research iteration works, it shortens the expected time to each subsequent capability milestone.