AI helping build more capable AI: Sergey Brin says Google models are reaching a critical moment of self-improvement
A cycle of model development
Recursive self-improvement, or RSI, describes AI systems helping to build better AI. That can create a cycle that accelerates the development of new models. Brin cited Google’s Flash model releases before Gemini 4 as evidence of progress. He said the company was “starting to get onto that RSI treadmill.”
Researchers estimate a timeline
Anthropic and OpenAI have also said AI may be approaching RSI. A Gladstone AI report published this week describes how AI companies are moving toward the process. Most researchers interviewed for the report, who work at Anthropic, OpenAI and DeepMind, believe RSI could begin within 3 to 12 months. They said it might happen sooner if there is no government intervention.
Safety concerns and Google’s efforts
Last month, AI researchers warned that people could lose control of the technology without sufficient guardrails. RSI was central to those warnings. Brin has become more involved in Google’s Gemini work and has used his influence as cofounder to shift resources toward projects he considers effective, Business Insider previously reported. That has included pushing for more focus on RSI.
Google teams see faster cadence
At a roundtable, Brin said he was excited by the pace of work at Google. He said that by the time viewers watched a video about Argon’s release, teams would be a couple of models ahead internally.
Key points
- Recursive self-improvement means AI systems help create more capable AI.
- Brin cited Google’s Flash releases before Gemini 4 as evidence of progress.
- Most researchers interviewed for a Gladstone AI report expect RSI within 3 to 12 months.
- Researchers have warned that insufficient guardrails could put human control at risk.