Brin says Gemini models now help build their successors

Google cofounder Sergey Brin says the company's AI models are now helping build their own successors. He described it as the early stage of recursive self-improvement (RSI), a long-debated point at which AI starts speeding up its own development.

Brin made the remarks in a roundtable on the making of Gemini 4 Argon, Google's new flagship model. Google shared the video on Wednesday, October 14, Business Insider reported. He said AI development is speeding up inside Google Alphabet Inc. ↗ 0.37%. He pointed to the fast run of Flash releases before Gemini 4 as evidence, saying the company was "starting to get onto that RSI treadmill."businessinsider.com

Models helping build models

"And now our researchers and the model work hand in hand to make the next version of the model," Brin said, according to Business Insider. He added that by the time viewers saw the video, Google would already be using models "a couple" of versions beyond Argon internally, and that he expects the pace "will only continue to accelerate."businessinsider.com

Google's safety report for Argon says parts of the AI research and development cycle are already automated. "AI agents now operate as high-leverage assistants that author a majority of production code and bug fixes, though code changes remain subject to human review before submission," the report states, per Business Insider. Vihan Jain, an RSI lead for Gemini at Google DeepMind, wrote on X that "the RSI flywheel has reached a new gear" now that post-training teams are asking for Argon usage quota rather than raw chips.businessinsider.com

A delayed flagship

Argon became available to Google AI Pro and Ultra subscribers in the Gemini app on October 13, Google said. It is Google's first top-tier model in about seven months, following Gemini 3.1 Pro in February. Before the wide release, it went first to vetted experts through the company's "Fairwind" security program on September 30, SBS reported. In the meantime, Google had held back higher-end models over concerns about their coding ability and released lightweight Flash models about every three weeks, according to SBS.blog.google+1

Brin has become more involved in the Gemini effort. He has used his influence as cofounder to steer resources toward RSI work, Business Insider has reported. Business Insider also reported that Google staff have been testing a stronger internal version of Argon called Carbon.businessinsider.com

Industry race and safety concerns

Anthropic and OpenAI have also said AI may be close to recursive self-improvement. A report released this week by Gladstone AI found that most of the researchers it interviewed expect RSI to begin within three to 12 months, and possibly sooner without government action, Business Insider reported.businessinsider.com

That outlook concerns some experts. Last month, AI researchers warned that humans could lose control of the technology if strong enough guardrails are not in place before self-improving systems arrive.