Brin says Gemini 4 Argon shows AI entering recursive self-improvement
Google co-founder Sergey Brin said in a roundtable about the new Gemini 4 Argon model that the company's internal AI development is accelerating and that models have begun helping build their own successors, saying Google is "starting to get on the RSI treadmill," according to Business Insider reports cited by IT之家. He pointed to the Flash-series models shipped before Gemini 4 as evidence of the trend, while Vihan Jain, who leads Gemini recursive self-improvement work at Google DeepMind, wrote on X that you know the RSI flywheel has shifted up a gear when post-training requests move from raw chips to Argon usage quotas. Google's Argon safety report says AI agents already write much of the production code and fix bugs, but code changes still require human review before being submitted; a Gladstone AI report this week said most interviewed researchers from Anthropic, OpenAI and DeepMind expect RSI to begin within 3 to 12 months, possibly sooner without government intervention.
It signals that leading labs believe the automation of AI R&D may already be underway, though the claim rests on internal accounts and interviewee expectations rather than independently verifiable technical evidence.