Google’s Pichai says AI self-improvement still in early stages

Google CEO Sundar Pichai says that AI systems that improve themselves, a process called recursive self-improvement (RSI), are still at an early stage. This means current AI systems cannot yet build and improve their own models completely on their own. Google's DeepMind team is working on this by using AI systems that repeat tasks and AI agents to help human researchers develop AI models. However, despite recent progress, Google has not seen major speed increases across the entire AI research and development process when measured using different tests. Pichai's comments suggest that the industry is still far from creating AI systems that can improve themselves fully without human help.

• Sundar Pichai's comments suggest that AI systems that improve themselves are still at an early stage.
• Current market prices for Anthropic's AI model show that people have less confidence that it will be the best model by October 2026.
• Google is still using partly automated methods to develop AI models, rather than systems that work completely on their own.

Watch for more announcements from Google and its competitors about progress in AI self-improvement. Work on Google's DeepMind programs, especially the Gemini 4 Argon model, could change how people view the market. New performance test results or model releases from Anthropic, Meta, and other major companies could also change how confident people are about which AI model will be the best by the end of October 2026.