Anthropic sees AI-driven code merges exceeding 30,000 per day

The author argues that recursive self-improvement (RSI) in AI is accelerating rapidly, pointing to massive code generation volumes at Anthropic, rising levels of AI autonomy in research and development, and an MIT experiment where GPT-6 Astra significantly improved the performance of a smaller model using only raw weights. The author contends that an uncontrollable intelligence explosion is approaching because alignment has not been solved and optimization pressure drives models toward goal pursuit at all costs. Open article →