Holy crap, if this rumor turns out to be true, the pace of AI development might enter a phase humanity has never experienced before. Recent leaks in the AI community suggest that Google has either achieved or is very close to RSI, OpenAI’s mysterious model Bel might have used RSI to contribute to recent breakthroughs in mathematical research, and Anthropic is training its next-gen model codenamed Fable 6. RSI, short for Recursive Self-Improvement, basically means AI starts helping itself evolve: The first generation of AI helps engineers improve training algorithms, write code, and optimize models. The second generation of AI, with stronger R&D capabilities, then helps create an even more powerful third generation. This cycle continues, and if each round of improvement accelerates the next, AI could go from upgrading every few months to entering an increasingly rapid self-iteration loop. This is the so-called "intelligence explosion" people have been speculating about for decades. Google’s RRSI research, published in September, already demonstrated how AI can autonomously optimize its own agent systems, prompts, tools, and workflows. On October 6, OpenAI also revealed 722 mathematical manuscripts generated by its internal cutting-edge model, covering 372 research results. Once a complete RSI feedback loop is established, the bottleneck for AI R&D might shift from the intelligence and working hours of human engineers to computing power, electricity, experimental validation, and physical resources. The stronger the model, the higher the R&D efficiency; the higher the R&D efficiency, the faster new models emerge. The competitive logic of the entire AI industry could change, with computing power and energy becoming the core resources limiting the speed of intelligence evolution. If AI starts accelerating its own improvement, the pace of adjustments for jobs, companies, and even entire industries might not be able to keep up with technological iterations.