AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
Today, we start a new series about one of the hottest topics in AI: recursive self improvement. Throughout the next few weeks, we will deep dive into the top research For about sixty years, arguing about recursive self-improvement meant arguing in the abstract. There was no system to point at. You cited I.J. Good, somebody cited Schmidhuber, everyone disagreed about definitions for two hours, and then you went home. It was a very pleasant way to spend an afternoon and it produced nothing. That era ended sometime in the last twelve months. In June 2026 Anthropic published an essay stating that as of May, Claude authored more than 80 percent of the code merged into its production codebase. OpenAI disclosed that GPT-5.3-Codex helped debug its own training process and manage parts of its own deployment. DeepMind’s AlphaEvolve has been turning up algorithmic improvements that land in the infrastructure other models train on. Three labs, three different flavors of disclosure, one direction. So the question is no longer whether AI helps build AI. It clearly does. The question is which parts of the job it took, how well it does them, and what checks the work. That last one turns out to be the whole ballgame, and it is what this series is about. The factory, not the robot Read more