【#Tech24H】Not long ago, a set of experimental data jointly produced by Tsinghua University and the Beijing Zhongguancun Academy went viral in the AI industry. During a oneweek experiment, AutoSOTA, an AI research agent, discovered 105 models that surpassed the current state of the art (SOTA) achieved by the latest human research, challenging conventional understanding of AI performance iteration efficiency. Unlike other AI for Science models, AutoSOTA essentially allows AI to “perform surgery” on AI, using AI to revolutionize the development and research of AI models. With the emergence of agents like OpenClaw, AI for Science is gradually evolving into Agent for Science. Artificial intelligence is transitioning from a mere “assistant” in the research process to a “research partner” and “AI scientist” capable of participating in reasoning, decisionmaking, and experimental closedloop tasks.  [ By Zhang Liyan | Tang Ruohan ]