AI technology is transitioning from an application stage primarily focused on chat and coding to a new phase of participating in scientific discovery. The new company Discovery Loop, founded by Jeff Dean, has been warmly sought after by the capital market, with its valuation rising rapidly. Meanwhile, cutting-edge technology institutions such as OpenAI and Anthropic have also made significant progress in this field. PhAI Labs released a technical report on Discovery Foundation Models (DFM), proposing a universal model system for open scientific discovery. The core team of this system boasts top-tier research backgrounds and model development experience. Currently, AI faces multiple challenges in participating in scientific research, including the lack of predefined task descriptions, difficulty in retaining experience, challenges in cross-domain adaptation, disconnection between prediction and decision-making, and the inability to reuse experience. To address these challenges, PhAI Labs has proposed a five-tier capability system aimed at building an infrastructure capable of carrying scientific research experience. The core reasoning system of DFM, through the Zetema reference architecture, forms a recursive discovery loop, enabling AI to participate in scientific research activities such as identifying research questions, constructing and revising scientific representations, validating hypotheses, and adjusting research directions based on evidence. In therapeutic peptide design experiments, the GALILEO system has been able to adjust research approaches and accumulate experience based on experimental feedback. Additionally, the team has launched the DFM Scientist Collaboration Program and plans to release relevant tool platforms to further explore how to enable AI to learn the ability of scientists to adjust research based on evidence, thereby participating more deeply in scientific discovery.