Nathan Lambert shares a farewell letter to the Allen Institute for AI (Ai2), reflecting on his two-year journey from a relative unknown to a key contributor on the OLMo and Tülu model families. He traces his career path from electrical engineering through a Berkeley PhD, HuggingFace, and finally Ai2, where he helped coin the term RLVR and led major open post-training efforts. The post argues for the continued importance of public scientists and open research ecosystems as AI increasingly moves behind closed doors at frontier labs. Lambert expresses concern that the migration of researchers to closed industry labs is eroding a social contract around neutral, accessible science, and warns that open post-training recipes are falling further behind. He plans to continue advocating for open models, coordinating the open ecosystem, and exploring medium-sized task-specific models as complements to frontier systems.
