O1 Replication Journey

The first open o1-replication program (2K★), from Pengfei Liu's GAIR lab in its pre-SII era — a "strategic progress report" documenting journey learning, search, and reward modeling in the open weeks after o1 shipped. Part 2's finding that simple distillation from o1's API plus SFT beats elaborate pipelines — the distillation "bitter lesson" — became one of the most-cited early observations of the reasoning-model era. (GAIR's later reasoning line ships under SII.)