Cooking with OpenAI’s Research Chief: AGI, o1, Evals, and Scaling Laws — Mark Chen

AI research at the frontier relies on a combination of rigorous scaling laws and creative problem-solving, where research taste is cultivated by replicating foundational papers and identifying novel cross-disciplinary breakthroughs. OpenAI maintains a stable high-level roadmap—focusing on pre-training, reasoning, and alignment—while empowering researchers to pursue high-risk bets through flexible compute allocation. As models saturate traditional benchmarks, the field faces an "evals crisis," requiring the development of new, adversarial evaluation methods to measure superhuman intelligence. The future of research is shifting toward "vibe research," where models handle complex execution and orchestration, allowing human researchers to focus on strategic direction and the development of intuition. Ultimately, the ability to solve long-horizon problems and integrate knowledge across domains remains the primary differentiator for both researchers and the models they build.