Max Schwarzer on AI Leadership and Research
Max Schwarzer, who led OpenAI's post-training team for models like the GPT-5 series, is moving to Anthropic to return to an individual contributor research role. His reflection highlights the challenge of balancing technical leadership and deep research, crediting collaborative work for breakthroughs in reasoning paradigms.
At OpenAI, Schwarzer's post-training team was crucial for taking powerful pre-trained models and preparing them for real-world deployment to millions of users. This final step of research and development ensures models are safe, efficient, and reliable for products like ChatGPT and the API. His team was responsible for delivering not just the GPT-5 series but also earlier reasoning models like o1 and o3. Post-training work involves a combination of reinforcement learning and product-focused research to refine model capabilities and performance before they are released. The move to Anthropic signals a return to hands-on reinforcement learning (RL) research, a field where Schwarzer has a deep background, including a PhD focused on sample-efficient RL. Many former OpenAI colleagues he trusts and respects have also moved to Anthropic over the last couple of years. This transition highlights a common career crossroads for senior engineers: the "technical track" versus the "management track." The leadership path involves managing teams and strategy, which can reduce time spent on hands-on coding and research—a trade-off many technical experts find challenging. Anthropic, founded by former OpenAI members, has a distinct focus on AI safety, interpretability, and developing reliable, steerable AI systems. Their research teams are organized around understanding model internals, ensuring alignment with human values, and studying the societal impacts of AI. Schwarzer joins Anthropic as a Member of Technical Staff, the same title he held when he first joined OpenAI before being promoted to VP of Research. This role allows deep focus on individual technical contributions rather than people management.
