Ross Nordby
Bio
Ross Nordby is technical staff at Anthropic working on AI safety. Before joining Anthropic, he was an independent AI alignment researcher funded by the Long-Term Future Fund, during which he worked on corrigibility frameworks, interpretability, and reinforcement learning environments. His background is in real-time graphics and physics simulation for video games; he created bepuphysics2, a widely-used open-source C# 3D physics engine, and runs Bepu Entertainment LLC. His published alignment work includes the paper "Soft Prompts for Evaluation: Measuring Conditional Distance of Capabilities" (arXiv, May 2025), exploring optimized input embeddings as a metric for latent capability discovery and automated red-teaming of language models, as well as LessWrong posts on using predictors in corrigible systems and AGI timelines. He received an honorable mention in the AI Alignment Awards Research Contest in the corrigibility category. He is based in Chicago, Illinois, and posts on LessWrong under the handle "porby".
Links
- Personal Website
- https://www.bepuentertainment.com/
- Twitter / X
- LessWrong
- porby
Grants
from Long-Term Future Fund
from Long-Term Future Fund
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.
Details
- Last Updated
- Mar 23, 2026, 12:45 AM UTC
- Created
- Mar 20, 2026, 2:57 AM UTC