This research continues many of the directions our team worked on prior to Anthropic, including: GPT-3, Circuit-Based Interpretability, Multimodal Neurons, Scaling Laws, AI & Compute, Concrete Problems in AI Safety, and Learning from Human Preferences. Be the technical voice on how Claude performs on science and engineering work, explaining results, limits, and tradeoffs clearly to researchers, engineering leads, and non-technical stakeholders alike.