You have delivered technology programs inside banks, insurers, or asset managers, you understand how model risk, security, and compliance review shape what ships, and you can speak credibly with a CIO, a chief risk officer, and a head of engineering. 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.