Anthropic’s AI R&D Measurements Point to a Broader Transparency Model

Anthropic’s public work on AI R&D automation points to a more demanding way to track advanced AI systems: measure not only what a model can do in an evaluation, but also how it is used in consequentia