An American distillation regime

What's this about?

Garry Tan wants smaller US open-weight labs to learn from frontier-model outputs with permission, countering Anthropic's push against covert extraction using false identities or stolen credentials.

What exactly is model distillation?

A smaller “student” model learns from examples produced by a stronger “teacher.” It resembles an apprentice studying thousands of worked answers, though contractual permission and automated extraction are disputed.

What's the community saying?

1. Supporters say labs trained on the public commons cannot fairly block others from learning from their outputs.
2. Critics ask who will fund expensive frontier research if competitors can immediately copy much of its value.
3. Others separate legitimate paid teaching from identity fraud, stolen credentials and undisclosed black-box biases.