Anthropic reveals"AI Trains AI"New method: automated alignment researcher lower cost and faster
The automated system first reviews existing literature, proposes training methods, then trains the model for 30 minutes according to the plan, and gradually improves test scores through multiple rounds of attempts. Effective plans are kept, ineffective plans are eliminated, enabling the entire research process to rapidly scale up.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open the original article on ITHome