Anthropic paper: an internal Claude model found the core algorithm with no human input

@__alpoge__ quoted a note from an Anthropic paper: "The core algorithm was found on October 16 by an internal Claude research model with 14M output tokens and no human input."

The note adds that Anthropic researchers worked with Claude to improve the presentation, which produced the paper.

Why it mattersThe claim is that the key research step was done by the model on its own, with people only helping write it up.