The Model Knows Your Project, Not You.

Per a post on LessWrong, language models recognize software projects far more readily than they recognize their creators. An experiment across thirty-seven models tested this pattern. A popular spaced-repetition tool called FSRS was identified by thirty-one of them, while the creator's name was recognized by only fourteen. When researchers added context — "the creator of FSRS" — recognition jumped to twenty-eight models. The finding reveals how LLM training data contains extensive information about tools and projects, but often lacks detailed coverage of individual people's identities and contributions.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton