Research4h ago
OriginBlame Enables Precise AI Training Data Deletion
The OriginBlame system reduces dataset over-deletion from 101x to 1.3x by tracking author identity at the token level. This research solves the practical gap in unlearning algorithms by generating precise forget sets via deterministic queries. It adds up to 19% throughput overhead in Datatrove. Practitioners can now honor data revocation requests without purging entire files.