CapTrack accepted to NeurIPS 2026, extends LLM forgetting tests to capabilities
The announcement says CapTrack evaluates what large language models forget beyond general knowledge and finds that forgetting varies widely across capabilities.
TLDR
The CapTrack announcement says the project has been accepted to NeurIPS 2026. It says CapTrack extends LLM forgetting evaluation from general knowledge to capabilities, finds that forgetting varies widely across them, and urges anyone changing model weights to check what changed.
Combined views
314
2 Sources, first seen 9h ago
likes
