CATT benchmark: negative log-odds loss boosts LM data attribution across methods
davidbau · x · 2026-10-31
A new arXiv paper by Steven Yu, Sheng-Yu Wang, Kimia Hamidieh, Antonio Torralba, and David Bau maps the design space of LLM data attribution, focusing on evaluation.
- Introduces the Controlled Attribution Benchmark (CATT) for fact tracing: LLMs are fine-tuned on entirely fictitious knowledge, giving unambiguous ground truth while varying one dataset property at a time (document redundancy, lexical distractors).
- Redundancy and distractors consistently degrade attribution performance.
- Key insight: causal evidence in a training document is highly localized, but standard methods dilute it into noise by averaging scores across the whole document.
- Simple fix: replace cross-entropy gradients with negative log-odds gradients to surface critical tokens, then aggregate only top token scores — this improves every attribution method tested.
Paper, demo, code, and models are all publicly available.
Related event: Negative Log Odds Loss Boosts LM Training Data Attribution(2 posts)→
More from Research
- Real human brain cells are being used for AI, and Reddit is buzzing — EmotionalGuess9229 · 2026-10-31
- LLM judges harbor systematic errors that silently corrupt model rankings and training pipelines — burkov · 2026-10-31
- New paper quantifies ineffable concepts: 65 of 1,270 can never be taught to LLMs via language — sanmikoyejo · 2026-10-31
- Unsupervised Method Measures Whether LLM Messages Are Received — sanmikoyejo · 2026-10-31
- Neural net flies a drone in Acro mode from streamed game footage on a tiny board — yacineMTB · 2026-10-31
- llmlip-rs: Rust MLIP inference engine runs 6000-atom sims 17.7x faster than PyTorch MACE — CatAstro_Piyush · 2026-10-31