Measure, Don't Optimize: Forecasting Recovery in LLM Unlearning

If you rely on unlearning for safety or compliance, clean outputs aren't enough — residual internal accessibility predicts how quickly removed knowledge comes back under further fine-tuning.

• fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.

• articleMemorization Diagnostics for Code LLMs Should be Scale-AwarePrateek Kumar Rajput, Abdoul Aziz Bonkoungou, Alberick Euraste Djir\'e, Xunzhu Tang, Yewei Song, Iyiola Emmanuel Olatunji, El Hacen Diallo, Jacques Klein, Tegawend\'e F. Bissyand\'e
• articleA Removal Based Approach to Improve LLM Faithfulness at Test-TimeQinglan Luo, S M A Nahian, John Guttag, S. Mazdak Abulnaga, Katie Matton
• articleFrom Detection to Refusal: Safer LLMs via Circuit-Guided Weight ScalingKuan-Lin Chu, Chung-En Sun, Tsui-Wei Weng

Checking sign-in…

Loading comments…