RepSelect: Robust LLM Unlearning via Representation Selectivity

Name: RepSelect: Robust LLM Unlearning via Representation Selectivity
Uploaded: 2026-08-31T05:25:24.645064+00:00
Duration: 460 s
Description: This episode discusses 'RepSelect,' a method for robust LLM unlearning. The hosts explore how the technique allows models to forget specific information by targeting its internal representation, rather than random weight deletion or full retraining. This approach significantly enhances AI accountability and trust.

This episode discusses 'RepSelect,' a method for robust LLM unlearning. The hosts explore how the technique allows models to forget specific information by targeting its internal representation, rather than random weight deletion or full retraining. This approach significantly enhances AI accountability and trust.

Open the full episode