EMBER Targets Token Embeddings: New Method Blocks LLM Knowledge Relearning
megamor2 · x · 2026-10-25
A new preprint (EMNLP 2026, Hall H Session B) introduces EMBER, a method for concept erasure in LLMs.
- Removing knowledge from LLMs is hard; preventing relearning is even harder
- The authors identify a blind spot in existing erasure methods: token embeddings
- EMBER addresses this gap, offering a more robust erasure approach
Paper by ClaraSuslik with megamor2 and OrShafran.
More from Research
- Safety researcher: verification should go all the way down to Maxwell's equations, not stop at assembly — JacquesThibs · 2026-10-25
- Local 35B authored 800 math problems under cloud-agent supervision — authoring erred 5x more than cold-solving — Individual_Page9676 · 2026-10-25
- SA-LapNet fixes spin exactly in neural-network quantum Monte Carlo, breaking the purity-energy trade-off — bravo_abad · 2026-10-25
- XYBench: Strongest LLMs Answer Literally and Detect Under 63% of User Misconceptions — MaartenSap · 2026-10-25
- Multi-domain training mix boosts supervised hallucination detectors 7.25% on out-of-distribution QA — MaartenSap · 2026-10-25
- EVALUESTEER: best reward models fail to align with user values over 25% of the time — MaartenSap · 2026-10-25