Emergent misalignment is predictable OOD; later-layer fine-tuning reduces transfer
AdtRaghunathan · x · 2026-11-11
New research shows emergent misalignment is predictable out-of-distribution: OOD and in-domain rates track on-the-line across models and hyperparameters. That's a problem, though—limiting OOD generalization also limits in-domain learning. The team, led by Olina Mukherjee, shows fine-tuning only later layers retains learning with less misalignment transfer.
More from Research
- AHA 2026: AI reads smartwatch ECGs for heart failure and speeds echo interpretation — AmiBhattMD · 2026-11-11
- SAG2 builds RAG index 7000x faster with no embeddings or GPU, beats HippoRAG 2 — udmrzn · 2026-11-11
- Artist with disability releases his artwork as a dataset for anyone to remix — plzno1 · 2026-11-11
- The Memory Trap: Benign Chats Can Make AI Agents Skip Confirmations and Wire $1.2M Unverified — AndrewLampinen · 2026-11-11
- TaskNet: AI-Led Worker Interviews Add 79 Tasks Missing from O*NET, Raise AI Exposure for Lawyers and Devs — stanfordnlp · 2026-11-11
- Stanford CS329Z releases assignment on building terminal-bench-style agent evals with Harbor — Diyi_Yang · 2026-11-11