LLM Digest

Story

arxiv_llm_reliability · Oct 26, 2026 · paper

Source brief

CANS: Capability-Preserving Neuron Steering for Large Language Models

arxiv.orgOct 26, 2026
original source linked

In brief

Activation steering offers training-free and dynamic control of target behaviors in large language models, but existing methods can impair general capabilities and require domain-specific intervention designs. We intr...

Feed lens

eval

Read the original at arxiv.org →Open in live feed Read that day’s brief

CANS: Capability-Preserving Neuron Steering for Large Language Models

Earlier in this thread 4 items

Spectral Smoothing and Spiking Perturbations Reconfigure Capability Profiles and Selective Task Performance in Large Language Models

Self-Assessment Reinforcement Learning for Search-State Awareness in Retrieval-Augmented Large Language Models

Hallucinations of Large Language Models in Cross-Language Code Translation: Phenomena and Mitigation

External Observers May See More Clearly: Cross-Model Span-Level Hallucination Detection in Large Language Models via Hidden State Probing