3 papers
cs.LG2026
Principled Under Pressure: Post-Training Decides Whether LLMs Act on Their Own Moral Judgment
Orion Reblitz-Richardson
An agent that can say an action is wrong and then take it anyway is a different failure from one that does not know better, and evaluations of stated values cannot see it. We ask h…
cs.LG2026
Refusal Reads Only a Slice of What the Model Knows: Harm-Keyed Routing and Its Exceptions Across Model Families
Orion Reblitz-Richardson
Alignment applied after pretraining is shallow in a measurable way: a single direction in a model's residual stream can be edited out, and the model stops refusing harmful requests…
cs.LG2026
Calibrating Interpretability Instruments Before Trusting Their Verdicts
Orion Reblitz-Richardson
Causal claims about large language model (LLM) internals rest on measurements. Those might include a projection, a cosine, an ablation delta, or an interchange patch among others.…