3 papers
cs.LG2026
Selective Safety Steering via Value-Filtered Decoding
Bat-Sheva Einbinder, Hen Davidov, Yee Whye Teh +2
While large language models (LLMs) are trained to align with human values, their generations may still violate safety constraints. A growing line of work addresses this problem by…
cs.LG2026
Knowing When to Quit: A Principled Framework for Dynamic Abstention in LLM Reasoning
Hen Davidov, Nachshon Cohen, Oren Kalinsky +4
LLMs utilizing chain-of-thought reasoning often waste substantial compute by producing long, incorrect responses. Abstention can mitigate this by withholding outputs unlikely to be…
cs.LG2026
Calibrated Predictive Lower Bounds on Time-to-Unsafe-Sampling in LLMs
Hen Davidov, Shai Feldman, Gilad Freidkin +1
We introduce time-to-unsafe-sampling, a novel safety measure for generative models, defined as the number of generations required by a large language model (LLM) to trigger an unsa…