analiza aiqbench
This research introduces Adaptive Multilevel Twisted Sequential Monte Carlo to improve the estimation of extremely rare unsafe behaviors in large language models. By using a hierarchical approach to learn twist functions, the method overcomes sampling limitations that often plague standard frameworks when detecting low-probability risks. The technique offers a more reliable mechanism for evaluating and aligning deployed AI systems against hard-to-observe failure modes.
Adaptive Multilevel Twisted Sequential Monte Carlo for Rare Events Estimation in Language Models
arXiv:2608.21736v1 Announce Type: new Abstract: Rare unsafe behaviors in large language models can remain practically significant even when their probability is extremely small, particularly at deployment scales involving millions or billions of interactions. Twisted Sequential Monte Carlo (SMC) provides a principled framework for rare-event probability estimation by learning twist functions that guide generation toward a target event. However, the standard twist learning framework relies on positive samples from the rare-event target distribution, which may be nearly absent before an informative twist has been learned, resulting in unreliable rare-event estimation. We propose Adaptive Multilevel Twisted SMC, which learns the rare-event twist through a sequence of progressively rarer intermediate events. At each level, the learned twist provides more informative positive examples for learning the next twist, ultimately leading to a more accurate final twist for the target rare event. Experiments across diverse tasks and model scales show that the proposed method produces more accurate rare-event probability estimates. By enabling more reliable discovery of hard-to-observe unsafe behaviors, our method provides a practical tool for strengthening the evaluation and safety alignment of deployed language models.
Abstract:Rare unsafe behaviors in large language models can remain practically significant even when their probability is extremely small, particularly at deployment scales involving millions or billions of interactions. Twisted Sequential Monte Carlo (SMC) provides a principled framework for rare-event probability estimation by learning twist functions that guide generation toward a target event. However, the standard twist learning framework relies on positive samples from the rare-event target distribution, which may be nearly absent before an informative twist has been learned, resulting in unreliable rare-event estimation. We propose Adaptive Multilevel Twisted SMC, which learns the rare-event twist through a sequence of progressively rarer intermediate events. At each level, the learned twist provides more informative positive examples for learning the next twist, ultimately leading to a more accurate final twist for the target rare event. Experiments across diverse tasks and model scales show that the proposed method produces more accurate rare-event probability estimates. By enabling more reliable discovery of hard-to-observe unsafe behaviors, our method provides a practical tool for strengthening the evaluation and safety alignment of deployed language models.
Machine Learning (cs.LG)
arXiv:2608.21736 [cs.LG]
(or arXiv:2608.21736v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.21736 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 270 cuvinte
Subiecte conexe
LatticeBridge: Rare-Event Sequential Inference for Faithful Structured Sequence Synthesis
arXiv · 12 iun. 2026
Robust Importance Sampling for Rare Events via Constrained Gaussian Mixtures
arXiv · 7 oct. 2026
Steering Diffusion Models to Rare Events with Sequential Monte Carlo
arXiv · 7 oct. 2026
Rare Event Estimation via Iterative Unalignment
arXiv · 22 sept. 2026