analiza aiqbench
This study provides a matched evaluation comparing behavioral alignment methods like DPO against representation engineering techniques for controlling and monitoring large language models. The findings indicate that while behavioral safeguards offer stronger overall control, representation probes serve as a cost-effective alternative for early detection in low-data scenarios and can complement standard monitors to recover safety degraded by subsequent fine-tuning.
When Do Model Internals Help? Exploring the Role of Representation Engineering in LLM Safety
arXiv:2609.34771v1 Announce Type: cross Abstract: Reliable AI safeguards require both control mechanisms that reduce unsafe behavior and monitoring mechanisms that detect safety risks during model interactions. Established behavioral safeguards include alignment methods that optimize model outputs and text monitors that assess interaction text. Representation engineering instead reads or modifies internal model states, but the relative strengths of these approaches remain unclear because they are often evaluated under different settings. We present a matched evaluation across two tracks. For safety control, we compare DPO, a behavioral alignment method, with three representation steering methods across robustness, practicality, and granularity. DPO provides the strongest overall control and generally improves with increasing training data, although its safety can degrade after subsequent benign fine-tuning. Representation steering remains competitive primarily in low-data settings, particularly with high-quality contrastive data. For safety monitoring, we compare representation probes with fine-tuned and open-weight text monitors across full-response detection, early detection, and computational cost. Specialized text monitors achieve the strongest overall detection accuracy, while representation probes remain competitive at substantially lower marginal cost. Finally, monitor-guided interventions recover much of the safety lost by DPO after benign fine-tuning, with little additional over-refusal. Overall, representation engineering does not generally replace behavioral safeguards, but offers practical advantages under specific conditions and can provide complementary safety benefits.
Title:When Do Model Internals Help? Exploring the Role of Representation Engineering in LLM Safety
Abstract:Reliable AI safeguards require both control mechanisms that reduce unsafe behavior and monitoring mechanisms that detect safety risks during model interactions. Established behavioral safeguards include alignment methods that optimize model outputs and text monitors that assess interaction text. Representation engineering instead reads or modifies internal model states, but the relative strengths of these approaches remain unclear because they are often evaluated under different settings. We present a matched evaluation across two tracks. For safety control, we compare DPO, a behavioral alignment method, with three representation steering methods across robustness, practicality, and granularity. DPO provides the strongest overall control and generally improves with increasing training data, although its safety can degrade after subsequent benign fine-tuning. Representation steering remains competitive primarily in low-data settings, particularly with high-quality contrastive data. For safety monitoring, we compare representation probes with fine-tuned and open-weight text monitors across full-response detection, early detection, and computational cost. Specialized text monitors achieve the strongest overall detection accuracy, while representation probes remain competitive at substantially lower marginal cost. Finally, monitor-guided interventions recover much of the safety lost by DPO after benign fine-tuning, with little additional over-refusal. Overall, representation engineering does not generally replace behavioral safeguards, but offers practical advantages under specific conditions and can provide complementary safety benefits.
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2609.34771 [cs.AI]
(or arXiv:2609.34771v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.34771 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 · 322 cuvinte
Subiecte conexe
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · 18 sept. 2026
Amazon is investing in the Lean Focused Research Organization
Amazon Science · 26 iul. 2026
Microsoft’s AI chief says superintelligence is near, but won’t take your job
The Verge · 8 iun. 2026
[AINews] The Future of Latent Space
Latent Space · 25 sept. 2026