analiza aiqbench
This empirical study investigates how preference tuning generalizes when language models encounter domain shifts, comparing five alignment objectives and adaptation strategies like pseudo-labeling. The research identifies a critical trade-off where methods that significantly reduce performance degradation often induce mode collapse, thereby limiting output diversity. These findings offer essential insights for practitioners aiming to balance model helpfulness and safety across different application domains without sacrificing generative variety.
An Empirical Study on Preference Tuning Generalization and Diversity Under Domain Shift
arXiv:2601.05882v2 Announce Type: replace-cross Abstract: Preference tuning aligns base language models to human judgments of quality, helpfulness, or safety by optimizing over explicit preference signals rather than likelihood alone. Prior work has shown that preference tuning degrades performance and reduces helpfulness outside the training domain. However, the extent to which adaptation strategies mitigate this domain shift remains unexplored. We address this challenge by conducting a comprehensive and systematic study of alignment generalization under domain shift. We compare five popular alignment objectives and various adaptation strategies from source to target, including target-domain supervised fine-tuning and pseudo-labeling, across summarization, question-answering helpfulness, and safety alignment tasks. Our findings reveal systematic differences in generalization across alignment objectives under domain shift. We show that adaptation strategies based on pseudo-labeling substantially reduce domain-shift degradation but induce mode collapse, revealing a generalization-diversity trade-off.
Abstract:Preference tuning aligns base language models to human judgments of quality, helpfulness, or safety by optimizing over explicit preference signals rather than likelihood alone. Prior work has shown that preference tuning degrades performance and reduces helpfulness outside the training domain. However, the extent to which adaptation strategies mitigate this domain shift remains unexplored. We address this challenge by conducting a comprehensive and systematic study of alignment generalization under domain shift. We compare five popular alignment objectives and various adaptation strategies from source to target, including target-domain supervised fine-tuning and pseudo-labeling, across summarization, question-answering helpfulness, and safety alignment tasks. Our findings reveal systematic differences in generalization across alignment objectives under domain shift. We show that adaptation strategies based on pseudo-labeling substantially reduce domain-shift degradation but induce mode collapse, revealing a generalization-diversity trade-off.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2601.05882 [cs.CL]
(or arXiv:2601.05882v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2601.05882 Focus to learn more arXiv-issued DOI via DataCite
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 · 230 cuvinte
Subiecte conexe
You can graft SDF changes from base models onto post-trained models
LessWrong · 6 oct. 2026
Frontier post-training recipe review with Finbarr Timbers
Interconnects · 16 iun. 2026
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026
The basic argument for making AI better at moral philosophy
LessWrong · 14 oct. 2026