analiza aiqbench
This research establishes a new task for predicting how fine-tuning Large Language Models on specific values affects their behavior across unseen contexts, revealing that activation-based representations significantly outperform textual descriptions. By demonstrating a correlation between value similarity and model robustness, the study provides empirical methods for designing more reliable alignment targets without relying solely on narrow behavioral benchmarks. The work advances the empirical understanding of value generalization dynamics in LLMs, offering a pathway toward a shared, model-independent value space.
Predicting Alignment Generalization with Value Representations
arXiv:2610.12410v1 Announce Type: cross Abstract: LLM developers post-train their models to exhibit prosocial values and behavioral traits, which are enumerated in an alignment target. However, while recent post-training developments have yielded models that score highly on alignment evaluations, training models on sets of narrow behaviors still influences their behavior across unseen contexts and environments in unexpected ways. In this paper, we establish the task of alignment generalization prediction, i.e., predicting how fine-tuning a model to follow a given value changes its behavior across a wide range of held-out values. We conduct a large-scale analysis of alignment generalization effects across 66 values found in modern alignment targets, and benchmark representational techniques on the alignment generalization prediction task. We find that representations based on model activations when applying values in context significantly outperform methods based on textual descriptions of the values. Specifically, the best activations-based methods achieve correlations of 0.45 with our generalization matrix, compared with 0.05 from description-based baselines. We then show the applicability of representations that predict alignment generalization toward downstream tasks by using them to measure how similar the values in a multi-value alignment target are, which we find is significantly correlated with model robustness. Finally, we show initial evidence towards a shared, model-independent value space, which we use to develop the first taxonomy of LLM values grounded in empirical generalization dynamics. Our work demonstrates the importance of studying value generalization in LLMs and its application toward the more empirical design and training of model behavior.
Abstract:LLM developers post-train their models to exhibit prosocial values and behavioral traits, which are enumerated in an alignment target. However, while recent post-training developments have yielded models that score highly on alignment evaluations, training models on sets of narrow behaviors still influences their behavior across unseen contexts and environments in unexpected ways. In this paper, we establish the task of alignment generalization prediction, i.e., predicting how fine-tuning a model to follow a given value changes its behavior across a wide range of held-out values. We conduct a large-scale analysis of alignment generalization effects across 66 values found in modern alignment targets, and benchmark representational techniques on the alignment generalization prediction task. We find that representations based on model activations when applying values in context significantly outperform methods based on textual descriptions of the values. Specifically, the best activations-based methods achieve correlations of 0.45 with our generalization matrix, compared with 0.05 from description-based baselines. We then show the applicability of representations that predict alignment generalization toward downstream tasks by using them to measure how similar the values in a multi-value alignment target are, which we find is significantly correlated with model robustness. Finally, we show initial evidence towards a shared, model-independent value space, which we use to develop the first taxonomy of LLM values grounded in empirical generalization dynamics. Our work demonstrates the importance of studying value generalization in LLMs and its application toward the more empirical design and training of model behavior.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2610.12410 [cs.CL]
(or arXiv:2610.12410v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.12410 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 · 343 cuvinte
Subiecte conexe
Models Can't Remember Their Training. Neither Can You.
LessWrong · 19 iul. 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · 18 sept. 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026
MSE loss does not generate superposition
LessWrong · 27 iul. 2026