analiza aiqbench
This research introduces a novel formal framework to measure AI "propensities"—the tendency of models to exhibit specific behaviors—using a bilogistic formulation that moves beyond traditional capability metrics. By defining an "ideal band" where performance is optimal, the study demonstrates that combining propensity and capability measurements yields stronger predictive power for model behavior on unseen tasks than either metric alone. This shift is critical for advancing AI safety and evaluation, as it allows researchers to quantify and manage both excessive and deficient behavioral tendencies in large language models.
Capabilities Ain't All You Need: Measuring Propensities in AI
arXiv:2602.18182v5 Announce Type: replace-cross Abstract: AI evaluation has primarily focused on measuring capabilities, with formal approaches inspired from Item Response Theory (IRT) being increasingly applied. Yet propensities - the tendencies of models to exhibit particular behaviours - play a central role in determining both performance and safety outcomes. However, traditional IRT describes a model's success on a task as a monotonic function of model capabilities and task demands, an approach unsuited to propensities, where both excess and deficiency can be problematic. Here, we introduce the first formal framework for measuring AI propensities by using a bilogistic formulation for model success, which attributes high success probability when the model's propensity is within an "ideal band". Further, we estimate the limits of the ideal band using LLMs equipped with newly developed task-agnostic rubrics. Applying our framework to six families of LLM models whose propensities are incited in either direction, we find that we can measure how much the propensity is shifted and what effect this has on the tasks. Critically, propensities estimated using one benchmark successfully predict behaviour on held-out tasks. Moreover, we obtain stronger predictive power when combining propensities and capabilities than either separately. More broadly, our framework showcases how rigorous propensity measurements can be conducted and how it yields gains over solely using capability evaluations to predict AI behaviour.
Authors:Daniel Romero-Alvarado, Fernando Martínez-Plumed, Lorenzo Pacchiardi, Hugo Save, Siddhesh Milind Pawar, Behzad Mehrbakhsh, Pablo Antonio Moreno Casares, Ben Slater, Paolo Bova, Peter Romero, Zachary R. Tidler, Jonathan Prunty, Luning Sun, Jose Hernandez-Orallo
Abstract:AI evaluation has primarily focused on measuring capabilities, with formal approaches inspired from Item Response Theory (IRT) being increasingly applied. Yet propensities - the tendencies of models to exhibit particular behaviours - play a central role in determining both performance and safety outcomes. However, traditional IRT describes a model's success on a task as a monotonic function of model capabilities and task demands, an approach unsuited to propensities, where both excess and deficiency can be problematic. Here, we introduce the first formal framework for measuring AI propensities by using a bilogistic formulation for model success, which attributes high success probability when the model's propensity is within an "ideal band". Further, we estimate the limits of the ideal band using LLMs equipped with newly developed task-agnostic rubrics. Applying our framework to six families of LLM models whose propensities are incited in either direction, we find that we can measure how much the propensity is shifted and what effect this has on the tasks. Critically, propensities estimated using one benchmark successfully predict behaviour on held-out tasks. Moreover, we obtain stronger predictive power when combining propensities and capabilities than either separately. More broadly, our framework showcases how rigorous propensity measurements can be conducted and how it yields gains over solely using capability evaluations to predict AI behaviour.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2602.18182 [cs.LG]
(or arXiv:2602.18182v5 [cs.LG] for this version)
Text integral sub licența deschisă a publicației · 260 cuvinte
Subiecte conexe
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
AI Alignment Forum · 23 sept. 2026
Towards embedded evaluations for scheming propensities
LessWrong · 8 oct. 2026
Why do models task game?
AI Alignment Forum · 6 aug. 2026
AI swarms are starting to pose indirect takeover risk
AI Alignment Forum · 12 aug. 2026