Search papers, labs, and topics across Lattice.
This study introduces a benchmark for evaluating large language models (LLMs) on logical inference involving probability operators, utilizing 14,320 procedurally-generated prompts across fifteen inference templates. The evaluation of 29 models reveals a concerning trend: most exhibit significant answer biases, favoring simplistic Yes or No responses regardless of the logical complexity of the queries. Only 9 out of the 29 models performed above random chance, highlighting a competence floor that underscores the limitations of current LLMs in handling nuanced logical reasoning tasks.
Most large language models struggle with logical inference over probability operators, showing pervasive biases that undermine their reliability in critical applications.
Both expressions of uncertainty and inferences are ubiquitous in natural language, and valid inferences over natural-language expressions of uncertainty are necessary for not only everyday conversations but also for high-stakes domains such as medicine and law. While large language models are increasingly evaluated on logical reasoning tasks, disentangling principled, symbolic reasoning from clever surface-level pattern matching is fraught with difficulty. We introduce a benchmark for reasoning over probability operators--inference over sentences with gradable epistemic modals (e.g., probably, might, must) containing 14,320 procedurally-generated English prompts across fifteen inference templates, systematically varying question form, negation strategy, and surface content. Evaluating 29 models, we find that most show answer biases independent of the logical form, a systematic preference for Yes or No. We summarize this with a competence floor: the worse of a model's accuracy on Yes-correct and No-correct items. Only 9 of 29 models exceed random chance. We also test variations in question form, verb phrases/activity, and both the gender and origin of names used in the prompts, finding biases across every axis.
