A Neuropsychologically Grounded Evaluation of LLM Cognitive Abilities

TL;DR

This paper introduces NeuroCognition, a benchmark based on neuropsychological tests to evaluate LLMs' core cognitive abilities, revealing strengths and weaknesses beyond traditional task completion metrics.

Contribution

It presents a novel neuropsychologically grounded benchmark for assessing fundamental cognitive skills in LLMs, highlighting areas for improvement.

Findings

Models perform well on text but less so on images and complex tasks.

Simple strategies can improve model performance on certain tasks.

NeuroCognition correlates with general benchmarks but measures distinct abilities.

Abstract

Large language models (LLMs) exhibit a unified "general factor" of capability across 10 benchmarks, a finding confirmed by our factor analysis of 156 models, yet they still struggle with simple, trivial tasks for humans. This is because current benchmarks focus on task completion, failing to probe the foundational cognitive abilities that highlight these behaviors. We address this by introducing the NeuroCognition benchmark, grounded in three adapted neuropsychological tests: Raven's Progressive Matrices (abstract relational reasoning), Spatial Working Memory (maintenance and systematic search), and the Wisconsin Card Sorting Test (cognitive flexibility). Our evaluation reveals that while models perform strongly on text, their performance degrades for images and with increased complexity. Furthermore, we observe that complex reasoning is not universally beneficial, whereas simple, human-like strategies yield partial gains. We also find that NeuroCognition correlates positively with standard general-capability benchmarks, while still measuring distinct cognitive abilities beyond them. Overall, NeuroCognition emphasizes where current LLMs align with human-like intelligence and where they lack core adaptive cognition, showing the potential to serve as a verifiable, scalable source for improving LLMs.