Triangulating LLM Progress through Benchmarks, Games, and Cognitive Tests

2025-02-20

AI Summary

Existing evaluation paradigms for large language models (LLMs) lack sensitivity to human-like cognitive capacities, limiting their ability to discriminate fine-grained differences in model quality. Method: We systematically compare three evaluation paradigms—large-scale QA benchmarks (MMLU, BBH), game-theoretic interactive tasks (Signalling Game, Taboo), and cognitively grounded tasks designed from human cognitive theories (e.g., working memory, theory of mind)—and introduce the first application of cognitive assessment frameworks to LLM evaluation, integrating correlational and causal analysis. Contribution/Results: Interactive games significantly outperform static benchmarks in discriminating executive functions and socio-emotional abilities. While causal and logical reasoning exhibit high cross-paradigm consistency, core executive functions and social cognition emerge robustly only within the interactive paradigm. This work establishes a novel, cognitively aligned evaluation framework that better captures human-relevant dimensions of LLM competence.

Technology Category

Machine Learning: Large Multimodal Models (LMMs)Cognitive Modeling & Cognitive Systems: Social Cognition And InteractionNatural Language Processing: (Large) Language Models

Application Category

Economics, Online Markets and Human Computation: Cost models of using LLMs in production systemsSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactionsUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

We examine three evaluation paradigms: large question-answering benchmarks (e.g., MMLU and BBH), interactive games (e.g., Signalling Games or Taboo), and cognitive tests (e.g., for working memory or theory of mind). First, we investigate which of the former two-benchmarks or games-is most effective at discriminating LLMs of varying quality. Then, inspired by human cognitive assessments, we compile a suite of targeted tests that measure cognitive abilities deemed essential for effective language use, and we investigate their correlation with model performance in benchmarks and games. Our analyses reveal that interactive games are superior to standard benchmarks in discriminating models. Causal and logical reasoning correlate with both static and interactive tests, while differences emerge regarding core executive functions and social/emotional skills, which correlate more with games. We advocate the development of new interactive benchmarks and targeted cognitive tasks inspired by assessing human abilities but designed specifically for LLMs.

Problem

• Evaluate LLM performance using diverse methods
• Compare benchmarks and games for model discrimination
• Develop cognitive tests tailored for language models

Innovation

• Interactive games surpass benchmarks
• Cognitive tests measure essential abilities
• Advocate new interactive LLM benchmarks