Proceedings of the Annual Meeting of the Cognitive Science Society

Large language models (LLMs) can appear impressively capable across conversation, reasoning tasks, and standardized benchmarks. Yet merely appearing intelligent on traditional metrics is not the same as possessing robust, generalizable cognitive capacities. This disconnect highlights the need for more principled approaches to model evaluation. To address this gap, the symposium brings together four talks centered on developing a cognitive-science--inspired approach to assessing LLMs.