Promoting interactions between cognitive science and large language models

Abstract

Infusing cognitive science into model evaluation can guide the evolution of large models and improve model safety in human society; in turn, large language models as an emerging research subject of cognitive science offer test beds while emulating human-like judgments and enabling efficient cognitive tests. Interdisciplinary collaborations are necessary to break down barriers between academia and industry to jointly provide task sets, datasets, and model sets.

Penghui Du

MSc Neuro-X, EPFL · Visiting Graduate Student, Buckner Lab, Harvard

Neuro-X master’s student at EPFL and visiting graduate student in the Buckner Lab at Harvard, working on precision functional mapping and multimodal neuroimaging.