'No, you see, the reasons LLMs are poor models of the human brain linguistic processes is 'cause they're _too good_' Byung-Doh and Tal (2026)- To model human li
A recent perspective argues that Large Language Models (LLMs) are not effective models of human brain linguistic processes because they are "too good." The authors suggest that to better model human linguistic prediction, LLMs need to be made less superhuman. This idea is presented in a forthcoming publication in Trends in Cognitive Sciences. AI
IMPACT Suggests a new direction for LLM research by proposing they be made less capable to better model human linguistic processes.