Can LLMs Comprehend Sequential Structure? A Controlled Experiment on Reasoning and Generation
A study tests LLM sequential behavior, finding models may recognize strategies yet fail to generate them faithfully, so evaluations must check both.
A study tests LLM sequential behavior, finding models may recognize strategies yet fail to generate them faithfully, so evaluations must check both.