Lost in Transmission: When and Why LLMs Fail to Reason Globally

TL;DR

Despite their many successes, transformer-based large language models (LLMs) continue to struggle with tasks that require complex reasoning over large parts of their input…

Opening excerpt from the authors’ abstract. source