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

Tobias Schnabel, Kiran Tomlinson, Adith Swaminathan +1

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. We argu…