1 articles with this tag
A new paper reveals transformer LLMs struggle with complex global reasoning due to limited 'effective bandwidth,' solvable by Chain of Thought.
