Reasoning in AI
Subbarao discusses the challenges faced by language models like GPT-4 when confronted with altered terminology in planning scenarios, highlighting their inability to reason effectively. He emphasizes that true reasoning should allow for problem-solving regardless of the labels used. Tim adds a definition of reasoning, framing it as the computation of knowledge through a world model, hinting at the deeper complexities of understanding and deriving truths.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Prof. Subbarao Kambhampati - LLMs don't reason, they memorize (ICML2024 2/13)
Related Questions