Magical Tokens in LLMs
Subbarao discusses the concept of "chain of thought" in relation to large language models, emphasizing that these models often mimic human-like reasoning without truly understanding it. He explores the idea of using specific "magical tokens" to enhance model performance and reflects on the challenges of sourcing these tokens, highlighting the role of human input and process supervision in guiding model behavior.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Subbarao Kambhampati - Do o1 models search?
Related Questions