Understanding LLMs
Subbarao discusses the misconception that large language models are performing complex Turing computations. Instead, he argues that they simply generate the next token in constant time, likening them to N-gram models. He also reflects on the implications of recent work by Greenblatt, emphasizing the need for caution in assuming that future models will solve every problem.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