Debate: Sparks versus embers

Sebastien Bubeck and Tom McCoy debate whether current LLM scaling methodology is sufficient to resolve major open mathematical conjectures.

Sebastien Bubeck (Open AI), Tom McCoy (Yale University), Anil Ananthaswamy (Simons Institute), Pavel Izmailov (Anthropic), Ankur Moitra (MIT) simons.berkeley.edu/talks/sebastien-bubeck-open-ai-2024-12-0… Unknown Futures of Generalization Debaters: Sebastien Bubeck (OpenAI), Tom McCoy (Yale) Discussants: Pavel Izmailov (Anthropic), Ankur Moitra (MIT) Moderator: Anil Ananthaswamy This debate is aimed at probing the unknown generalization limits of current LLMs. The motion is “Current LLM scaling methodology is sufficient to generate new proof techniques needed to resolve major open mathematical conjectures such as p!=np”. The debate will be between Sebastien Bubeck (proposition), the author of the “Sparks of AGI” paper arxiv.org/abs/2303.12712 and Tom McCoy (opposition) who is the author of the “Embers of Autoregression” paper arxiv.org/abs/2309.13638. The debate follows a strict format and is followed by an interactive discussion with Pavel Izmailov (Anthropic), Ankur Moitra (MIT) and the audience, moderated by journalist in-residence Anil Ananthaswamy.