Reasoning in AI
The discussion delves into the complexities of sampling methods in language models and their implications for reasoning tasks. Jan highlights the distinction between the probability of paths versus answers, emphasizing the need for correct reasoning steps in tasks like code generation. Tim expresses intrigue over the tractability of using language models to guide searches, challenging assumptions about the exponential nature of the problem.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Daniel Franzen & Jan Disselhoff - ARC Prize 2024 winners
Related Questions