Language Model Reachability
Tim explores the fascinating concept of reachability in language models, highlighting the misalignment between sampling methods and the true capabilities of transformers. Jan adds depth by contrasting language contexts with more constrained frameworks, emphasizing how the limited token space allows for efficient sampling and solution evaluation. Together, they reveal the creative potential often overlooked in greedy sampling approaches.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Daniel Franzen & Jan Disselhoff - ARC Prize 2024 winners
Related Questions