Compositionality in Meta ICL
Daniel and Sewon discuss the role of compositionality in meta ICL and explore how models can be trained to perform seemingly disjoint tasks. They mention recent papers that tackle this problem and express the elegance of the implementation. The conversation highlights the importance of understanding the extent of compositional generalization and the potential limitations that still remain.In this clip
From this podcast
The Gradient
Sewon Min: The Science of Natural Language
Related Questions