3 ms·Data Distributional Properties Drive Emergent In-Context Learning in Transformer1 points by tech-sucker 4y ago