Large-scale cloze evaluation reveals that token prediction tasks are neither lexically nor semantically aligned

Add code
Oct 15, 2024

Share this with someone who'll enjoy it:

View paper onarxiv icon

Share this with someone who'll enjoy it: