CausalLM is not optimal for in-context learning.

Nan Ding, Tomer Levinboim, Jialin Wu, Sebastian Goodman, Radu Soricut

Browse the full ICLR paper archive.