mwatkins comments on Linear encoding of character-level information in GPT-J token embeddings

[mwatkins]

Thanks for pointing this out! They should work now.