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

[Joseph Bloom]

Fixed, thanks!