do you think the "bug" exist in the latent space? i don't think all bugs does.. the bugs exist as long as variant exist in the trained weights. until we have some kinda rl env for verifying bugs.. its never gonna work "well".
Very interesting one! So many cool things have been made in this field! However I think it's not relevant for the board. It's not open-sourced (at least yet).
Anyway thank you for checking it out and suggestion!