New paper reveals reducing "bias" means making ChatGPT stop mirroring users' political language.
Source: