Do language models favor their home countries? Asymmetric propagation of positive misinformation and foreign influence audits

Comments

• muntedcrocodile@hilariouschaos.com ⁨1⁩ ⁨day⁩ ago

  Language models will favour whatever is in the training data. Their home country will have a majority of the language they are targeting. Thus is will favour its home country. Who could have seen this coming. Come on guys critical fucking thinking.

  source
• trolololol@lemmy.world ⁨10⁩ ⁨hours⁩ ago

  This research sounds legit but it should be called something like " our llms vs deepseek: does deepseek have a bias against English culture?"

  Since my culture is neither English nor Chinese, I think this is a d*CK measuring contest.

  source