Language tuning doesn't protect AI models from disinformation

Original: Friend or Foe? Language as an ideological switch in open-weight LLMs under Russian disinformation stress

Source: arxiv.org ↗