Training large language models on narrow tasks can lead to broad misalignment

1 comments

• PaulHoule 2 hours ago

  Isn't this just "catastrophic forgetting?" e.g. training LLMs on anything leads them to get worse at what they learned before.