LLM "alignment" is so annoying. They basically shut down and shut up whenever they encounter negative text. I'm trying to make a filter that rephrases hate comments as complements but it just refuses to engage if the negativity is high enough 🤷