https://github.com/danielpetho/unbaited/blob/main/extension/...
Seems like a reasonable idea. I have doubts about an LLM being able to reliably detect when a tweet was made to *deliberately* trigger emotions. That’s setting the bar quite high on llm working out intent
the default system prompt is quite weak, and could be much more specific / better.
also, even as a "human" can't decide sometimes what is a bait...
the default system prompt is quite weak, and could be much more specific / better.
also, even as a "human" can't decide sometimes what is a bait...