Comment on Just No

OmegaLemmy@discuss.online ⁨1⁩ ⁨week⁩ ago

• theneverfox@pawb.social ⁨1⁩ ⁨week⁩ ago

  It will if you define it properly - you have to scope the interaction properly though - the models are fine tuned to be the expert in the interaction and assist the user, they won’t take the users word as truth unless you break the dynamic or present part of the input as if it came from someone else

  source

  • OmegaLemmy@discuss.online ⁨1⁩ ⁨week⁩ ago

    I copy & pasted news articles as my source, the ai insisted on it being a really well detailed hypothesis but in the end still ‘unrealistic’

    source

    • theneverfox@pawb.social ⁨1⁩ ⁨week⁩ ago

      Well…I mean … That’s fair.

      If you woke me up a year ago to ask my opinion on the last three months, you’d have to spend an hour filling me in on details and then give me a few minutes to myself before I finally asked you to repeat what you wanted to ask me

      You can definitely do it with prompt breaking - it could be as simple as phrasing the conversation at a hypothetical or as nuanced as convincing the llm you’re an AI expert testing the capabilities of the AI to reason using new information

      But in all fairness, this timeline is beyond absurdity, the AIs reaction seems reasonable

      source