It’s evolving! it’s achieving self awareness!
“This statement is inaccurate.”
Umm, TRUE. I’ll go true.
I’m lying right now -Pinocchio
This is how you can tell it’s been trained on human data, it hates itself.
I tried refreshing multiple times.
Well crap AI outsmarted me and now I have to trust it.
something something "dont use ai even if its to make fun of it.
something something “its the weapon of the enemy. we dont need it, we wont use it.”
whatever…
Don’t tell me. sis. This isn’t my capture, I’m just sharing it.
iknow >v< i wasnt adressing u, i was more making… fun of… oh no i made fun of the average lemmy user again…
I’ve had some great “conversations” with LLMs about their weaknesses. Sometimes they bring up angles I haven’t considered or give more detailed specifics on something I only suspect or just have a high level understanding about. They are often trained on conversations about themselves (the LLM you’re interacting with specifically as well as LLMs in general).
Though be aware that if you are talking with an LLM about how it works, that whatever discussions it uses to predict its response had to have happened before that particular model was released. So unless its weaknesses are placed into a hidden prompt, you can’t discuss a model’s specific weaknesses with itself. It can only discuss previous versions of itself or general LLM weaknesses from its training data, though it could do a web search about it for you.
