I decided to align this and the post following in this order because as I've said, I've been playing with LLM's a tonne in the recent days, and I have Thoughts.
First the good part: LLM's are amazing at a few things. Like mind-bending, scary, incredibly futuristic amazing, the sort of intelligence people in the sixties thought was only like a decade or two away. The computers can write essay, on any topic, even without material if you don't supply it, and they're darned good. And if you specify a specific style or format, they'll follow that. And explain to you what they did as well! Chat with you, they can do that too! More on that in the next post.
The limitations are...they really really don't seem to know what they're doing and lose the 'thread of conversation' pretty freaking quickly. And you have to keep reminding them the conversation, the context, and how to set it up. If you're expecting a super human genius who doesn't require babysitting, it's not here yet. It might get there, but that's not the case currently. Sometimes they are stuck in a loop. They do something stupid, you remind them to do the other thing, they apologize, and go back to doing literally the same thing, over and over and over, until you give up in frustration because if you wanted to experience this you'd get a child of your own!
And oh the unpredictability, that one's massive. Like on one occasion it'll give fantastic results, but the next time around it'll be totally different and you won't know what's different and why. To put it differently I guess, in more human terms, they have zero fucking personality and a disjoint identity. Which...makes sense, but not what humans are used to dealing with.
The one fear I have with 'big picture' is people forgetting to talk to and interacting with humans after spending too much time on ai, but uhhh there's bigger issues out there, probs.
No comments:
Post a Comment
Tell me what you think. I'll read, promise.