Last week, I tweeted this:
So, I guess I’m saying imagine some kind of, “mechanical turk” situation where it turns out OpenAI wasn’t actually using AI at all:
… imagine it was a person behind the scenes all along!
First of all, I would freak out but also be massively disappointed by the whole thing. It would also be such an epic troll borderlining on commercial and likely academic fraud.
But my real underlying reason for posting this tweet - I think we take for granted how compelling GPT-3 is. Just the fact that it can coherently respond so fast to the queries we give it, it has in many ways already surpassed human abilities.
It would simply not be possible to have this many people behind the curtain responding to every user query instantly in all these different ways. They’d have to type so fast and would run into limits of human cognition and some form of hand arthritis. GPT-3 isn’t just a compelling language model, it’s scaled out and doesn’t have hidden human realtime dependencies. It’s not an elborate illusion powered by low wage workers overseas, it’s really here, right under your fingertips. I certainly can’t say the same about Alexa, Google Home, or any of the other so-called, “intelligent agents”.
This is a tremendous achievement and as the technology improves it will only continue to blow us away. Speed and coherency at this level is no easy feat. I love the era we live in.