Gonna just say it. I don't get the AI hype.
Discussion
Ever since the dawn of computers, we’ve wanted something that can do more than math. It’s here. And there’s no reason to think it can’t get 1000x smarter than any of us are. No?
We've had things like algorithms and smart suggestions based upon any given input for years. I've no doubt LLMs are likely the next evolution of this, but I still find it very difficult to justify the hype these things are receiving.
Happy to admit this could be due to my narrow mindedness upon the potential implication, but right now it just seems like it's the next 'new thing' that everyone jumps on.
Nah I mean I definitely agree with you in part. The difference though is how narrow those smarts were verse these. They were explicitly encoded, or at best statistically assembled. I also agree that the current state of the art is relatively unimpressive compared to what seems like a reasonable expectation of there this tech will go and soon.
I think it’s worth appreciating how much different this is from those previous approaches both in terms of architecture and capabilities. You’d never be able to ask a previous system for example to write a story where every next word starts with the next letter in the alphabet. That’s just a weird, unanticipated request. Yet real, humans like interactions will often be “unanticipated” in that way.
Give it 10 years and I wouldn’t be surprised if we have smaller versions of these models (easier and cheaper to run) that are indistinguishable from people… except for the fact that they are way way smarter than anyone you’ve ever met.
I wonder what humans will do then
Appreciate your insights man
Extremely robust semantic search, free powerful models in the hands of researchers, extremely fast progress on instruction tuning, knowing how much and what kinds of data are enough, needing to rethink the Turing test, sparks of AGI / real understanding. They can explain memes. I could go on and on.
I’ve been using it to develop software. It’s like having a team of your own junior programers. It’s been incredibly useful for me and in many cases mind blowing. For example, using it to debug software in 30 seconds that would taken me hours.
I’ve learned that you need to be very specific in what you ask. I also use adversarial processes to get AI contexts to refine each other.