Two questions about Chatgpt and the current "AI" hype

February 11, 2023

Context

I am a software developer by trade, and thanks to the Kodsnack podcast I frequently get to talk to a lot of developers way smarter than I am. Some are very enthusiastic about Chatgpt and predict everything changing at once, any minute now. I was quite excited at first, earlier this year when image generators suddenly became fun to use, but now it feels like the hype has got completely out of hand. We recently recorded an(other) episode discussing the present and possibilities, and it occurred to me after recording that these two questions did not surface the way they perhaps should have.

1

Why is everyone so excited right now? Is there more "there" there than before? Perhaps the wave of entertaining image generators primed us to believe a somewhat better text generator had some kind of powerful new technology behind it. Perhaps the tech industry really thinks it needs a new buzzword. I am not sure we can reach crypto currency levels of scams, but this may be the closest thing in a long while.

2

Is there any reason to believe the problems can be solved? Nothing about the tech I have seen today gives me reason to think any of its downsides and challenges are easily solved. People are talking about fact checks and references and things like that. References I can kind of see, but is there a technical foundation for believing that anything about the language models of today can support or handle any kind of fact checking? The way I understand it, they are slightly glorified Markov chain generators and have absolutely no idea what their input or output text is about. Building an idea about the actual content, now that would be a big thing, but I see nothing pointing toward that in all the hoopla going on today.

Side question

Why are Microsoft and Google so desperate to wade into this? Both are suddenly willing to add a major new component to their search engines, of all things, a component which would probably most honestly be labeled "lies".

Would you like some checked facts about global warming from Wikipedia, or would you prefer some machine-generated improvisations on the subject? Why would I ever want this?

I do think there are useful areas for this type of "AI", but I think they look much less exciting and world-changing than the current hype circus wants to make us believe. Carefully constructed and applied in the right context, they will be magical, but not in any generally applicable sense.