Truth.

ChatGPT cannot write an application. But it CAN write blocks of code that I, with my knowledge, can assemble into an application, and it can do that much faster than I can. And those code blocks are often better thought out than what I would have written.

Working with it has made my coding speed go up by about 30% AND made my code cleaner. I still have to fact check and debug everything it does (it gets things hilariously wrong sometimes). As I get more used to using it, I imagine my output will go up even more.

This thing is very early. I could imagine a model that uses ChatGPT, in it's current state, as a sort of sub processor... Like a human being defines an application, the model (trained on a bunch of open source software) looks for applications similar to what was defined, then starts asking ChatGPT (as a separate layer) for blocks of code it can assemble into the final app. When it runs into emergent bugs when these blocks conflict, it asks ChatGPT to solve the underlaying problem. Then it runs the final output through a bunch of benchmarks and optimization layers. It could even ask something like Stable Diffusion for graphical components.

I actually don't think that kind of capability is that far off. I can imagine it and I think it could be assembled from parts we have right now, given time and effort. And yeah, the final result might need some human input to clean it up (and pentest it!), but the effort saved would be phenomenal.

The long term effects of this tech on the economy are going to be... Weird. Probably painful at first. The capitalist mindset is not well equipped to deal with the disruption these kinds of tools can cause. But it's also going to cause an explosion of human expression and creativity, and give people the ability to do things they couldn't before (thanks to Stable Diffusion, I can make art in a way I never could before, I haven't even begun to scratch the surface of what I might want to do with that). What an exciting, fun and scary time to be alive.