I can't post anymore on twitter today unless I pay up. Which I'm not going to do.
I've been thinking about Fable, and I'm honestly not impressed. Unless these LLM prompt engineers know some kind of shortcut in design, you're just letting the AI do the design and praying that it sets everything up correctly.
And maybe I'm thinking about projects on a great scale, but consider something like this. You tell a construction crew to build the Empire State Building, using their best judgement.
No, you don't. You have architects and engineers who design, and then you have the manual labor.
I guess people might argue that AI is so good now that it can do the architecting and engineering part. And maybe it can.
But the more you leave the abstractions to the AI, the more trust you have to place in it. The more design choices it makes, the more trust you have to have that it made the right design choices. And I'm not saying it'll hallucinate. I'm sure it can architect much better than some humans.
The problem though is that things need to be reviewed and tested. Especially if you're trying to sell something and take people's money. It's one thing to make Evangelion demo scenes for fun on a portfolio, and it's a whole nother animal to sell SaaS to customers who you are obligated to support when shit hits the fan.
If my demo site breaks, who cares. I mean, if I really did care, I would review all of the code and make sure that the logic is sound. Otherwise we run into this trust issue again.
Fable sounds awesome, and I really do I want to fuck with it. But at the same time I am concerned how much autonomy it has to just make decisions, and we're expected to trust those decisions because it is a Mythical class artificial intelligence technology that is smarter than all humans combined.
My prediction is that the review process is the next phase of AI coding agent development. If you could make it easier to review AI generated code, then you can assign responsibility to a human who is obligated to explain their actions, instead of an AI saying 'oh my matrix math went wrong somewhere and I fucked up'
But who knows, the whole point I made all those study guides is so I could learn patterns. And then with the AI I can scan through the code looking for patterns instead of reading line by line.
Anyways rant over. More tweets tomorrow, maybe, or not. I guess it just depends on what interesting news happens in the end.