LLMs are in a similar place, and still a long way from doing the whole job reliably by themselves. The current state of AI is one or several unknown unknowns away from real AGI. We're missing something fundamental.
Likewise with autonomous vehicles, the mythical SAE level 5 vehicle that goes anywhere and everywhere there is a road is still very much science fiction.
I ride in waymo all the time, they are SAE 4.
I know many people who take a waymo every day.
What percentage of driving represents the delta between 4 and 5?
SAE 4 level LLM/Ai (if we can really even make that comparison) would have far less difficulty in deployment and would be far more disruptive in a far shorter period of time than SAE 4 self driving cars.
Waymo's robotaxis, likewise, are able to do the driving task in geo-fenced areas. Waymo does a lot of hand-built code and testing to deal with particular problems, such as the 5-points in the Cairo district of SF, where it's a completely unique intersection, there's no other intersection quite like it. A ton of effort went into dealing with just that one intersection, and the bespoke effort doesn't generalize to any other intersection.
So if you want to be able to say, have a platform for producing working video games out of prompts, well, I believe that can be done with our current AI, but it will depend on a lot hard work making tools and hand-built code that do not generalize to other domain specific jobs.
Now if you want to make a movie worth watching out of prompts, that could be done too, but it depends on solving a whole different set of bespoke problems that once again need to be solved the hard way using more conventional software.
I think there’s too much hype and money, but LLMs are useful right now.