I feel like the thing a lot of people are missing about the AI explosion is that there will be new opportunities that didn't exist before. Sure, things will look a lot different than they used to, but we're going to be able to create systems that are far more complicated than anything we could make before. We're not going to simply automate 2022 levels of producing software and stop there, we're going to leverage this new technology to move faster and build more complicated things that wouldn't be realistic with hand-coding. I don't know what that will look like yet, but human history has shown that when we get some advanced technology, it's leveraged to create more than was ever possible before, not just stopping at making what existed in the past automatic.
Maybe compilers? instead of using them to make 1950s software faster, we used the abstraction they created to build vastly more complex systems like operating systems, browsers, cloud platforms, and modern games. AI could be another jump in abstraction, making software possible that would previously have been too expensive or complicated to build.
I disagree, a compiler is a deterministic tool while LLMs are not, you can be 100% sure that if you write "if a>b then c" the compiler will produce exactly that same statement in machine code, with LLMs not so sure, maybe it will give you that, maybe it will give a different validation, maybe it will hallucinate and do something completely different, and the worst part is that if you run it again with the same prompt you'll get a different answer, how can you build a solid building if your can't be sure that your workers are just making shit up with confidence, even worse somethings the do fine so is a russian roulette
I mean you can never be sure your workers aren't just making shit up with confidence. Like, the humans who work for you. And yet somehow, we've built all this software.
How to produce systems that work reliably out of a collaboration of large group with one person at the head who ultimately has responsibility is more or less the study of most human endeavors, this does not seem like an insurmountable problem.
jrflo · · focus · HN ↗
biophysboy · · focus · HN ↗
jackbrookes · · focus · HN ↗
throwaway1995v2 · · focus · HN ↗
joshjob42 · · focus · HN ↗
How to produce systems that work reliably out of a collaboration of large group with one person at the head who ultimately has responsibility is more or less the study of most human endeavors, this does not seem like an insurmountable problem.