I don’t doubt that the amount of code will not reduce, it’ll just be easier and easier to get AI to fix it.
We are still less than two years into widespread use of this technology, and it’s surprising how good it is.
I am a ‘greybeard’ compiler guy and modern LLMs fix compiler bugs better than me, to a large extent. And it keeps slightly getting better every few weeks.
I firmly believe having LLMs generate code for current languages is a transition step, just like Assembly devs had to be convinced optimising compilers were generating the same kind of code they would write themselves.
They are not there yet, but the day will come.
But ignoring that for now, LLMs can already do better than modern compilers when it comes to optimizing code.
It has become a neat way of finding opportunities to implement in a compiler.
Give the LLM a compiler generated piece of assembly and it will sometimes spit out a slightly better version. Then you try and figure out if it's possible to implement it in the compiler. This works really well for blind spots in compilers like generating good vectorized code.
Yep, I guess you can train an LLM on a bunch of binaries to get it to mimic a SotA compiler with some accuracy, which may or may not improve over time, but come on. Times where there were free performance increases are gone, and this is not the area where shipping any bullshit real fast will get you any sort of advantage.
* Which are unlikely to happen again in the foreseeable future.