Ok let’s give a little bit of context. I will turn 40 yo in a couple of months and I’m a c++ software developer for more than 18 years. I enjoy to code, I enjoy to write “good” code, readable and so.
However since a few months, I become really afraid of the future of the job I like with the progress of artificial intelligence. Very often I don’t sleep at night because of this.
I fear that my job, while not completely disappearing, become a very boring job consisting in debugging code generated automatically, or that the job disappear.
For now, I’m not using AI, I have a few colleagues that do it but I do not want to because one, it remove a part of the coding I like and two I have the feeling that using it is cutting the branch I’m sit on, if you see what I mean. I fear that in a near future, ppl not using it will be fired because seen by the management as less productive…
Am I the only one feeling this way? I have the feeling all tech people are enthusiastic about AI.
Could you elaborate? I don’t have a deep knowledge of the field, I only write rudimentary scripts to make some ports of my job easier, but from the few videos on the subject that I saw, and from the few times I asked AI to write a piece of code for me, I’d say I share the OP’s worry. What would you say is something that humans add to programming that can’t (and can never be) replaced by AI?
I think the need for programmers will always be there, but there might be a transition towards higher abstraction levels. This has actually always been happening: we started with much focus on assembly languages where we put in machine code, but nowadays a much less portion of programmers are involved in those and do stuff in python, java or whatever. It is not essential to know stuff about garbage collection when you are writing an application, because the compiler already does that for you.
Programmers are there to tell a computer what to do. That includes telling a computer how to construct its own commands accordingly. So, giving instructions to an AI is also programming.
Yeah that’s what I was just thinking. Once we somehow synthesize this LLM into a new type of programming language it gets interesting. Maybe a more natural language that gets the gist of what you are trying to do. And then a unit test to see if it works. And then you verify. Not sure if that can work.
TBH I’m a bit shocked that programmers are already using AI to generate programming, I only program as a hobby any more. But it sounds interesting. If I can get more of my ideas done with less work I’d love it.
I think fundamentally, philosophically there are limits. Ultimately you need language to describe what you want to do. You need to understand the problem the “customer” has and formulate a solution and then break it down into solvable steps. AI could help with that but fundamentally it’s a question of describing and the limits of language.
Or maybe we’ll see brain interfaces that can capture some of the subtleties of intend from the programmer.
So maybe we’ll see the productivity of programmers rise by like 500% or something. But something tellse me (Jevons paradox) the economy would just use that increased productivity for more apps or more features. But maybe the needed qualifications for programmers will be reduced.
Or maybe we’ll see AI generating programming libraries and development suits that are more generalized libraries. Or like existing crusty libraries rewritten to be more versatile and easier to use by AI powered programmers. Maybe AI could help us create a vast library of more abstract / standard problem+solutions.
Generative neural networks are the latest tech bubble, and they’ll only be decreasing in quality from this point on as the human-generated text used to train them becomes more difficult to access.
One cannot trust the output of an LLM, so any programming task of note is still going to require a developer for proofreading and bugfixing. And if you have to pay a developer anyway, why bother paying for chatgpt?
It’s the same logic as Tesla’s “self-driving” cars, if you need a human in the loop then it isn’t really automation, just sparkling cruise control that isn’t worth the price tag.
I’m really looking forward to the bubble popping this year.
This year? Bold prediction.
You’re assuming we only keep trying the same things we’ve already tried. We aren’t.
We’re constantly improving the accuracy, efficiency, size, complexity, capabilities of this tech.
I agree with you about trust and everything else, but don’t assume it’s only getting worse, the transformer based architecture is getting a lot of focus and research by basically every institution in the world.
There’s a new math LLM trained on synthetic problems and about to do math at a university level, we might end up making it up new data like how you teach a kid to read from basic sentences.
But they are getting worse, specifically because their own outputs are getting fed to the next generation of language models as input. Google search, for example, is rapidly becoming worthless now that LLMs can be used to optimize for SEO. Meanwhile, Google’s own LLM is getting rebranded 'cuz last year’s tech demo couldn’t even regurgitate basic facts accurately and cost the company $100B.
It can’t reason. It can’t write novel high quality, high complexity code. It can only parrot what other had said.
90% of code is something already solved elsewhere though.
AI doesn’t know if the code copied is correct. It will stright up hallucinate non existing libraries just because they seem to look good at first glance.
Depends on how you set it. A RAG LLM verifies up against a set of sources, so that would be very unlikely in state of the art.