Email or username:

Password:

Forgot your password?
Top-level
Tobias Denkinger

@eugenialoli
The chief problem with code is not that it takes long to write, but that its quality is bad (and we therefore have to debug a lot). I can see how an LLM may be writing code faster. I don't see how an LLM can write better code. After all, the LLM learned to write code from examples of human code and has, in contrast to a human, no way of reflecting on what it has learned.
@nixCraft

20 comments
Tobias Denkinger

@eugenialoli
I believe we are close to what neural networks can do. We have been tinkering with them for the last 50+ years. In their current state they are not fit to replace a programmer. They probably never will be.

While there is a lot of other machine learning going on, this usually requires domain specific modelling. Such a model will be able to replace a programmer only after we have figured out how programming should actually be done.
@nixCraft

Eugenia L

@denki @nixCraft AI is with us since the mid-60s, yes. But AI has not yet reached its limits. It started slowly, and it's now on an exponential curve. We are NOT close to what neural networks can do. We're are the beginning of the exponential curve.

Tobias Denkinger

@eugenialoli
I was specifically talking about LLMs and neural networks, not AI in general.
@nixCraft

Eugenia L

@denki @nixCraft the market will use any tool in its disposal to further its goals.

crazyeddie

@denki @eugenialoli @nixCraft And the code models are trained on stackoverflow answers so... yeah.

Eugenia L

@denki @nixCraft Only a few software houses will care about "better" code, most would care about "faster-written" code. That way, they can reduce costs. This alone, makes the chance of someone young getting employed very slim, making the advice of "don't become a coder in the age of AI", a sound advice. You can still learn to code for fun, sure. But not to make a good buck.

Ariaflame

@eugenialoli @denki @nixCraft So it can fail faster. You do however need to know the basics of code to become a debugger, and those I suspect will be in demand.

Eugenia L

@ariaflame @denki @nixCraft Look at the state of modern furniture, household devices, clothes. Their quality is a far cry from the stuff you could buy in the 1970s. And yet, no one cares, everything is replaceable now, a commodity. Apart from some important code for banks/taxes/etc which will employ humans, all the other "apps" would be low quality AI code. They will fail often, and no one will care. People will move on.

Eugenia L

@ariaflame @denki @nixCraft It's not cynicism, it's objectivity. I'm old enough to remember the washing machine of my ex-mother in law, that she bought in Germany in 1963, and she still had it working until the mid-2000s. That's durability that is currently not built in to ANYTHING these days. Why would consumer apps be any different? Capitalism always strives to commodify everything, so it creates sizable markets with low entry point. That's how it "grows". It's not good, but it's what it is.

Eugenia L

@phi1997 @ariaflame Maybe you're right. Diogenes has always been my favorite philosopher. He told it like he saw it.

Ariaflame

@eugenialoli @denki @nixCraft Besides, learning coding isn't just about being a coder. It's about learning to problem solve, to break down a problem into pieces so small even a computer can understand them. At which point hopefully you do to.

Eugenia L

@ariaflame @denki @nixCraft Yes, as I wrote in the thread, learning to code for fun, is fine. But not to create a career that would give you the money you could make if you were born 20 years earlier. These days are gone, forever. The programmer as a sure fire professional with lots, LOTS of money per year, is gone.

Ariaflame

@eugenialoli @denki @nixCraft Most of the tools I'm using these days in my work didn't exist until I was in postgraduate study. I'm not sure that all programmers made lots of money anyway. People will adapt.

Eugenia L

@ariaflame @denki @nixCraft You can only adapt around tools that allow you to manipulate the tool in different directions. That's how you "adapt". AI does not allow you to do that. I estimate that eventually, within the next 30-40 years, AI will be having 85% of all jobs. Only a 15% of the population, the "overseeers", will have an actual job.

Ariaflame

@eugenialoli @denki @nixCraft I think you're overestimating AI's abilities.

Tobias Denkinger

@eugenialoli
Writing worse code faster is cheaper only in the short term. It produces technical debt that will have to be paid later. Many managers don't care about that because the problem is not big enough with human programmers that care about their craft. The problem will, however, become big enough to care once AI produces enough code.
@nixCraft

Eugenia L

@denki @nixCraft I replied on this in the thread elsewhere. Some crucial code will use humans. The majority will just use low quality AI code. Because it'd be commodified to pieces.

Fish Id Wardrobe

@eugenialoli @denki @nixCraft That ... wouldn't help. A bug in *any* of the code can be disastrous. It *all* has to be free of (serious) bugs.

Go Up