AI code assistants consolidate the most widely used languages.

[Picuino]

This seems logical. AI code-generating assistants require substantial amounts of human code for training. However, assistants for new languages have no previous code to train on. Given that generative AI significantly helps programmers, are we entering an era in which new languages will encounter significant resistance to adoption?

[Picuino]

This problem also applies to C programming in new microcontroller environments. When I first started using STM32, ChatGPT was very helpful in explaining the issues I was having with my programs. However, new microcontrollers or environments will not have that advantage, which could result in slower development of new features.

Ultimately, Asimov was right in his 1957 story "Profession." Reading and doing the slow work of learning still makes a lot of sense.

Edit: Here is a link to Asimov's short novel, a good read for this summer.
http://employees.oneonta.edu/blechmjb/JBpages/m360/Profession%20I%20Asimov.pdf

[Siwastaja]

Given that generative AI significantly helps programmers

Citation needed. Building a discussion based on a very controversial claim, which is very likely to be untrue, as some sort of self-evident assumption, makes very little sense.

[Siwastaja]

This problem also applies to C programming in new microcontroller environments. When I first started using STM32, ChatGPT was very helpful in explaining the issues I was having with my programs.

Nothing new here - just the same old thing using new words.

Why do you think ChatGPT is able to help you with STM32 problems? Because it scavenged the ST's forum, and many others, and is able to combine those pieces of practical experience. This is exactly where LLMs are most useful.

But this was same even before LLMs - then the thing was just called "googling". It's the good old argument: picking a popular microcontroller helps with availability of information, support, and especially peer-to-peer support on discussion forums etc.

For example, I started with Atmel AVR microcontrollers exactly because they were recommended on newsgroups in late 1990's - help was plentiful. That was the sole reason.

STM32's #1 reason for popularity is, I believe, the popularity itself. It already is that way - has always been - both in professional and hobby world, which makes it very difficult for new companies to push new product lines on the market: no one will use them because everyone uses what is familiar and which you can Google ask ChatGPT about.

[Picuino]

This problem also applies to C programming in new microcontroller environments. When I first started using STM32, ChatGPT was very helpful in explaining the issues I was having with my programs.

Nothing new here - just the same old thing using new words.

Yes, it's something new.
It's like saying that when Amazon appeared, it was not new because catalog sales already existed, and Amazon is the same thing.
However, new technologies are gradually turning Amazon into something completely different. Amazon is used much more widely than catalog sales, and it is much more concentrated in the hands of a single player who has complete control. This completely distorts the market.

I believe LLMs will follow the same path. Their ease of use multiplies the useful of old forums several times over and, for that reason, they are making forums (reddit, stack overflow, etc.) less useful for asking questions. It's similar to what Amazon is doing to small businesses.
Without forums for asking questions, LLMs won't be able to learn, and new technologies will decline.

It's just my opinion. I don't have any documentation to support it, but I believe it's accurate because it aligns with my personal experience with LLMs and Internet.

Edit:
Of course, I could also be completely wrong. Perhaps all of this is simply an irrational fear that LLMs will take away our creativity.
Asimov's novel depicts a similar world in which studying books is still necessary, even though almost no one believes in them.

[radiolistener]

The current capabilities of code-generating AI are very highly overestimated - what we're seeing is largely driven by hype. While AI can sometimes help you get started or offer ideas when you're tackling an unfamiliar problem, it's far from being a reliable assistant or autonomous coder.

In reality, AI lacks true understanding of software architecture, context, performance constraints, or long-term maintainability. It doesn’t reason like a human developer - it predicts plausible text based on patterns in its training data. This means its output contains subtle bugs, security issues, inefficient constructs, or completely non-relevant code that only an experienced programmer can catch and correct.

[Siwastaja]

I believe LLMs will follow the same path. Their ease of use multiplies the useful of old forums several times over and, for that reason, they are making forums (reddit, stack overflow, etc.) less useful for asking questions. It's similar to what Amazon is doing to small businesses.
Without forums for asking questions, LLMs won't be able to learn, and new technologies will decline.

Oh this I do understand and fully agree with. People asking questions from LLMs instead of discussing with other human beings has this fundamental problem that not much new data is coming in.

Even in "good old discussions" most of the comments were just obvious repetitions of others, easily replaceable by LLMs, but the hidden gems based on own testing / research / non-trivial thinking pattern are the most valuable. If people forget about the benefits of human interaction, these forums are in risk of dying out, and then AI will stay roughly similar to what it already is today; I don't believe in expectations of large breakthroughs because the LLMs already are trained on pretty extensive set of material.

[SiliconWizard]

TLDR; LLMs are a plague, to put it mildly, especially when misused. And "code assistants" is a straight case of misuse.

They can be used as a glorified search engine, to give you some pointers and ideas in sometimes more efficients ways than "googling" - you always have to double check anyway - but they are NOT designed for *correctness* whatsoever. This is why they'll just spit out stuff that may look remotely plausible to the layman but is often incorrect in one or many ways, and using "generators" have the huge drawback of making people avoid learning. Which is disastrous. I think life is all about learning.

Now of course they'll just "promote" already "popular" languages as that's what they have the most material to train on. In that regard though, it's not hugely different from the situation with only humans involved. Popular languages are those most taught, talked about online, so that promotes them in a feedback loop. The difference though is that people still learn along the way, and that's an opportunity for getting better, having new ideas and getting to know stuff that you haven't seen before too, so opening you to new things. Using LLMs, that opens you to nearly nothing at all. Just cut and paste.

And while there can be a lot of noise in human discussions as well, there is at least real interaction, with people (even if not all of them, at least some) genuinely willing to help and to share their own *real* experience of things that do work and they'll pride themselves in sharing that. And there is peer interaction too, where peers can correct whatever looks wrong, add insight, etc. LLMs have no pride, no morals, no ethics, they are not cross-checked by any "peer", they just generate a sequence of tokens in a one-sided way.

[radiolistener]

I believe LLMs will follow the same path. Their ease of use multiplies the useful of old forums several times over and, for that reason, they are making forums (reddit, stack overflow, etc.) less useful for asking questions.  It's similar to what Amazon is doing to small businesses.
Without forums for asking questions, LLMs won't be able to learn, and new technologies will decline.

Over 30 years ago, back in university, we had academic projects applying neural networks to industrial process control systems. I still recall the control system diagrams for an industrial process, which included a neural network module trained using system feedback signals. I remember asking a professor, who had been working in AI for decades whether it would ever be possible to create a program that could truly understand human language and solve problems. He assured me that the entire theoretical foundation for such systems had already been worked out in detail, and that it was only a matter of time and engineering effort to make them practical. And this was before the year 2000.

So in that sense, there's nothing fundamentally new about neural networks or even the ambitions of AI. What we're seeing now is not a revolution in concept, but a leap in practicality - particularly in natural language processing, thanks to larger datasets, faster hardware, and architectural refinements. Putting too much trust in LLM risks a harsh awakening.

Your vision of technological progress feels like a fairytale. The rise of powerful platforms and centralized control isn't just a question of technological progress - it's more deeply tied to questions of ethics, power, and ideology. It’s less about what the tool can do, and more about who holds the hammer and why.

“Digitalization” paints a picture of progress and innovation, but it is a mask. It often evokes an idealized image of high-tech progress - something modern, efficient, and intelligent, something related to digital electronics and innovation. But in reality, it refers not to innovation and not to digital electronics, but to digits as instruments of dehumanization. At its heart lies a process of peeling away identity, reducing people to cold digits, much like how prisoners in concentration camps were stripped of their names and assigned digits as a form of dehumanization, the same way people today are assigned these numbers. Similar to how a company owner conducts an inventory of assets by assigning identification numbers to items for managing their property. Where people cease to be free souls and instead become inventory - tagged, tracked, and optimized like livestock.

Seen this way, the current frenzy over LLM feels like a carefully composed spell - conditioning us to surrender critical judgment to faceless oracles devoid of conscience. Responsibility vanishes into the machine’s cold gears, allowing those in power to hide behind “neutral” code while casting morally bleak choices as mere outputs.

At the core of this machine beats a brutal heart: the wish to reduce human beings to obedient cogs, laboring quietly for another’s profit. In this system, there is no place for your individuality, no room for human rights - you are an object, someone’s property, someone’s slave with no rights.

This reality is far darker and more unsettling than even Orwell could have imagined when he wrote 1984. And it demands vigilance and serious reflection to not accept imposed rules in a form of numbers and structures to avoid the slide into a such bleak and bloody totalitarian system.

[JPortici]

Given that generative AI significantly helps programmers

bullshit. If anything it's the opposite in middle-long term.

using STM32, ChatGPT was very helpful

maybe so, because there is a lot of training material. Do that for any other architecture, it's extremely frustrating.
For example, microchip did their own agent which are pushing, which is invariantly wrong.

Well.. for one thing it mimicks the bad habits of doing one thing (assign input to X) and writing another in the comment ("this assigns input to Y") so the training material worked.. i guess?
and almost anything interfacing with the peripherals is the most absurd pile of shit

[DimitriP]

Similar to how a company owner conducts an inventory of assets by assigning identification numbers to items for managing their property. Where people cease to be free souls and instead become inventory - tagged, tracked, and optimized like livestock.

People have been "inventory" for a while now...whether they are called voters, consumers,prospects, viewers, web site visitors,job seekers, tourists,  patients or cell service subscribers depends on the business you are in ; unfortunately

[Siwastaja]

People have been "inventory" for a while now...whether they are called voters, consumers,prospects, viewers, web site visitors,job seekers, tourists,  patients or cell service subscribers depends on the business you are in ; unfortunately

For a while, yes...

I'm certain we heard this same discussion about this new recent selfishness and utilization of people thing already in ancient Greece.

[newbrain]

Well.. for one thing it mimicks the bad habits of doing one thing (assign input to X) and writing another in the comment ("this assigns input to Y") so the training material worked.. i guess?

That's a trait that I witnessed in a number of team members - AI is just getting more and more human!
One of them teammates brought the art to the extreme, comment differed from code, both did not match the specification.

For a three lines change, that was quite the record, and, from my very limited AI experiments, that's what often happens with code assistents.

Sometimes, nay, rarely, I get pleasantly surprised, I asked copilot to:
"Write a DSP filter based on equation b), in this markdown document with embedded LaTeX formulas, operating with frames of 512 samples and keeping all the needed state from one invocation to the next."
The result was a correct, clear and efficient C routine at the first shot.

[cfbsoftware]

are we entering an era in which new languages will encounter significant resistance to adoption?

ChatGPT seems to think so I put your question to it and the concluded 'Net Effect' was:

Unless mitigated, the AI era creates a high barrier to entry for general-purpose new languages. A language without robust AI support will:

Appear immature.
Require more manual effort.
Be harder to learn and trust.

As a result, the cost of innovation in language design rises, possibly leading to stagnation or consolidation around a few AI-friendly languages.

[Analog Kid]

Hmmph: I wonder if that isn't something like giving "AI" the following prompt:

"are you lying?"

[Siwastaja]

ChatGPT seems to think so ...

A language without robust AI support will:
... Require more manual effort.

Since the whole reason to use AI coding assistant is to fill in boilerplate, isn't this a great opportunity to start designing new generation of programming languages which make sense - with the focus on being able to describe the programming intent in clear, short, concise, unambiguous way, with least amount of boilerplate, so that no middlemen - LLMs or "classic" autogeneration tools - are required anymore?

Writing getters and setters for a C++ class is a classic example of programming gone horribly wrong, but after that, boilerplate has got much worse. New language designs should be solving the problems for programmers, not enslaving programmers by requiring them to write boilerplate, and then offload that to AI.