@jaredwhite I assume that you mean without AI?
But with compilers on a computer and not with pen and paper.
What is the main motivation for not using LLMs as part of the development process?
I am asking mainly to understand as I am head of studies at a department of computer science and we are currently getting ready to put more coding with LLMs into courses.
@UlrikNyman @jaredwhite My main motivation in not using LLMs is to maximize my own learning and retention, to avoid offloading critical thinking to digital landlords, to avoid paying for a service that will only ever continue to cost more, and to resist contributing to the inflated prices of hardware being caused by this bubble.
@skippy @UlrikNyman @jaredwhite Vibed code is also a hit to productivity in the sec & infrastructure space esp, much time lost to auditing. If I need automation, I reach for shell scripts that I have written. These I can trust. Gigantic black box simulation machines in datacenters in foreign countries, not so.
Further, software development is, in part, self development. I don't want to lose that through cognitive offloading
Be careful when pushing such tools on kids:
@JulianOliver @skippy @jaredwhite
We are not pushing them on kids.
We are investigating how the software industry are using them in practice now.
What they are using it for, which types of models.
And then we are adjusting the course curriculum of our computer science and software engineering educations such that the students that we educate can work in the industry that will hire them.
Jul 27, 2026, 4:39 PM@UlrikNyman @jaredwhite LLMs are non deterministic, so you can't know what it's doing and it potentially does different things when you run it a second time. They have no concept of architecture or design, just what is the statistically most likely next symbol. They are extremely inefficient because they have to re-read the whole codebase for every output token. Overuse encourages de-skilling, so you become more dependent on the LLM
The "main" reason is because they hurt people including the programmer.
AI data centers driving the major coding LLMs are publicly and visibly hurting people and the environment where they are built and exacerbating climate change with their energy use.
Coding LLMs are being used as a justification for mass layoffs, so again, they're hurting people by breaking labor power.
Coding LLMs are part of a hail mary to keep the infinite money glitches that tech had relied upon going and in the process is enriching outright fascists, deepening inequality, and setting us all up for a painful economic collapse. So, once again, it hurts people.
Also, the various technical reasons other folks will list, which again, hurts you as the programmer.
To be fair, some of this can be remediated through use of local, open models, but I don't see local LLMs used much in workplaces and that still doesn't remediate all the harms. Maybe academia could surprise me and would put some focus there.
It still has issues that the hardware for local coding LLMs is by no means cheap, but academia still charges hundreds for books and had dedicated labs with high grade hardware for running CAD software when I went, so it's not unreasonable to do.
But also, from personal experience with the output, at a very mundane level it destroy comprehensibility of the code base and documentation. People will use LLMs to generate documentation that is voluminous but also is targeted towards other LLMs rather than human comprehension.
The net result is that even if a stated rule is that people will review and understand the code, that becomes functionally unachievable. This becomes doubly problematic when grappling with generated documentation for integrating components in a larger system. The code changes downstream then become larger than they needed to be because the documentation was more verbose and less comprehensible than it needed to be.
These are negative outcomes that I have seen in action.
@UlrikNyman @jaredwhite there are a myriad of reason: technical, practical, business, legal, environmental, political, ethical, psychological ... . take a look at the the work of Emily M Bender, Timnit Gebru, and Paris Marks (all on the fedi) for a more rigours go over the above
@UlrikNyman @jaredwhite what does prompting a stochastic parrot have to do with the study of computer science?
Even if your program is less about theory and more about the tradecraft of programming, how does LLM use advance the students’ craftsmanship?
Unless you’re a trade school for a specific provider, the skills you impart should be general; what LLM-use skills generalize across providers?
What’s your plan for when investors stop subsidizing LLMs?
@UlrikNyman I do expect that to some extent AI may become a new level of abstraction. I do enjoy working in assembly or even in some cases toggling in raw instruction words on my Altair or SCELBI. Just as I enjoy high level coding in other contexts. I personally find the level of AI abstraction useful as well, but less rewarding. It's very interesting!
Assuming this is a genuine question, here are some reasons:
- keeping a planet we can live on
- antifascism
- respect for the work of creators
- not fond of speculative bubbles
Jul 27, 2026, 7:20 PM@UlrikNyman @jaredwhite
I see Aalborg University has a pro-AI head of department… good to know, future students can safely avoid that organisation for the purpose of learning about software development because they know this university won't teach them.
Industry needs people who can actually reason about code for themselves, not endlessly "chat" with a bot about what the code "might" do.
Jul 27, 2026, 8:21 PM@UlrikNyman @jaredwhite
If I wanted to be a manager, I would have studied for an MBA. I studied to become a developer because I like writing code.
Maybe try asking the phys ed teachers why they don't use cars in their courses.
@leeloo @UlrikNyman @jaredwhite I like that anology. Learning to program is very similar to training a muscle. That will not happen by using LLMs, similar to a sports student won’t gain muscles by driving around in a car…
Many reasons. Honestly the main reason for me lately is that it makes coding a miserable experience. Gambling, waiting, retrying. Then reviewing garbage and fixing things and updating agents.md, but the LLM ignores a lot of it anyway. Then more reviewing large PRs from peers that they didn't review well and so they are full of bugs that look good and need extra scrutiny to find. After correcting them, no one learns anything unlike the mentoring in traditional reviews.
Then there are the ethical concerns. LLMs break a social contract - people share their work on GitHub, Stack Overflow, etc and it lends to mutual learning and perhaps some connection and notoriety. Now that is vacuumed up by well funded organizations to be put in a blender and stripped of it's connection to you and sold back for profit to the already wealthy organizations.
(Open weights models are slightly better here, but it's still stripping the human connection and anonymizing)
@UlrikNyman
The LLMs also seem generally harmful to people using them, from deskilling to ruining relationships to propping up misinformation. GenAI typically includes image generation which seems to have no positive value at all.
The fourth paragraph here has a lot of relevant links: https://anthonymoser.github.io/writing/ai/haterdom/2025/08/26/i-am-an-ai-hater.html
Why would you integrate coding LLMs into courses? Whatever is the best practice now will be obsolete in a few months. And trying to learn CS by having an AI doing the work is like using a robot to lift weights for you.
@UlrikNyman @jaredwhite Jeg har undervist andre i software udvikling. Og AI frityresteger folks hjerner. (Og lærer dem dårlige design patterns.)
Hvis eleverne ikke lærer at tænke abstrakt om kode, så er jeres undervisningsmateriale ubrugeligt.
@UlrikNyman @jaredwhite at deprogrammere alt det dårlige information som AI giver folk tager langt mere tid og energi end at undervise dem ordenligt fra start.
Den kognitive overgivelse som kommer som et resultat gør det også svært for eleverne at lære nye koncepter, hvilket udruster dem dårligt til arbejdsmarkedet.
@UlrikNyman @jaredwhite Your question led to an interesting thread. Thank you!
Because we have a truly interesting dilemma.
The students that we had, that were in the middle of their studies when ChatGPT was released were kind of in the best situation. They had learned to program without AI, but know got to experiment with the new technology.
The choice of just telling the students not to use AI at all, does not seem like an option.
What we have been doing is keeping the courses kind of static and then letting the students experiment in projects
@UlrikNyman @EdTheDev you said "we are currently getting ready to put more coding with LLMs into courses"
Which means *you* are the one making the choice to have your students do that.
Or you could make the choice to teach the craft of writing software.
Setting your students up for failure right as the hype bubble is crashing seems unwise to me.
@UlrikNyman @jaredwhite "What is the main motivation for not using LLMs as part of the development process?"
The same reason you're not supposed to read the Cliff's Notes when you're studying literature: it defeats the purpose entirely. Your students won't actually *learn* a damn thing if they just get it all from the plagiarism bot.
But, here you are putting up a straw man argument.
Are intention is of course to just let them use LLMs from the beginning. But we are already experiencing that they are using LLMs for the exercises in our regular programming courses. They fail or get bad grades on the exam where they cannot use LLMs.
In their projects (half of all credits) they use can use LLMs, but here they still have to be able to explain and defend their code during the 5 hour oral group exam.
@UlrikNyman @jaredwhite if they fail when they can’t use the LLM then they’re not actually learning or retaining what you’re teaching. That seems like a pretty damning indictment of the thing in an academic setting…
@UlrikNyman "They fail or get bad grades on the exam where they cannot use LLMs." - that's just horrible and immensely disappointing.
@UlrikNyman @jaredwhite LLMs produce bad code. LLMs get in the way of understanding what the computer is doing, so you most definitely can't learn how to program with an LLM. LLMs cause a drastic loss of problem solving and critical thinking skills in their users which makes them highly addictive. They're trained on stolen material and are enormously damaging to the environment making their use unconscionable.
Personally, I feel like any one of those would be reason enough not to ever use them.
Jul 28, 2026, 5:22 AM1. GenAI er ødelæggende for rigtig mange ting på en gang:
- Scrapers står for 95%+ af internettrafikken til open source
- Kognitiv decline for dem der bruger det - ikke begrænset til det område brugerne anvender det
- Forstærkning af bias
- Kontroltab til uigennemsigtige firmaers agendaer
- Ingen reproducerbare resultater (universiteter burde jo stå helt af her)
- Moralske og etiske problemer ved udnyttelse af det globale syd (data-vetting) og ingen respekt for IP
Jul 28, 2026, 5:25 AM2. Det er et DSL, hvor du ikke kan få en manual til - og det vil give forskellige resultater til forskellige brugere. DSL'en ændrer sig igen ved næste version.
Som udvikler skal du prøve at holde en geleklump i form af en kugle.
Jul 28, 2026, 7:34 AM@UlrikNyman
when amnesty international calls for a global ban on using it because of how it damages human rights and environment, it's a good time to consider perhaps not using it.
https://www.amnesty.org/en/documents/pol40/0996/2026/en/
@UlrikNyman @jaredwhite Do you really still need to ask that?
Yes, because I clearly see a future where people use local LLMs on their own computers to help program the system that they want to create.
This can be from completely open source and open weight models in open source projects to company owned models on company software.
I do not like the centralization of LLMs to a few (mainly USA based) companies. This is definitely not the future that I want.
Our department are educating software engineers and computer scientists.
@rhialto @jaredwhite And the local software industry clearly tells us that our graduates need to both be able to read and understand code, write it "by hand", and be able to proficiently use LLMs in a integrated development method in some form of agentic software development.
So the question for us as a department is how do we take high school graduates and give them these competences within 5 years.
@UlrikNyman @rhialto "agentic software development" wasn't even a thing five years ago! Who knows what will be a thing in another five years.
Your job is to teach a craft that will last a lifetime, not Silicon Valley's hype du jour.
@UlrikNyman @jaredwhite If people need to be able to write code by hand, you certainly should not use LLMs. It has been shown that LLMs cause brain rot (less informally worded: deskilling). Many people have complained that they don't learn anything after having used an LLM. So that is a reason (on top of the immorality of LLMs and the fact that they destroy the world) not to use them, certainly not in education.
Apart from that, all arguments against LLMs in this thread should be well known already, including to you, so there really is no reason to demand that they are produced again and again.
@UlrikNyman @jaredwhite I'm making ecological choices to limit global warming. This includes: not using tools consuming huge amounts of water, electricity, rare metals, forest space...