What happens when programmers get to vote on a complete ban on slopcoded software on their code platform? Members of Codeberg were asked to vote on a proposal to completely ban slopcoded projects from Codeberg, and in what should not be a surprising outcome to anyone not overcome with “AI” hysteria, the vast majority voted in favour of the complete ban: over 70% of Codeberg members voted to ban slopcoded projects entirely (358 in favour, 144 against, 14 abstentions).
Of course, this is not a surprising outcome. Stripped down, programming is a form of artistic expression, and programming is no different than writing, painting, composing, or any other artistic endeavour. I think it’s safe to say writers, painters, composers, and similar creatives are against “AI”, so it’s only natural programmers feel the same; poll after poll shows the overwhelming majority of respondents – usually well over 70-80% – are against “AI”.
The pro-“AI” accounts on OSNews often try to paint my anti-“AI” position as extremist, but in reality, I’m just voicing how 70-80% of people clearly state they feel. I have zero skin in this game, zero outside pressure to please any bosses to get that promotion, zero pressure to conform to avoid getting laid off, zero pressure to not contradict upper-management. I can speak freely, openly, and without fear of retaliation. And as Nikhil Suresh explains in his harrowing from-the-trenches article AI Mania Is Eviscerating Global Decision-Making, that’s a massive asset.
The vast majority of people – including your friends, family, and co-workers – really hate “AI”. You can either accept this, or be left behind.
Reading the comments, the obvious reason for it was provided, AI produced code is heavily affected by copyright infringement and hence it represents a liability to host it, use it, redistribute it, sharing it … Even most open source licensees at minimum demand author attribution. And we all know, as it was publicly admitted, everything in existence and regardless of the license was thrown into it and now AI is sharing it with all original conditions stripped away.
Thom Holwerda,
This is clearly about my recent post. I don’t think you realize just how popular AI assistants are becoming with ordinary users who don’t have a horse in the race.
The 70% vote in the article was for the following amendment, which doesn’t actually ban AI coding assistants. It only bans projects consisting of mostly generated code. It’s a significant difference that AI helpers are still allowed.
I’m not really bothered by the policy. Ban projects generated without much thought and without an author who understands the code…I’m fine with that. But this doesn’t necessarily mean those 70% are against the use AI assistants altogether. If a developer works with AI to solve a problem, then it’s allowed by the policy and I don’t have a problem with this either.
In reality things are much more nuanced that you are admitting though. Most people are FOR AI assistants that answer questions and help them do tasks, find bugs, but against AI competing for resources, don’t want data centers in their backyards, AI spam and slop videos, and so on. In short there are good and bad aspects of AI, a single value is just not going to fairly reflect this nuance. I for one hate some of these bad AI use cases more than I like good AI uses cases. So if someone asked me if AI is good or bad, well the question itself is inadequate: AI is actually both.
I know you’re unlikely to change your ways, but I do think there are problems with your coverage that should be addressed. Like putting quotes around “AI” and pretending that context doesn’t matter when you trash it. Now it may well be your opinion that all AI is bad (you’ve refused to clarify), but I do think that is solidly into extremist territory and is not representative of how most people feel about AI. AI has so many applications that ordinary people would approve of, like cancer screening, helping the blind, helping the deaf, planning crop rotations, etc.
Of course it’s possible your opinion is actually more nuanced and less extreme than your posts suggest. Maybe your actual criticism is for specific uses of AI and not everything under the sun. If so, that’s fine, I’m genuinely curious to hear more about that. However I’d put to you that if there is confusion over nuances of your opinion, your own words are to blame since you routinely summarily blame all of “AI” as a generic concept. If this is not what you mean to say, then why not actually say what you do mean? Use more specific words like LLMs when what you are criticizing is LLMs.
Alfman,
The meme was
“Why isn’t AI doing more of household chores, but less of intellectual work?”
(Some image showing roomba vs office is usually with this, but need to search it to find an exact one)
People always assume “somebody else’s work” would be automated, and they would reap benefits. This is like someone else will pay taxes, or someone else will be laid off in the next round.
But none of us are actually safe.
Found it!: https://www.reddit.com/r/ChatGPT/comments/1jlryrw/its_an_older_tweet_sir_but_it_checks_out/
>I don’t think you realize just how popular AI assistants are becoming with ordinary users who don’t have a horse in the race.
How popular are they becoming? I’m finding figures like 15-25% of Americans using chatbots daily, and most of that seems to be because AI summaries and chatbots have essentially replaced google search for many people (though there’s talk of sites taking the nuclear option and delisting from google entirely).
https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/
LLMs have not really made a big difference in most peoples’ lives from a technical perspective. Now instead of reading the top 2-5 hits when they search for information they read a synthesis of those same top five hits. Instead of going to wikipedia they get basically get wikipedia read back to them by a chatbot. Hardly a game changer, just a time saver. The only respect in which I can see it being a game changer is for people who always sucked at using google effectively (e.g. my mom) can now have a natural language query.
“Such projects having an unclear copyright status ” – there is nothing nuanced about this. It’s just true, in a binary way. That is FAR from settled territory. You might have an opinion about how useful the tool is, and about how the law should follow based on that, but it’s NOT settled, really at all.
CaptainN-,
I take a bit of an issue with your citing one quote that talked about copyright status and then citing a different point about nuance, which was in response to a different quote altogether. You are effectively rebutting something I did not say.
As far as copyright status goes, in principal FOSS and LLMs are compatible. I have been saying that although it can make sense to be critical of LLMs/developers/etc who don’t abide by copyright terms, on the flip side we should be promoting LLMs that do respect FOSS license terms. Do you disagree with this view, and if so why specifically?
What I agree with is completely irrelevant to what the outcome will be. The outcome will be driven by power, not by any righteous understanding of copyright law.
CaptainN-,
Sure, my own opinions aren’t going to shape the outcome either. Nevertheless we still have opinions and I was asking about what yours is. Would you be ok with LLMs designed to explicitly respect copyrights? Or would you still be against LLMs that respected copyrights?
If you don’t want to answer, that’s fine too. The main reason I ask though is because while we can (and IMHO should) create LLMs that address the copyright concerns, it’s not immediately clear to me whether addressing those concerns in good faith would even matter in people’s minds.
I’ve been reading this site for a very long time – I very rarely jump into these types of conversations, but I feel like I need to here – despite the fact that this has practically become a religious argument and I know it’s probably not going to change anyone’s mind.
I’m actually disappointed that you, and this website by extension, are so vehemently against what is a technological advancement – whether y0u personally like it or not. It smacks of being like the old mainframe grey-beard operator who was never moving to a client/server model, the Netware guy who was “moving to that Windows crap” over his dead body, the tech who wanted to run dedicated hardware 1Us and 2Us in a rack instead of moving to a virtualization platform, or the sysadmin who dismissed cloud services as “just someone else’s server” while desperately trying to maintain uptime on-premises. They all got left behind.
Is there a lot of hype in the “AI Space” right now? Yes. If I go to the grocery store, are my Lucky Charms and Cool Ranch Doritos now coming with AI? Probably. The same thing happens with every tech cycle that I’ve seen in my life. Then the hype goes away. Then the real work gets done. Then it just becomes what it was supposed to be all along – integrated into our computing workflows.
My opinion is that using these LLM’s/weak AIs is exactly where computing was supposed to go, as far as I’ve seen it throughout my life. It’s another step towards the ultimate “future of computing”, whatever that is. They’ve been telling me that this was 5-10 years away since I read about it in my “Weekly Reader” in first grade 50+ years ago – along with nuclear fusion and the cure for cancer. And it was *always* 5-10 years away, until ChatGPT landed and truly disrupted. Is this the final version of this stuff? Absolutely not. I’ve seen ideas resurface over and over again – the first iteration is too soon, the second iteration is usable but isn’t quite right, the third iteration is where it becomes truly viable and usable. Platforms came and went. “Cloud Storage” popped up twice before we hit OneDrive or iCloud. The tech industry is littered with examples of this across hardware, software, and the web/cloud/et cetera. We’re probably in the second iteration of this now, but it is evolving quickly.
I’m guessing that the majority of the readers on this site have seen an episode or two of Star Trek: The Next Generation. I don’t once recall watching Mr. Data throwing on a hoodie, jacking a laptop into the holodeck, and then hacking away on C++ code for hour upon hour to build a program. The computer was prompted. You are hung up on, from what I have read, that it’s trained on copyrighted materials – that it uses FOSS code without attribution – it’s using proprietary knowledge without permission – it’s not creating anything – it’s all slop and crap and garbage. And I agree that some of it is. But, some of it is not. Using LLMs/AI can certain automate and orchestrate routine tasks – it CAN be a creative endeavor if the person behind it understands exactly what he or she AND the LLM is actually doing.
I’ve said this over the years more times than I can count – about PCs in general, about the web, about the cloud – This is a bus – either you get on it or you get run over by it. You don’t have to like this transition, but if you work in technology or are a technology enthusiast, don’t you owe it to yourself and the people who will come after you to understand it well enough to help shape what comes next? The conversation that is worth having is not whether AI is pure, ethical, intelligent, creative, or worthy of the hype. The real conversation we need to be having is “How do we shape it into something useful while limiting the damage it can cause”? The stand we need to take isn’t calling it slop, garbage, or crap and wishing it would go away. It’s recognizing that this is a social, legal, economic, and technological transformation – and then insist that we all get a say into how it develops.
We need to demand that transparency, attribution, privacy, and accountability is built into this. We need to ensure that we implement protections against abuse. We need to determine where these tools work, where these tools actually belong, and establish the boundaries of their use. We need to expose them when they fail – and examine how and why they failed, and what we can do to improve them and keep them from failing again. Crying foul and not contributing constructively is just leaving the decisions to the people, companies, and governments that are willing to make them.
Gee I was going to write a reply to your argument but then you cited a fictional android from the fictional future. Using fiction to back up one’s argument is very much in keeping with the LLM age.
Seems like a reasonable stance by Codeberg to me. If you love AI you can use Github and everyone will be happy.
I do agree with you: nothing wrong with creating some safe space, where you can cover your eyes and suck your thumb while pretending that the danger was not out there.
That is certainly a nuanced and moderate opinion but you forgot to include the nazi allegations.
What’s the point of childishly insulting people who actually care about code quality, maintainability, and fostering a community of developers? Why are AI-enthusiasts so satisfied at trashing skeptics? Why do good manners disappear when AI boosters like you engage with us?
If you’re right and we’re wrong then you don’t have to do anything, just wait for history to pass us by. If you’re right then you should just ignore us. The fact that you seem to have something to prove, in spite of your show of certainty, is telling.
Codewho?
You are so emotionally angered by AI that the positions you take seem particularly wild to me. You say…
> programming is a form of artistic expression, and programming is no different than writing, painting, composing, or any other artistic endeavour.
If that is so, than programming belongs to everyone because everyone is artistic.
> I think it’s safe to say writers, painters, composers, and similar creatives are against “AI”, so it’s only natural programmers feel the same;
You are playing gatekeeper – literally saying that only degreed writers can write, only trained painters can paint and only graduates in CS are allowed to program. Even daring to suggest that only certain people deserve the mantle of the word creative if they meet certain societal qualifications in a specific field. Wild…
Everything belongs to creatives and everyone is a creator. Full stop. Not every creative is going to program. Not every painter is an artist even if they can draw well. Not every degreed “programmer” is an architectural artist – many are just brick-layers.
Bricklaying is an incredibly difficult and complex trade that is highly gatekept and monitored for really fucking obvious reasons. The fact you use this skill in a derogatory sense is quite telling.
In what way is the comparison “software architect”+”code monkey” vs “architect” (actual one, for houses)+”brick-layer” derogatory?
Edit: bill has a point. not every programmer is an artist, who thinks about every line twice before they put it into writing. why would anyone for the major amount of work? there are so many code monkeys who work on tickets to implement features, fix bugs etc.
Thats a craft, not artistry
Edit 2: I really love to spent lots time with certain pieces of code… and then I’m very happy to produce art. 😀 but for work, most of it is UI glue etc that needs to work, that is testable,…
Bricklaying is a very basic job.
I think you are mistaking bricklaying for civil engineering.
Thom,
My father is a old-school learned bricklayer, who has had to earn is money as a bricklayer before studying civil engineering (HTWK Leipzig). I hold a Master in Civil Engineering (Universitaet Leipzig, 2003) and bricklaying was part of the required curriculum.
What you are saying is simply not true. Bricklaying is rather simple, you can master it by doing in less than 1 year. Plastering and Tile laying are much more delicate, but again not “incredibly difficult”.
Having standards that are actually relevant to the role at hand–like having skills and experience–isn’t what is usually meant by “gatekeeping”.
funny how only anti-AI news are posted around here… while the Major Pinguinista in Command himself tells AI haters to fuck off.
https://www.theregister.com/ai-and-ml/2026/07/15/linus-torvalds-tells-ai-haters-to-fork-off/5271894
Oh, I understood this was covered on OSNews, although possibly not in the manner you might have been hoping for: https://www.osnews.com/story/145548/follow-the-money-especially-in-open-source/
yes, and this decided me to pause my osnews patreon (only 6e but still paused.)
– I agree that AI is mostly slop-producing-tech, supported by neoliberal techbros
– I agree that using AI to replace artists is bad, machine produced “art” is not art, and don’t compare that to photography and others, it’s not. At least until the artists are given the possibility to train their own model, from scratch, then we can discuss but not before… And I say that because with friends we manage a small TTRPG publishing house, and we have to be clear that it’s a no-AI house, or only D0 and D1 on Dalia scale (grammar and language checks only). (https://daliascale.org/) But the point is that the whole endeaviour is a human one. TTRPG with players, books, writers, illustrators, cookies, beer, etc.
However, if I take computer programming, and most of the IT section of engineering and science (because it’s something I’m working on everyday)
– Computer programming is not art. It’s engineering, even if you are not an engineer.
– Computer programming is not art. It’s a craft, also, because designing things is not enough, you need the touch to assemble them properly.
– Computer programming is mostly copying what others have written, or making efforts to understand what previous coder left for you to work on.
– With AI tools, programming can also be a bit more secure. It so effing easy to botch security with current compilers and programming languages, that having a robot catching CVEs for you is really a boon !!!
– However, letting AI agent doing the code, e.g at D6 level, is dangerous : AI has not clues about what are the needs and what are the implicit requirements. This is the domain of the worst slop and shitty vide coded script kiddies stuff. No-code bullshit, resulting as always in more complexity and bad practices. This is also the realm of randomly managed cloud resources… well, that cannot go well.
To use AI agentic tools, to write code or manage infrastructure, you need strict specifications, to reduce unplanned changes caused by the chaotic nature of the neural networks. So, that means that you have to *write code*, but this time in English (or French, or Klingon, or Japanese), then the AI will compile that to a source code result, that will be compiled as executables by the compiler du jour. But, to get proper result, you have to think properly too. Otherwise, and to paraphrase the Monty Pythons, it’s slop slop slop ! lovely slop ! wonderfull slop !
Now, what I want is a proper local model to not depend on frontier thieves.
People tend to fear what they do not understand, whether it is a scientific concept, a new technology, or an unfamiliar tool.
Old Usenet flamewars provide some hilarious examples. People who did not understand compilers would rant against using them in production OS development, just as others confidently dismissed the web as a passing fad, etc, etc.
AI is simply the latest train a great many people seem determined to miss.
I’m really tired of AI-pushers claiming that skeptics “fear what they do not understand”. In my _personal_ experience, the people most excited about generative AI are those who understand it the least. They might have a vague idea of what a NN is, but they’ve never even heard of gradient descent, or know what interpolation is, how nonlinearity is introduced into NN, what a context window is, etc. IME they make unfounded assumptions about how generative AI works behind the scenes (e.g. that they are somehow training it just because it remembers stuff from its context window/local metadata) and then make unwise extrapolations about future behavior.
Suspicion about AI is totally warranted. As yet we don’t know how it will turn out – what it’s effects on our society will be. I’m quite amazed at the number of people who happily get on the band-wagon, seemingly without a moment’s critical thought. There _are_ examples of tech turning out be be detrimental to our society – social media is a big one. Results are hard to predict. Unfortunately, it looks like AI will overtake us, and only afterwards will we see what the outcomes for us all are.
Shifu,
Of course. I am convinced that AI is going to make make millions of long hull truckers obsolete within a decade (at least if states don’t stop it). Generative AI is already creating high quality VFX today. The truth is VFX artists were already having a hard time, AI could be the nail in the coffin for them. AI is still in it’s infancy for large programming projects from scratch,, but it’s already being used to successfully find and patch vulnerabilities.. In a few years time more software developers could also forced to contend with their own redundancy.
It’s not just jobs, sometimes people are putting AI to bad uses. Spammers are latching onto AI because of it’s effectiveness at generating custom tailored advertising spam. Foreign governments are probably using generative AI to gaslight the public. Meanwhile we’re dealing with AI related parts shortages and price hikes. The dating scene is so bad that even relationships are being replaced by AI, and I’m sure the sex robots are coming.
I think the view that AI is all good or all bad is incomplete because AI is just a tool. Its goodness or badness really comes down to how we use it, which is why I want to see more people involved in shaping AI positively.
It’s always interesting to read your posts because you generally try to take a balanced line. I’m sure most people would agree that there are positive and negative uses for AI. Issues around bias aside, many uses for classic non-generative AI are pretty uncontentious.
But this discussion between you and Shifu really exposes the fault line. You think there’s scope to shape the future direction of AI. That’s a very optimistic view. I suspect a lot of people feel quite powerless when faced with AI development and don’t have faith that the people and organisations who actually will shape future direction have their best interests at heart.
Bingo. That’s the real issue. I don’t think there’s anyone who disregards the hypothetical usefulness of LLMs, but the practical and ethical concerns are so massive it’s impossible for – apparently – 70-80% of people to set them aside. The destruction of the environment, massive data centres causing a whole slew of problems, the wholesale theft of other people’s work on an industrial scale, the further concentration of wealth in the hands of the few, the wave of slop destroying the web even further – and that’s just a selection.
If you can solve these issues, people will change their minds. Up until then, the cost-benefit analysis just does not favour the vast majority of people.
Thom Holwerda,
The thing that I would like to convince you and others to do, to the extent that it’s even possible at all, is to offer more constructive criticism that helps shape AI into a better future for humanity. I think the media plays a critical role here but it does mean covering the good, and not just the bad. If you take issue with how models are trained, then go proactively call for/find/promote/encourage the world to develop & embrace community friendly models that aim to respect FOSS licenses. Osnews could be a valuable asset in promoting projects to tackle the issues, but obviously to date it’s all been blanket hate with no desire to distinguish between good and bad AI efforts. 100% of the coverage is hateful and non-constructive. AI is going to be a part of the future though and our shared responsibility to the future is to make it the best future we can. We need to promote good AI today so it has a chance to succeed in the future. Otherwise our failure to positively shape AI around our common social needs will leave AI and the future in the hands of technocrats who you know won’t give a damn about any of us. Unfortunately I think the future is leaning in their favor because we are doing such a bad job at promoting good alternatives.
Alfman,
What makes the discourse even more difficult is that adversaries of the Western countries, China, russia and Iran are actively trying to sabotage the public perceptions
https://www.nytimes.com/2026/07/09/business/china-russia-ai-data-centers.html
For example China was found to be funding the anti-datacenter hysteria (They only have 1/10th the capacity, and cannot catch up organically. So they artificially try to slow us down). And most people who believe in nonsense like “datacenters use up our water” do not actually try to check the real data on water usage beyond the propaganda veil.
This won’t be an easy process, it will actually be rocky, but I believe there is hope, as Open Source and Open Weight models are now barely ~6 months behind the SoTA from frontier labs. The hardware is also accessible from many vendors, and some models are now focusing on “clean” data heritages (like Hugging Face’s SMOL)
https://huggingface.co/collections/HuggingFaceTB/smollm3
https://huggingface.co/HuggingFaceTB/SmolVLM2-2.2B-Instruct
I criticized you earlier but this is the type of article you should be writing about AI. These are legitimate real concerns. I have them too. It’s a completely different subject matter vs the gatekeeping suggestions about who has the right to enter a field – which causes an escalation.
AI is not going away because governments are keen to weaponize it. They have already done so – domestically and internationally. Guardrails are being removed for a long time for the ruling class. Businesses are even being given the right to vote in the US because they lean pro-fascist. The US government, for example, has demanded that AI guardrails be removed for military operations.
There are lots of substantive arguments that can be made against AI. Maybe the only solution you see is to destroy all of AI but your AI articles dance around this conclusion.
sukru,
> And most people who believe in nonsense like “datacenters use up our water” do not actually try to check the real data on water usage beyond the propaganda veil.
So where is the real data behind water AND electricity usage except behind a paywalled NYT article?
OlaTheGhost
This is pretty easy to look up
Globally only 1.5% of power usage is in datacenters:
https://ourworldindata.org/how-much-energy-do-data-centers-and-artificial-intelligence-use
For water they consume less than 1% of the US supply
https://www.nature.com/articles/s41545-021-00101-w
Please do not fall for documented Chinese astroturfing campaigns.
https://americanenergyinstitute.com/docs/aei_data-center_report_v2.pdf
We should be better than just following propaganda, and be able to look up actual numbers.
sukru,
Data is important, but the trouble is that most data centers just aren’t reporting these figures directly and there’s a wide variance in how one can extrapolate data in the gaps. Much of the data precedes the recent boom too. Water use can technically go down with solar energy, which is good news and data centers like amazon’s have pledges to source all their energy from solar in the coming years. In reality though the the grid is a big pool and paying the solar farms that fill the pool looks good on paper yet in practice might be little more than an accounting maneuver that shifts payments for coal to other consumers without having materially changing anything at all.
Most of these things are acknowledged in your article, and my point isn’t really to be contradictory, just to illustrate that it leaves enough room to make the case that things are getting better or things are getting worse. We don’t really have a nice easy answer.
I agree with your point, we do need to be very careful and I appreciate your calling it out. That said though the best propaganda always has a nugget of truth. And personally I don’t think it’s wise to divert ANY water for data centers in areas of the country that are already depleting water sources.
It is well understood that farming is a huge consumer, but it’s also extremely important that farms not fail. Global warming exacerbates this. Yet even on the farming side things are complicated since many farms are growing crops for Ethanol for energy purposes rather than food. The more you dig, the more you unravel a web of complications.
flypig,
I am pessimistic by nature as well. I concede the difficulties and skepticism are real. That said I genuinely believe there’s a lot we can do working together. Momentum builds on unity. but too much infighting and things just peter out.
Obviously on an individual level not everyone is going to want to participate in building a better future for whatever reason, such as “too difficult”. It is what it is. But the real loss of momentum happens when we start discouraging others from trying to build a better future – like a pessimism multiplier: “I’m not going to try and you shouldn’t either“.
Many long haul truckers already are obsolete and have been since before long haul trucking was invented. The amount of long haul trucking we do vs. rail (which is far older) is primarily a function of government regulations, massive government subsidies for cross-country highways, rail unions artificially driving up headcounts and wages. Rail unions, air traffic unions, longshoremen unions etc. already successfully block many labor-saving innovations in America that the rest of the developed world has had for decades.
We already know how this is turning out: AI tools are firmly embedded in many design and development workflows within industry.
E.g. Several of our architectures have already been developed using AI-first processes, and the resulting gains in agility and productivity have been monumental, and they continue to improve at a pace we have not seen before. There is no going back. Abandoning these tools/flows now would amount to professional or commercial suicide, particularly when every one of our competitors is adopting the same approach.
Everything and every tool can have a positive or negative effect. Fire can either help you cook your food, or burn your house down.
Why is it that I never heard of any real world success stories in this supposedly world-changing technology that aren’t either orchestrated by the LLM-providers themselves or aren’t just vague unconfirmed anecdotes like yours?
What other world-changing technology not only failed to turn a profit in spite of having plenty of infrastructure and user access (the dot-com bubble era is _not_ comparable) but also relied so heavily on tech demos from the vendor and “trust me bro” stories on social media to justify the hype?
Why is there still no vertical taskbar in Windows 11 if these features were so powerful for anything other than generic CRUD apps? That should be an easy one-shot fix according to every tech demo and “trust me bro” story I’ve seen.
Thom, you are still missing the point. I don’t want to give up on you, because I have too much respect for all the good work that you did before becoming a fanatic embarked on an anti AI crusade, so I will try again.
Can AI produce slop? Yes. Is a lot of AI generated code sloppy and crappy? Yes.
Is a coding assistant in the hands of a capable engineer a productivity multiplier? Yes.
Can coding assistants if used critically produce high quality code? Also yes.
Coding is no art. It is 99% grudge. Coding agents take that grudge away, and let you spend more time on the big picture. They are not perfect, but they are already pretty good tools, and will keep improving.
Do they raise concerns about control, copyright, openness, energy efficiency and what not? Certainly yes, though not all in the same measure and in the same way. But these are all orthogonal issues that are not inherent with the technology.
Open-weight, ethically trained models that can be served on a single GPU do exist. They are not as good as frontier models, but they will get better over time, and I bet that eventually will become the go-to solution for most engineers (because we do like control, and do not want to pay a subscription if we can do without it).
With your inflammatory, religious attitude you are missing an opportunity to help your readers form an informed opinion and navigate the landscape critically.
I come to OSNews for a balanced view. But I don’t expect to see it in the headline articles. After more than two decades, I’ve come to appreciate how Thom sticks to a principled line that often diverges from the mainstream.
The balance comes from reading the thoughtful replies in the comments. All I’m saying is that I appreciate and enjoy reading the healthy discussion from both sides.
> The balance comes from reading the thoughtful replies in the comments. All I’m saying is that I appreciate and enjoy reading the healthy discussion from both sides.
Too bad Andreas is banned now
Not yet, although I respect Thom for only deleting comments (so far), but not striking the hammer yet.
Ah okay – I saw your comment gone, and on your profile your avatar is also replaced by a broken “banned_user.jpg” haha. That made me jump to that conclusion.
Nice that you’re still here. I do not alway agree with what you post, but I appreciate the perspective.
I would like to return this compliment to you and everyone else posting here, including Morgan.
One can argue that this _is_ the balance to all of the hype you can easily find everywhere else.
> The vast majority of people – including your friends, family, and co-workers – really hate “AI”.
Then surely there’s nothing we can do about the majority of people who support right-wing parties in the UK and other countries, and we shouldn’t even try, right?
Imagine that LLMs will come up with the cure for cancer. Would you still be against them? Is this so unlikely? One thing is relatively certain: the next drugs will benefit from the results from AlphaFold. Yes, AlphaFold is not LLM-based. But the people who brought us AlphaFold believe it’s a good avenue to explore. Further, LLMs have started to produce new math results. Yes, they’re still limited, and right now they’re mostly coming up with counterexamples disproving conjectures (really interesting, if you want to read for example Terence Tao’s recent article about the Jacobian conjecture), but not only (see for instance Donald Knuth’s paper on Claude Cycles).
Regarding LLMs being trained on stolen data. I just don’t understand this argument. All the advancements we make are based on previous work. We are all “standing on the shoulders of giants”. Would a mathematician consider his work stolen if it was used to come up with something useful?
I get that there’s an artistic side to programming and I hope it endures. But when the NHS pays a programmer they’re not paying them to be artistic, they’re paying them to help cure someone.
Sorry, I would have provided some links on the subject, but last time I tried to post them my comment got flagged for moderation and disappeared. Sent Thom an email about it but it seems to have gone into a black hole.
I wish I could say I respect other’s opinions on AI, and I like to think that for the most part I do. But if I am partly optimistic about LLMs (not least because I have ulcerative colitis and LLMs my help produce my next treatment) and someone calls me an “AI hysteric”, what position should I have towards these people?
I’m sorry to play that card, but I believe it’s important. It may be that there isn’t a scenario in which humanity gets to hold on to the positive outcomes of AI without preventing very bad negative ones, but it would be a really sad scenario, wouldn’t it? If you hold this position it’s not entirely indefensible, but I sure hope you have some strong reasoning behind it. Hell, you’re making a lot of decisions for the rest of us.
I think people’s attitudes toward AI are often shaped by two things: how well they understand the technology and whether they expect it to enhance or threaten their professional security.
You rightly see AI as a tool, something that can help create products and services with the potential to improve your wellbeing substantially. Thom, if I remember correctly, works as a translator, so he may perceive the same technology primarily as a direct threat to his livelihood. Those different incentives naturally produce very different emotional responses.
There is also a group that desperately wants AI to be a bubble, partly because they lack the opportunity to invest in it. Rather than accept that others may benefit from something they cannot participate in, they hope the entire sector collapses and everyone else fails with it. Misery loving company etc, etc.
What fascinates me is how emotionally charged and reflexively cynical much of the online discourse around AI has become. There are legitimate criticisms, obviously, but experience has taught me that technological cynicism is often used to disguise a lack of understanding, education, or practical experience. People perform certainty by becoming increasingly dismissive, as though the harshness of an opinion somehow establishes the authority behind it. That seems to be a fairly universal on internet tech discussions at this point.
Something that isn’t an LLM being useful is not evidence for an LLM being useful.
This is why we need to stop talking about “AI” and start talking about the specific types of tech that have been lumped in under the same term.
You don’t understand how somebody could sink weeks, months, years of their life into a project and then graciously share it with others for free in the expectation that it will be used by a fellowship of like-minded developers acting in good faith and that the terms of the license will be adhered to (whether it be simple attribution or sharing of modified code) and that the people who build on their work will in turn continue to service that community and then get upset when it’s used for commercial purposes with zero payment, attribution, sharing of modified code and the thing they’re using it for is designed to replace you and your community with crude imitations and also drown you in slop PRs before finishing you off?
Thom,
Here’s a fellow ex-translator. Yes, in the past two years I lost my means of living to AI (or “AI” as you’ll put it). I had been getting 800 days worth of jobs a year, now I don’t get 80. Next year I’ll not get any. And I didn’t prepare for this looming collapse in the past decade. In that sense, you did much better than me, now that at least people pay you money to see you “suffer” with Windows for a month. I’m not jealous about it though. Good for you. You worked hard on Osnews, and this is at least well earned relief for you.
I am also not a fan of most uses of AI in their current form. I even took the time and built an adblock/ublock filter for AI-generated music channels on youtube (if anyone is interested, https://github.com/cevvalkoala/CevvalYoutubeAIBlocklist/ ).
On the other hand, whenever I read an AI-related comment by you, my brain screams “Luddite”.
LLM-based AI paradigm won’t disappear. It will grow and grow and grow. This year it took my job. Next year it will take a graphic designer’s job. Then it will come for a coder’s, a truck driver’s, even for some doctors’. This is obvious for the short-to-medium term. Maybe a pediatrician’s job will be safer compared to that of a radiologist. Perhaps carpenters will still get to retire safely (assuming our economic paradigm survives that long) while accountants will have reason to panic. Teachers’ jobs will be harder to take over for the AI, but the demand for teachers will surely fall as students (and their parents) will no longer see the point in studying for decades. And so on and on. We’ll have to accept the laws of physics even if we may not like them. I’m not looking forward to being pulled into a black hole for instance, but if one appears nearby, it will take me in complete disregard of my feelings. Facts are pretty stubborn things. They don’t care. And I’ll build the rest of my comment on what Bernard Baruch once wrote: “Every man has a right to be wrong in his opinions. But no man has a right to be wrong about his facts.”
Nope. Some are afraid of it. I’m terrified of it as I’m deprived of my career and probably I won’t have the chance to build a new one at age 48. And I don’t know how I will earn a living for the next 20 years. This gives me shivers. But I don’t hate the existence of AI, because that’s a fact. My wife is also not happy with me losing my income, but she’s happy when Chatgpt gives her advice for the next baby stroller we need. In the same vein, I’m also happy as Claude made it possible for me to build a VPS-based system to save $200 a year for my various subscriptions. That was an impossible task for me earlier, and I used to pay namecheap, bazqux, surfshark and so on a lot of money each year. Now, I pay a minuscule sum to a VPS provider, and get comparable services I designed and built, with the help of an LLM.
Everybody has some story like that. My cousin likes offloading grunt work of coding to LLM. My neighbor is now able to scan through hundreds of CVs to find good applicants for jobs listed by her HR firm. My aunt really likes the tip of adding some lemon juice to the lentil dish she used to cook for 60 years, and can’t boast enough about how Chatgpt made her cooking better. My friend who organizes congresses and conventions loves avoiding excessive sums he used to pay to simultaneous interpreters.
To be sure, they all have some reason to be worried about. My aunt is worried that my cousin won’t find a good job because all jobs are being eaten by AI. My HR-firm-owner neighbor worries that her market will shrink as firms will no longer hire that many human beings. My wife isn’t happy about how her boss pushes more and more tasks to her, saying “just do these, AI will make it possible for you”. The rest of the society have similar stories. Sure, these are all anecdothal evidence. Maybe I’m in an echo chamber. But maybe you are. And the burden of proof lies with the one claiming that “The vast majority of people – including your friends, family, and co-workers – really hate “AI”.” And no, that futurism.com article doesn’t count. Do you think the NBC News poll it refers to is something to build your argument on?
You’re right, there are a multitude of reasons for being suspicious of, even hostile to AI. Datacenters are one. Black-box processes are another. Growing dependence on a techno-capitalist clique is also terrifying. And don’t let me start with RAMageddon. These aren’t issues that can’t be solved though. A good deal of sensible regulation can make all these better. And regulation will surely come, first in the EU, and then the rest of the world once that orange lunatic is gone.
But you’re not right in claiming “programming is a form of artistic expression, and programming is no different than writing, painting, composing, or any other artistic endeavour”. Art and craft aren’t the same. Would it make sense if Luddites claimed their textile products were works of art? Why would it make sense for a programmer to claim he or she engages in art? The society is aware of this fact. That is why we have been training programmers in engineering schools rather than art schools.
And here’s a rather radical argument, one which will not find many subscribers: Assuming LLMs are based on violation of intellectual property rules, so what? Intellectual property is the invention of the past 3 centuries, and it has been growing to an unreasonable extent in terms of scope and length. I have no qualms about it being curbed down. If LLMs will be the pretext for doing away with that anomaly in human history, let them be. Sure, replacing the multitude of intellectual property ownership with 3 evil mega corporations will not be good, but we have a good opportunity for reining them in through regulation. Let’s take this opportunity to cut back the scope of intellectual property, and expand the scope of public domain.
One of the best comments I have ever read here.
Respect man. I wish you all the best. Hopefully you can figure out how to make money in this crazy world we are in.
Very happy to see OSNews staying strong as a bastion of sanity, Thom, and equally delighted to see this Codeberg decision. Let the boosters yell all they want. What else can they do?
FWIW, in my personal life I do not know anyone who hates AI. Everyone from grandma to kids uses it for all sorts of things. Are people worried where it might go? Yes. I’m worried too, but the utility is pretty clear.
My objection is not that Thom is an extremest, which he is. It is that he is uninformed. To properly criticize the thing you have to study it and use it, otherwise you are taking a purely political stance, and I don’t like that.
Run a fundraiser that makes you use AI for everything you do for a while. I for one will give you some money.
There’s now a blog post available covering the change in Codeberg policy. I don’t think the intention is to necessarily explain why people voted the way they did (I don’t suppose anybody really knows that) but it does detail why the proposal was put forwards in the first place:
https://blog.codeberg.org/protecting-our-floss-commons-from-llms.html
flypig,
This is informative, thanks for posting!
While I don’t agree with everything they say (most FOSS licenses are actually compatible with LLMs in principal). But at least they do acknowledge that some of their users are using or accepting some LLM code and codeberg doesn’t intend to block it in cases that don’t really cause project problems. They recognize that not all LLM users are being problematic.
Consider…
IMHO it’s really “causing problems” that should be the focus of the policy regardless of whether they use LLMs or not. Anyway it sounds like they’re being reasonable about it.
Thank you for sticking it out, Thom. I’ve always just loved computing and computers. I swore off retrocomputing years ago, and have recently plunged back into it just because “modern” became synonymous with “broken” as a relative descriptor (i.e. “more modern” = “more broken”).
There’s little good or interesting tech news any more, and I come here because what little there is, you post, and you post sane opinions about AI. It’s a shame the comments aren’t usable any more, but I’m guessing your stats back up my guess that there are more like me that just read your post and skip the comments flooded by zealots indoctrinated into the new corporate religion, like the probable 20 attacks that will reply to this.
The irony that many of those commenters comment on every single article, and each comment they make is at least five paragraphs is its own source of amusement to me. That sure feels like evidence of the use of, and a decent illustration of the problem with, a certain technology to me.
Right on the money my friend. It’s almost as if they expect an actual human to digest these walls of weaponized statistics, or perhaps it’s tactics to drown out all dissenting voices in floods of gibberish while advocating for nuanced discussion.
Hey Thom, OSNews is constantly logging me out, and when I log back in, I am not taken to the article I wanted to comment on, I am taken to some crappy profile page – can you ask Claud to fix this. It’s a really sloppy user experience.
Almost nobody I know hates AI. Almost all the coders I know now realize that if you want to code in the future, you are going to use AI, and that these tools are amazingly useful. The users of Codeberg are of course entitled to do whatever they want to do. What’s codeberg again?
Savage dude.
I for one wish someone slopcoded a WordPress replacement with a more usable comments section.
You are asking a man for proving himself wrong. You can’t do that, come on.
Did they also massively vote they’ll only accept manually-written assembly code? Rejecting AI-generated code in 2026 is like rejecting compiler-generated assembly in 1996: nonsense except for a few very specific and special cases that not even 0.1% of th the projects using Codeberg will ever scratch.
LLMs are _not_ comparable to compilers. Compilers don’t rewrite your algorithm, they don’t randomly decide to ignore what you told it or hallucinate libraries or delete entire directories on a mad whim, they take the deterministic code that you carefully chose and generate machine code that implements exactly what you asked for (albeit with some play in the joints for optimization but that only affects performance and not behavior). It’s not a layer of abstraction, it’s a layer of haze.
I applaud the Codeberg community for this move. I am glad there are spaces still left for people who care about coding as a craft and about open source software as a fellowship, as the entire software world seems to be turning towards a hypertrend-following grift-driven space that values shipping more features and pointlessly recreating existing functionality with NNs over maintainability, readability, and stability.
That said, you weaken your credibility (and the credibility of other NN skeptics such as myself) by overstating your argument and misrepresenting facts. On that latter point, the ~70% figure is only about people who voted on this issue in Codeberg, it’s not an opinion poll of the general public. If you have such a poll you should cite that and not misrepresent (intentionally or unintentionally) this one. If one mainly hangs out in webdev/enterprisedev-dominated spaces or MBA-dominated spaces all of the time, you would assume that everybody is absolutely gung ho about NNs and that things like coding by hand are dead. If one mainly hangs out in spaces that aren’t that then you’d probably assume that everybody hates NNs. It can also vary broadly by age group as well as profession/interests. It’s important to not get one’s perspective warped by one’s bubbles.
I also disagree that coding is an art, I see it as a craft. Code is trying to perform a job, not communicate something. In art, the communication itself is the thing of interest, whereas in programming the communication (between the programmer and the computer) is the implementation of the real work which is whatever job the program is doing (spreadsheet, simulation, etc.).
If I were you I would just turn down the temperature and absoluteness of your rhetoric.
Now this is a good post from “the other side”, which I can get absolutely behind — even when we don’t belong to the same party (which is not necessary anyway).
One small detail to add:
`Codeberg` is a “Eingetragener Verein”, a kind like a Union Trust under German Law — and the members/stake holders voted (not the user common base). In this context, the count makes absolutely sense, especially when reading further: they main concern was the cost for their provided infrastructure. And I do very much understand and support this.
Yes, except for the Codeberg position, I agree with you. I don’t mind people being against NNs or others AI tech, because it’s a set of technology that can be problematic if pushed by techbros interests… Oh… what did I just said here… But, well, AI in general has its uses, it would be better though that we globally as a species kept a cool head about this instead of trying to drink the koolaid as promoted by the techbrosphere.
And big yes, programming is a craft, not an art, and you explained why 🙂
Also, about Codeberg, it’s the choice of its community, so be it. I wanted to move to something with more features and that doesn’t try to be a smaller version of github, Some platform are interesting, I’m looking into them (tangled, self-hosting forgejo, radicle, etc.)
> I also disagree that coding is an art, I see it as a craft. Code is trying to perform a job, not communicate something.
An important purpose of code is the communication between developers. Also, code (and its design and architecture) does reflect the programmer’s views on “how things should be done”. In that sense it’s a creative expression of oneself and thus has an artistic aspect to it. It’s also an engineering artifact. Both things can be true at the same time.
More counterbalance against the LLM hype is sorely needed to get past the “peak of inflated expectations” ASAP. Thanks, Thom, for your contribution.
The thing is, the cold, hard data shows that the hype is overblown. No, LLMs don’t make you a 10x developer. They’re currently in fact a net negative, causing bugs and incidents on an unprecedented level. I don’t understand why reports like “Acceleration Whiplash” by Faros and “The Maintainability Gap: AI Code Quality in 2026” by GitClear aren’t all over the news.
OpenAI ‘accidentally’ hacking their open source competition is just the cherry on top.
mightshade,
Most of us agree there is too much hype and I for one don’t mind countering specific instances of it. However it’s also true there’s a misunderstanding of how they are useful and a barrage of strawman arguments used to put down LLMs. Given that it’s new technology, some confusion is to be expected. It is very clear that some people have absolutely no interest in LLMs ever succeeding. I can respect this opinion, however I do think there has been a tendency to project their own views and misrepresent the value of LLMs for others who actually find them helpful as time saving assistants.
I am an old school dev and really not comfortable trusting LLMs to commit code without oversight. However I can’t deny that LLMs are proving useful at finding those pesky bugs and quickly getting custom tailored API answers that are much more laborious to do manually. LLMs are great at this. This isn’t to negate the concerns, however I think that those who use these tools effectively do have a productivity advantage.
I agree. I’ll add that the misunderstandings and strawman arguments are on the pro-AI side, too. You can see it in the comments right here. E.g. “people fear what they don’t understand” gets thrown around, as if being familiar with LLMs and being sceptical precisely because of that couldn’t possibly be a thing.
So, I’ll go into more detail why I say “the hype is overblown” and “net negative”.
The 2025 METR study showed that even seasoned developers can grossly misjudge the influence of LLMs: Participants estimated they were more productive when they actually were less. Thus, productivity claims need to be taken with a grain of salt. I prefer objective numbers.
Which brings be to “Acceleration Whiplash”. There indeed is a productivity increase on average. However, it is more than nullified by the rework that LLM-written code causes. We’re talking about a 33.7% increased task throughput per developer, but 28.7% more bugs per PR and a whopping 242.7% more incidents per PR. It’s of really bad quality, confirmed by “The Maintainability Gap”. Additionally, LLMs rewrite code more often, so it’s never “battle tested” for long. That all taken together is a net negative for software development in my book.
I should add that this is just the current-ish state of things. I’ll update my opinion when new data points in the other direction. That’s why I read these reports, after all.
Also, I’m not denying that LLMs may be useful in other areas. They are, after all, good pattern finding and completion machines.
mightshade,
I agree numbers are needed. Although even numbers from 2025 are getting dated because LLMs have been improving at a rapid pace and people are getting better at using them.
This is very dependent on how we use LLMs. Like many developers I have been stubborn to change, which includes coding by hand. I still take pride in it. Still I’ve found that LLMs are significantly more productive than trying to find and decipher API information myself. Even allowing for occasional mistakes, the LLMs are so good at finding pertinent information and compiling it into a useful form that no humans, even experienced professional coders, can compete at that task IMHO. The LLMs even generate custom tailored examples specific to our needs. This part of the job used to be a drag and now it’s a breeze.
I’m not using LLMs in a vibe coding fashion, but I do see first hand where things are going in the industry. Having already experienced cheap offshore coding companies taking away huge chunks of the domestic job opportunities, I kind of expect AI is on a trajectory to do the same thing. Traditional software jobs aren’t safe. The corporate incentive structure doesn’t care who (or what) does the work, even if we genuinely believe we are better. Just as my opinion that I could do a better job than the offshore companies didn’t protect my job from offshore farms in the past, I don’t believe people’s opinions that they can do better than LLMs will protect them going forward either.
I’m certainly not saying this is how things should be, but I do believe the cost savings will be too alluring for executives and project managers who are swayed by “cost savings” to a fault.
> Although even numbers from 2025 are getting dated because LLMs have been improving at a rapid pace and people are getting better at using them.
I know, I didn’t quote the 2025 numbers because of that. My point is that there is an effect, demonstrated in said study. “numbers from 2025 are getting dated” isn’t an argument against that, because even if current LLMs are better, the effect may still apply.
I’m not sure whether you agree or disagree with my points. I think you’re responding to my “LLMs may be useful in other areas” with a couple of examples. Thanks, I appreciate them. If I misunderstood you, feel free to correct me.
> but I do see first hand where things are going in the industry (…)
You might be right. You see, speculating isn’t really what I’m here for, so I’ll leave it at that.
mightshade,
That seems a bit handwavy to me, which is ok, but I think timely studies are important to reevaluate conclusions as the technology evolves.
I’ll leave it to you to qualify “other areas” 🙂
That’s fair enough. Although I do think people should try to prepare for the future, which is why predictions matter.
> That seems a bit handwavy to me
I don’t dismiss an effect demonstrated last year, and I trust objective numbers more than “trust be bro”. Neither is “a bit handwavy”. I don’t know what you’re referring to.
> I’ll leave it to you to qualify “other areas”
“Not programming”.
mightshade,
Well, I just think it’s inherently problematic to dismiss progress based on a study that doesn’t cover that progress. It’s stale data and using it to make conclusions today is handwavy.
In any case though, the same METR group published a 2026 update using the same methodology as before and they measured a 18% speedup for the original participants and a 4% speedup for new participants.
https://metr.org/blog/2026-02-24-uplift-update/
So given that you wanted to treat METR’s data as god when it showed a a slowdown, will you still treat it as god now that it shows a speedup?
I’m not ready/willing to let it do my whole job. But it’s undeniably useful as a programmer’s assistant. Some are loath to admit LLMs make good programming aides, but there are just some tasks that LLMs are exceptionally proficient at even if you don’t trust them to take on the whole enchilada (which I have reservations about). The ability to get useful & actionable answers on a moment’s notice is unparalleled and yes it saves a lot of time.
> I just think it’s inherently problematic to dismiss progress based on a study that doesn’t cover that progress
I don’t dismiss progress. After all, I talked about newer reports, too, didn’t I.
> the same METR group published a 2026 update
Indeed. Its new productivity findings are consistent with Faros’ ones. However, as I already wrote, that’s not my reason for referring to the 2025 METR study.
> So given that you wanted to treat METR’s data as god
Come on, dude. This is a caricature of what I wrote. I think I expressed my actual position clearly.
All of this taken together, I’m getting the impression that this conversation is devolving into trying to score gotchas against me. Thus, I don’t plan to comment here again unless someone shows up with something new on-topic to say. Do not construe any silence from me as agreement with your points. Feel free to take the last word if you require it.
I genuinely don’t get your position. METR’s latest data show productivity gains, and you were trying to use their older data to dismiss developers who claims that LLMs make them more productive. This seems highly contradictory to me and I don’t know why it’s not contradictory to you? It warrants an explanation at the very least.
Well yeah, I think I have to respond to defend myself. My intention has always been to keep things respectful even in disagreement. Why is it ok for you to override my arguments with older data and then take offense when I point out that the newer data leans in the other direction?
I don’t like ending discussions unresolved like this, but you seem to resent the fact that there’s room for me to reasonably disagree with you.
Let’s put the arguments on ice, I’d really rather end on friendly terms anyway and it would be a shame if we can’t be friendly going forwards 🙂
Interesting vote. The distinction between “AI-assisted” and “AI-generated slop” is difficult to define objectively, but it matters. Quality review processes that assess the actual output rather than how it was created seem more sustainable than blanket bans. That said, Codeberg’s community is welcome to set its own norms — it’s their platform.