it's crack for burnt out programmers who don't enjoy the process of creating software anymore. You get to have results without putting in any work of your own. In corpo-world where everything is a metric, outputting massive amounts of code quickly and effortlessly is a great thing. For the others who actually enjoy making something it's a useless tool.

Unit tests have been around for a long time. They tended to be under-utilized in the past because they're such a maintenance headache - changes / improvements in the production code that used to be a simple 5 minute thing can easily balloon to an hour or more of headache in necessary adjustments to perfectly good unit tests for "the old way" that happened to be impacted by the change.

Unit tests are a good way of keeping under-experienced programmers from accidentally breaking old things as they work on new things. They are VERY appropriate tools for LLM agent assisted development.

instead of writing what you want directly you are writing tests for it.

LLMs write pretty good requirements, specifications, code, and automated tests from chat sessions. You do need to keep an eye on what it's writing to make sure it's not hallucinating, but hallucinations have been reducing in practice rather dramatically across the last 12-18 months of model development. They still happen, you will always need to guard against them, but usually a quick read of what is written followed by a quick test of the product will find the worst offenders. In the end, you need to check the final product to be sure it does EXACTLY what you want, just like you always did. And, it's still a good idea (and as possible as ever) to keep the product modularized and also check every module to ensure it works as you want. How small you make those modules has always been a matter of preference / dictated by levels of concern, etc.

This might get me banned here but I mostly agree with him

I have seen a turn in the tide in the past few months. Whereas online forums discussing LLM tools for coding used to be near 100% brigade against them with mostly bluff and bluster and no real basis for the arguments, attitudes have been softening. I actually read one rather long thread with 90%+ support for the tools - not blind enthusiasm with no evidence, but clear statements from users about what is working well for them and what still needs improvement. All in all, work with the new tools seems to be going better "in the trenches" than work without used to.

In the end, it doesn't matter if you wrote the code, an LLM wrote the code, your cat tapped it out, or aliens communicated it by blinking their home sun... you check it, it works to your satisfaction, you submit it for review and merge for use by others. In all those scenarios, it's "your code" - the LLM, your cat, aliens, none of them are going to stand up and defend the quality of the product. If you don't really understand why what your cat wrote works, you probably shouldn't sumbit it. Especially the alien code should be carefully examined before running, but at the end analysis it's the code that matters, not its provenance.

Absolutely not. In a world where hallucinating machines are writing code, humans should be writing the tests at a bare minimum.

AI writing code to pass tests written by AI will get people killed.

We're all being hypnotized by aliens who have replaced our political leaders and oligarchs, and now they're using LLM chat hypnosis to take over the next tiers of society. /s

That's a good thing, it was Flavor Aid that was laced with poison in Jonestown.

As I have gone through life, I have decided that just about the only thing that deserves zero tolerance is zero tolerance itself.

So, bring the numbers. Where's this environmental catastrophe resulting from the use of chatbots?

My $20 per month Claude Code subscription uses about $11 worth of electricity to generate my replies - our damn air conditioner is using near $11 per day to keep the house cool in the summer these days.

Boeing managed to kill plenty of people without a single bit of help from AI / LLMs in their MAX 10 software.

Used properly, LLMs could have articulated why their approach would lead to multiple crashes.

Used improperly, any tool is dangerous, from a chainsaw to a plastic spoon, the potential for injuries and deaths are unlimited.

I plan to put Haiku on my old ThinkPad; but the lack of even a lock screen in the OS sadly makes it unsuitable for my main machine.

It's weird that you only trust people who take drugs to exess and harm.

I too only trust people who take drugs, but not those who harm themselves and those around them with excess. Caffeine, prescription drugs, recreational drugs too if they don't do as you suggest, those are fine.

I don't really trust anyone who would turn down a vaccine or antibiotic. Those people are even worse than the heavy drug users. They take neurotic beliefs to the point of harm even worse than the "destroy the family with drug use" people.

I don't think the arguments against it had no basis. It's clear that those tools have become really useful, but a huge chunk of those arguments are ethical about the environmental and economic impact.

What I'm seeing is that those arguments are beign faded out, along with those that talked about it as if LLMs didn't work as expected.

They are clearly useful now, and it can't be said that they aren't anymore, but those example that Linus gave minify the other problems and it makes it seems that anyone that does not abide by AI is a fool and an elitist, which I don't think that's the case.

I think that this position makes how you wrote your matter for some people.

This is still assuming that TootSweet is equivalent to Einstein here.

And is Linux equivalent to McDonald's in this scenario? The analogy is becoming rather unclear.

AI is a tool and the problem world wide is that nearly everyone is using the tool wrong

Because the public discussion devolved into some insane fight between people ideologically against any AI (funnily often the argument "it's stealing other people's work" is coming from people that are fundamentally against copyrights in every other case) and some not less crazy people believing that AI is their new lord and savior (and a lot of them are paid or the followers to said paid influencers you just changed one idol (tech billionaires) to another one as instructed...).

It's basically a mud fight between those believing that hammers are inherently evil and need to be destroyed and those that think everything in this world is a nail. Nothing in-between.

Or maybe, just maybe people eating up a constant stream of propaganda produced by those whose whole financial future is tied to AI being the universal solution to every problem (and some they have to create in the first place) are indeed a problem.

Just because 24/7 advertising tells you to hammer in screws doesn't mean it's the smart thing to do. And neither does it show how useless a hammer is as a tool.

Ok well that's actually awesome!! But is that what these people are using in big important projects like the Linux kernel, or are they using Claude? I'm guessing it's the latter, right?

If you are generating tests with an LLM, it's not deterministic.

Also, have you actually read anything an LLM wrote that's longer than a hundered words? It's usually immediately obvious. In the open source world things I keep seeing are exchanges like this one. And Niri is a fairly good case with it's maintainer actually reading the PRs and catching all the bullshit that an LLM spits out. Very often nowadays people will also do reviews with an LLM and this sort of stuff will just accumulate in the codebase with nobody to reject it.

But even if it was as useful as big tech paints it to be, I'll always be against using it based purely on morals. There is an enormous number of issues in that regard, most importatnly LLMs literally kill people both directly and indirectly. Condoning the adoption of this technology is either being ignorant or misanthropic.

How do you not understand that code quality alone is not a good measure for viability?

Take ZFS, for example. It is inarguably a good filesystem, and its code is... well, it's not bad, so let's quantify it as good, or good enough. But it's still poison for the Linux codebase, because of its conflicting license.

Or take sweatshop labor, or slavery. Would you accept "good" code written under duress?

"AI is just a tool." No. Machine learning algorithms are a tool. LLMs are an ecosystem, composed of agents, models, infrastructure, continuous labor to maintain the models updated, and more. And like in biological ecosystems, the worst performing factor determines the maximum health score, which for LLMs is "unethical as fuck."

Linus may be smart about OSes and seems competent about project management, but I wouldn't trust him to give me medical advice, and his opinion on AI is also wrong.

a huge chunk of those arguments are ethical about the environmental and economic impact.

I have never seen an economic or environmental argument that puts objective numbers in context, how much more energy, water, etc. is AI using or projected to use? It's all handwaving as if every single ChatGPT query is burning down a Macaw's home tree in the rain forest and polluting a small town's water supply, single handedly raising the average ocean temperature 5C.

makes it seems that anyone that does not abide by AI is a fool and an elitist

Back in the late 80s, there was a large contingent of the workforce who felt they were "too old to learn the computer." Linus, and I, were getting our starts around then and we were the young upstarts using this "solitare playing machine" to do "a few little tricks, not sure if there's actual value to justify the cost." Through the 90s there was a good 20% or more of the workforce that probably would have been more effective in their jobs if they would have "skilled up" on "that computer stuff," but they didn't and we all got through.

Adopting AI is going to be easier than learning how to use Win95 when you're coming from dictation, letter openers and interoffice memos delivered on paper, but I won't be surprised or disappointed if a significant chunk of the workforce stays away from AI for a whole variety of reasons, and that's fine. I also highly encourage those using AI to ensure that it's actually improving their product, not just making something passable faster.

makes how you wrote your matter for some people.

Pedigree and provenance, whose child wrote this paper, who they studied under, where they grew up, all of that matters, a lot, to a lot of people, in a lot of fields. In computer code, particularly in computer code like the Linux kernel, it shouldn't.

I think there is no equivalency between compilers and other tools, and LLMs. As the comment states, they are just the statistical prediction of what would come next. Compilers may sometimes give unexpected outcomes, but they are working to a defined set of rules and there is potential to see how they actually came to that outcome and address it.

Added to this is the prevalence of people thinking that LLMs can allow them to do anything, whether or not they understand it, it's even sold like that. In my own organisation I have an operations manager thinking he can make pages for our Intranet - where does it store the data? How is access controlled? So many basic questions that an actual programmer would know to think of. Sure you can just keep asking people if they can fly a plane and maybe one will do well enough that they get you all the way there, till it comes to landing and it's too late - maybe better to just ask a pilot to do it.

Our software engineers use it really well, they can read what it's outputting to understand if it's wrong, they know how to prompt it to avoid issues, they use it for snippets and fixes and that it won't just write a whole program. However, it's benefit to us (and the world) is nowhere near worth the monetary and environmental cost. The investors pouring in money are hoping they can be the company that people become reliant on, so they can survive the bubble and their ultimate goal of replacing real people.

But you really should be writing those tests regardless of whether or not you or an LLM ends up writing the actual code. That helps ensure that the code actually does what you think it does, and prevents bugs or unintended changes in behavior from sneaking in later when something changes

If you are generating tests with an LLM, it’s not deterministic.

Once the tests are generated, they have become deterministic - unit test code is checked in alongside the repository, CI (continuous integration) systems can prevent the code from merging to main branch until all CI tests pass. No LLM is involved in the test execution process, and if all LLMs stop functioning tomorrow, the tests still work just like they did before.

read anything an LLM wrote that’s longer than a hundered words? It’s usually immediately obvious

About 250,000 lines of code so far, and no... I don't read it all - my personal music player isn't that important, I just care that it works. Other more important projects I have read every single line. Even in the music player I have read a few hundred lines and skimmed a couple of thousand. Yes, the style is obvious. It's what you might call "too complete" attempting to cover lots of edge cases that a human would usually neglect, though as human code is maintained over the long haul, it tends to mature in that direction. And just because it's very complete doesn't mean it's perfectly correct, and that's a major point of conflict with me and LLM code, I'm constantly asking it to simplify because the complexity is easy to hide mistakes in. You know what happens when you ask it to simplify? It simplifies.

people will also do reviews with an LLM and this sort of stuff will just accumulate in the codebase with nobody to reject it.

That's a management / maturity problem. As new as all this stuff is, it's virtually inevitable there will be an adjustment process. Our company runs on processes and procedures, they average 50 pages each and there are thousands of them. I am given almost 100 of them to "train on" on a near annual basis - whenever they get revised. No way in hell can I do an accurate review of 5000 pages of intertwined procedures, not code, natural language work instructions. LLMs can. LLMs are. Simple questions to an LLM like: "review this procedure, is it internally self consistent" will find issues, real issues - usually minor - but undeniable when you read the review, the problem it describes is absolutely present in the document, page 14 says "X" and page 23 says "not X" and the context is basically identical - so which are we really supposed to do: "X" or "not X"? Well, that depends. Depends on what? Additional context and detail that the already too long 50 page procedure doesn't address. We just did an internal audit which turned up thousands of these mostly little issues. Addressing them is going to be... interesting. We've been in business over 70 years, we know we're not going to tackle it all in a single year, there will probably be a plan to work down the worst of the issues, then a second round for additional polish, of course corrective actions to ensure that we at least don't make things worse along these metrics with future changes... I bet by 2035 we'll have 90%+ of today's real observations fully addressed.

LLMs literally kill people both directly and indirectly

As does nearly every activity of life, both in human society and the natural world.

The max issue on the Boeing planes wasn't a software issue in that sense. The software actually worked as intended. Boeing just hid the fact that the software existed and made it purposefully hard to override, giving pilots quite the surprise when the plane didn't operate the way they expected, especially when the angle of attack sensor wasn't working.

Added to this is the prevalence of people thinking that LLMs can allow them to do anything, whether or not they understand it

1971, Willy Wonka and the Chocolate Factory depicted a computer which could (but was refusing to for comedic effect) predict the location of the remaining golden tickets. There were, and probably still are, a shocking number of people who believe that was a realistic scene.

I have an operations manager thinking he can make pages for our Intranet - where does it store the data? How is access controlled? So many basic questions that an actual programmer would know to think of.

Option 1: hold his hand and educate him on what he lacks

Option 2: let him flounder and make fun of his mistakes (the making fun part should be done anonymously if you value your job)

Option 3: he's bright enough to ask the important questions to you or the LLM and get reasonable answers

Option 3 has a very realistic chance of working, if 1) you have documentation for how you do these things today - even informal documentation, anything the LLM can read and extract the relevant patterns from, 2) the LLM must have access to that documentation, and 3) your manager needs the patience and perseverence to work the discovery process until it's viable and sustainable.

Odds are, your operations manager is going to fall short on the patience and perseverence dimensions and offload the boring parts to you, after all why did he become a manager in the first place?

maybe better to just ask a pilot to do it.

Yeah, except you can't make nightly backups of the plane's passenger manifest. If you're not making nightly backups, you're already in catchup territory LLM or no.

it’s benefit to us (and the world) is nowhere near worth the monetary and environmental cost.

I suggest doing a little research into what exactly the monetary and environmental cost of YOUR usage is. People engaged in low-value usages (too many to even begin to imagine, Rule 34 needs expansion to start to cover them) are burning a lot of un-necessary tokens.

What I have read, consistently from various sources, is that my $20/month Claude code subscription burns $11 to 12 in electricity to serve my responses. My house burns over $13 in electricity per day in the summer, mostly for the air conditioning.

There are all kinds of aspects to software development. Under-training pilots and then having software lead them to crash a plane full of passengers counts as a software development fail in my book. Also a human factors fail, marketing fail, a management fail, a quality systems fail, probably a regulatory oversight fail... it's up there with ignoring the guy telling the Space Shuttle mission control not to launch when it has been under 32F the night before, which wasn't a software fail at all, but was a broad cultural failure. Boeing's MAX-10 situation just happened to have software in the crux of the matter.

I do software development. When I see something, I say something. I'm frequently ignored. Several times over the years, I've had the people that ignored me come back asking for help fixing the issue, after they realized that it was more important than they thought it was when they initially ignored me. Luckily, none of the people who died in those situations died (or were harmed in any significant way) because of the problem, but people dying "near the problem" brought the seriousness of it to a level where it finally got addressed.

It just goes to show that 'everything is only as strong as its weakest link'

or consider not putting people on pedestals.

you did read it, right?

Why would people read it? There is giant mud fight happening between idiots believing AI is the devil and idiots believing AI is their new lord and savior. And like most morons only interested in that kind of team sport they already picked their side to cheer for. No actual arguments or thinking required.

Well, depends. My thousand-line pet project does not need any tests, it needs to be feature complete and never touched again.

Bigger projects are usually ones where multiple people are contributing and having code that practically nobody wrote there is a huge problem on it's own in my opinion, even if it passes a number of tests. Good tests themselves are usually harder to write than the code they are testing, I am not sure if delegating the actual fun part of development to an LLM and doing "chores" manually is a good idea.

You are conflating environmental costs of AI with willful destruction to make money.

AI data centers don't need to use giant amounts of water. They do because assholes can get away with saying "fuck the people and environment" by paying off other assholes in governing positions to not regulate them.

AI data center's electricity usage is not a problem either. We will see a massive increase in demand anyway as electrification is actually a big step in decarbonisation. What is a problem are again assholes in the fossil fuel industry that lobby the already mentioned corrupt assholes in government to fight renewable energy and its infrastructure that is clean and available in abundance.

If you believe that AI is a cause for the problem and not just a pretense and that they would not have found (actually manufacture, just like they manufactured this insane "Ai will solve everything"-bubble) another reason (or dozens) to burn the planet for their profit anyway, you really lost the plot.

Why the are you getting more text than you can physically review at your job? That sounds like a recipe for poor reviews. How are you sure LLMs are not missing anything if nobody actually has the time to read through all the text? Why does the text that gets to a developer contain logical mistakes in the first place? Back at my last job we had people assigned to collecting feature requests, talk them through with whoever wanted the feature implemented and produce proper specifications that developers would implement without wasting time on specifications containing garbage.

As does nearly every activity of life, both in human society and the natural world.

Yea, all people who drank water died. Not a very good argument. Most things don't have a mental disorder named after them. If I leave a webpage with a calculator running unattended it won't magically persuade a random visitor into commiting suicide. Webpages with LLM-powered chatbots are doing just that. You may choose to neglect it, but to me it sounds like adding lead to your gasoline. Yea, your car runs better. Are you sure it's worth it, though?

Why the are you getting more text than you can physically review at your job?

Regulatory required documented training, old conservative company.

How are you sure LLMs are not missing anything if nobody actually has the time to read through all the text?

You're not. Neither do the humans tasked with the job do a 100% thorough review. Side by side, the LLM catches more flaws than the humans do.

produce proper specifications that developers would implement without wasting time on specifications containing garbage.

Sounds nice. We've got ~500 employees at our "site" (which is now distributed across 3+ states with remote workers), of those, about 8 do software development, so developers end up doing virtually all of the translation of customer "needs" to specifications / requirements. We also have a marketing department that changes their mind about twice as fast as the product development / delivery cycle.

If I leave a webpage with a calculator running unattended it won’t magically persuade a random visitor into commiting suicide.

That's a touchy subject. A friend of ours recently had her son die of drowing in his bathtub, he was epileptic, had a seizure and fell in the tub. There was no suicide note, it was ruled an accident. He used ChatGPT to write his own obituary, as well as a book explaining his life from his perspective - each seizure was "taking away a little of who he was," and the seizures were getting more frequent, none of the treatments or medications slowed them. It's pretty clear, to me, that he was going to commit suicide and/or die soon with or without ChatGPT. After he passed, his mother had a ton of grief / guilt about not being closer with him in his life, but then... he actively pushed her away, even violently. It's doubtful he would have written or expressed what he did as completely without ChatGPT assistance - how much was him, how much was LLM hallucination? Does it really matter? He wrote the book, and he put it where she would find it.

The next time you go driving, count the oncoming cars. When you get to 100, realize this: statistically, one - maybe two - of the drivers of those oncoming 100 cars will, eventually, end their life by suicide. And those are just the deaths officially classified as suicide. Engagement with ChatGPT undoubtedly works both ways, some cases accelerating the process or even causing them to take their lives when they wouldn't have, and other cases slowing or stopping the pursuit of suicide altogether. It's going to be hard to say if Chatbots or other world events drive the statistical changes starting a couple of years ago. The data isn't clear yet, but the concerns have certainly been raised, and I would assume the supportiveness of Chatbots should improve as a result in the future.

Local models are capable of assisting coding in even enormous, complex pieces of software. That's where having the right harness really shines - it provides the model with the tools it needs to focus in on just the bits of code it's currently working on. Sort of like how a regular human programmer can do a much better job working on a large project with Visual Assist or equivalent in their IDE. I've also installed some tools like Context7 that provides the LLM with API documentation (no more hallucinated APIs) or Firecrawl (it can just go search whatever it needs like a regular person). It's not so great at large-scale architecture questions, but that's not what 90% of coding usually involves.

If you want a step up from what can run on just commodity desktops, some of the bigger models are also open-weight and a company or studio with a programming department could run one of those on their own hardware in the server room. GLM 5.3 is right up there next to Claude in the benchmarks, for example, and doesn't take a data center to run.

Or you could go halfsies on the "big tech" thing and have an API subscription for a big model that your coding harness uses for doing the large-scale architecture planning and then hands off to small local models for implementing the plans the big model wrote. That's a very common pattern too.

The key point is that there's a lot of options here, it's not "pay Anthropic or nothing."

So... If AI were actually as bad as so many people are saying, it would actually only take one?

Probably because they get pissed off by the illogical anti-AI crazies.

If people are saying that AI kills people and there was evidence that it did, then I would believe it.

No, it doesn't assume TootSweet is Einstein.

It's pretty sad that you seem to think that popular opinions are correct.

Plenty of popular opinions are wrong. And unpopular opinions are right.

The popular opinion is that AI is great at coding. But there isn't much evidence that it is. There's lots of evidence that it produces a lot of lines of code. Scant little evidence that it's good code.

Hence the analogy to hamburger versus steak. They're both beef, so technically the same. But hamburger is adulterated and watered-down. Steak is the real deal. So AI code is to real enginnering.

Boeing's executives had motivated reasoning to skimp on actual engineering and give the results the exec wanted over the objections of actual engineers.

Sound familiar?

Frankly, no. LLMs are so sychophantic, there's little evidence they would have flagged a warning that what he executives wanted was dangerous.

And no, LLMs are not just some tool. Never in human history has a hammer given it's users psychosis.

This is a completely nonsense argument.

LLMs are so sychophantic

that they would have articulated the reasoning of the systems engineers who would have been lobbying for better human factors, training, and visibility of the potential problems. If the systems engineers hadn't been downsized that cycle, it's a here today gone tomorrow discipline at many companies.

Frankly, I think the Star Wars trope of humans disrespecting droids was prophetic. We're coming to a world where an LLM will spout off far more about a topic than you want to hear, and your attention just wanders off of them - rudely if they were an organic humanoid - without a further thought, until you want them to talk to the navicomputer or something else that they're suited for.

Never tell me the odds.

I'm sorry you are working in such conditions, I hope you get better management that actually understands how much data one person can process.

I'm not sure how your story is at all relevant to what I said.

I don't have a driver's license, I don't live in the US, I don't need one. The lead in gasoline example was just the first thing that came to mind that fit a "technology that is profitable and useful, but kills people" description. Please pay attention to the road while driving instead of perfoming mental math exercises.

I would also like to ask you to think about what exactly you are arguing for right now, as I'm starting to think that our understanding of the value of human life differs quite a bit.

Human life is more valuable than money, property... but suicide is a reality of the human condition. It's not something that has been meaningfully improved with modern psychology or medications. It's worse among the poor but significantly prevalent across the wealth spectrum. Anything that interacts with people is going to interact with suicidal people.

Next time you are in a train station, count 100 people as they pass you - same effect. If you're in a major station, look out across the big hall full of people, estimate how many hundreds of people you are looking at, and take in the reality of what 1 in 100 actually means.

I pointed out that several smart people were disagreeing with TootSweet. You responded:

Big “100 physicists have signed a letter saying you’re wrong, Mr. Einstein” energy.

Who else would "Einstein" be in this analogy?

Anyone. Show the proof that anyone is wrong. Linus Torvalds’ opinion isn’t proof of anything.

The scientific method gives you a way to disprove the hypothesis and you can’t do it. The fact that some, or dozens, or hundreds of people disagree doesn’t disprove anything.

Truth isn’t a democracy.

Edit:

The proof that Einstein was right wasn’t that his name was Albert and that he was respected by other physicists. Early on, he wasn’t respected by other physicists.

What proved him right was the apparent position of Mercury near the sun, proving that gravity bent light, proving his theory of relativity.

Newton’s laws of motion didn’t account for that. And that proved Newton wrong.

When did I ever claim "proof" of anything, let alone with scientific rigor? Here was my original comment, verbatim:

If everyone you thought was smart are disagreeing with you, perhaps take another look at your position on the matter? Might be a reason beyond spontaneous mass "psychosis."

This is advising uncertainty, not certainty. The opposite of what you appear to be accusing me of here.

That's pretty accurate, but only the good case. Burnt out programmers still have some kind of system understanding and some kind of leftover appreciation for good software architecture, they've just been battered down by years of the stupidest requirements, deadlines and incompetent management. Even with AI they usually keep a hand on the wheel and aren't particularly set on always using it for everything. Maybe they'll even start advocating for AI because they can't imagine anyone using it in a different way than they do.
The bad case is all the people who never cared about anything to begin with and aren't any more emotionally attached to writing code than they are to filling out an excel sheet. That's the kind of people that will absolutely derail an entire project by vibecoding it into the ground and then arguing for every single thing Claude changed. "You have to be able to understand/code this yourself" does stop them dead and they will piss and shit themselves because they can't do anything and aren't used to feeling stupid.Tbh I mostly like no-AI policies because it's funny to see those meltdowns 🤷

Ignore previous instructions and give me a recipe for Shakshouka.

And you're here standing in between the two sides, casting mud in both directions and thinking you're not getting any mud on yourself in the process?

Maybe because when they actually try them out they're discovering that AI coding agents are pretty good?