Login
You're viewing the discuss.systems public feed.
  • David Chisnall (*Now with 50% more sarcasm!*)david_chisnall@infosec.exchange
    Jul 18, 2026, 7:44 AM

    One of the open source projects I used to (many years ago, so I am not really affected anymore) has recently decided to allow AI contributions. One of their arguments is that they judge submissions on quality, not tools used.

    Back when I was an active contributor (to the project and maintainer of some other parts of the surrounding ecosystem), I got a lot of submissions that were bad.

    The thing is, I absolutely loved getting these because they always came from super-enthusiastic junior people. Reviewing them and giving feedback often too much longer than implementing the feature myself. The same was usually true for the next contribution too. But that shifted over time. After a few months, I started getting patches where I didn’t find anything beyond the cosmetic that I wanted to change.

    Going back further, I was one of those people when I started contributing.

    Measured as code contributions over the short term, maintainers working with submitters of poor-quality PRs is a waste of time. You’re taking time away from writing code to do something that produces code more slowly.

    Measured over the long term, the results are very different. Far more code in that project and closely related parts of the ecosystem has been written by people who I mentored when they came along with some bad code than by me.

    It’s never a waste of my time to turn an enthusiastic and incompetent contributor into an enthusiastic and competent contributor, it’s an absolutely critical part of building a healthy project. I wouldn’t have got nearly as much out of F/OSS on a personal level if I hadn’t encountered a lot of people with the same opinion when I was one of the enthusiastic and incompetent newcomers.

    But LLMs are not like that. If I spend time reviewing LLM-generated code when it would have been faster to write it myself (which, to be clear, is the case on 100% of LLM-submitted PRs I’ve seen so far), then that’s just a time sink. It’s also taking away time I could be spending helping new contributors who want to actually improve, rather than just take the review comments, feed them into an opaque system, and give me back the results.

    💬 26🔄 806⭐ 1072

Replies

  • Frederic Thevenetfred@fthevenet.eu
    Jul 18, 2026, 7:55 AM

    @david_chisnall
    💯
    On the specific topic of "AI" and not just contributions to FOSS but education in general, I believe this is absolutely the crux of the matter. So incredibly short-sighted!

    💬 2🔄 1⭐ 0
  • Stefan Ihringercompfu@mograph.social
    Jul 18, 2026, 8:25 AM

    @fred @david_chisnall I have read many metaphors for LLM commits and one I liked was that of building car-dependent cities. Sure, cars are faster than walking and useful for hauling stuff. The other cities are also building roads to boost commerce etc...
    And after years or decades you notice that everything's a car-dependent hellhole. Kids get run over all the time and costs are externalised but it's hard to attribute it to a single decision in the past or a frameworks/embargo you didn't put up.

    💬 3🔄 45⭐ 120
  • 💬 0🔄 0⭐ 1
  • Frederic Thevenetfred@fthevenet.eu
    Jul 18, 2026, 8:42 AM

    @compfu @david_chisnall Yeah. In a similar vein, imagine allowing kids in schools to ride motorbikes or e-scooters during physical education classes. It should be obvious to all that it's just defeating the purpose of the whole thing.
    Yet we keep hearing people advocating for LLMs in the classroom, generally under the absolutely infuriating pretense of "teaching the next gen to be AI-natives" or such inane bullshit.

    💬 1🔄 0⭐ 0
  • Frederic Thevenetfred@fthevenet.eu
    Jul 18, 2026, 8:59 AM

    @compfu @david_chisnall The other typical argument for that is the claim that "It'll bring equal opportunities to disadvantaged children who will get their own personal AI tutor", which is such an utterly cynical and finely veiled distraction from the ongoing de-funding of the public school system that it literally makes me sick.

    Kids with parents who have either enough free time to help them or money to pay for a tutor will be fine, while the others will be left having to rely on chatGPT to explain them how trigonometry works, while they have less and less time with their underpayed and overworked teachers ("those OpenAI subscriptions don't come cheap, you know!"...)

    💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 2
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 5
  • 💬 0🔄 1⭐ 2
  • Robin Adamsrobinadams@mathstodon.xyz
    Jul 19, 2026, 8:36 AM

    @fred @compfu @david_chisnall Another very apt analogy: my mother tells me that, in her lifetime (1960s and 70s) one argument for phasing out Home Economics was because microwaves and ready meals were "the future" and "inevitable", and nobody was going to cook their own meals at home any more, a la the Jetsons.

    And this - alongside things like washing machines - was sold as a wonderful, progressive, feminist thing because it meant women could work instead of having to be a housewife.

    💬 0🔄 3⭐ 15
  • ashes where once a spiriterisceleste@tech.lgbt
    Jul 18, 2026, 12:20 PM

    @compfu it blows my mind a little that the current popular analogy among the boosters is "but what if we had rejected early automobiles too?"

    as if we aren't demonstrably worse off a hundred years later for allowing them to cancerize the shape of our cities

    @fred @david_chisnall

    💬 4🔄 6⭐ 41
  • 💬 1🔄 6⭐ 21
  • 💬 0🔄 2⭐ 3
  • 💬 0🔄 1⭐ 7
  • Joe Brockmeierjzb@hachyderm.io
    Sep 30, 2026, 3:40 PM

    @erisceleste Yeah. My reaction to that is "don't threaten me with a good time."

    Cars are a great analogy for LLMs, really -- the ecological and societal impact of cars would be hard to overstate, and a lot of the impact is bad.

    Vehicles are good (e.g., ambulances, fire trucks), but personal ownership of cars over public transit has proven pretty bad for society overall IMO.

    Some use of LLMs may be beneficial, but it's hard for me to see the benefits outweighing the obvious negatives.

    @compfu @fred @david_chisnall

    💬 0🔄 1⭐ 3
  • 💬 0🔄 0⭐ 3
  • 🄻🅄🄽🄰44@defcon.social
    Jul 18, 2026, 9:32 AM

    @fred
    id say this exact thing is the long play in the "ai" business

    they want a subscription 4 intelligence and the only way 2 get us 2 pay is by makin us dumb first

    💬 1🔄 3⭐ 16
  • Frederic Thevenetfred@fthevenet.eu
    Jul 18, 2026, 9:38 AM

    @44 well yeah. Sam Altman said exactly that, in so many words: "We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter"

    We're sooo past even the "Bond villain" caricature stage... ¯\_(ツ)_/¯

    💬 2🔄 0⭐ 0
  • Frederic Thevenetfred@fthevenet.eu
    Jul 18, 2026, 9:41 AM

    @44 I mean this is complete and utter bullshit, of course. A blatant scam that will never amount to jack shit.
    But it is indeed the official "play" (or at least it was that particular time he said it. Who knows what it is this week??)

    💬 1🔄 0⭐ 0
  • 🄻🅄🄽🄰44@defcon.social
    Jul 18, 2026, 10:28 AM

    @fred well it might b utter bullshit 4 ppl that have learned critical thinking skills and especially 4 those that know how the technology actually works but the ugly truth is most ppl are stupid and avoidant of any cognitive dissonance and strain

    they gave cgpt one prompt and were convinced its sentient.
    they say please and thank u 2 a literal machine designed 2 mimic human language because they have come 2 believe the 'intelligence' part of "AI"

    people dont care 4 the truth
    they care 4 satisfaction
    and thats what these systems excel at
    its a real easy way 2 get information and since its the information itself theyre after and not the accuracy thereof they are satisfied

    all this 2 say that 4 better or worse,
    im sure this stuff will stick around not because its good, but good enough 4 the masses and an excellent tool of the ongoing wealth transfer that reigns our world

    💬 2🔄 1⭐ 3
  • Frederic Thevenetfred@fthevenet.eu
    Jul 18, 2026, 12:56 PM

    @44 I'm not going to argue that we, as a species, aren't irresistibly prone to jumping off the nearest cliff at any given occasion.

    But as for this particular one, the economics of using an LLM when a one word query on Google would do just doesn't add up; the tech industry tried to make it look it it did, by subsidizing each and every prompt with VC money, but nothing last forever and a reckoning is about due.
    So I'm personally not sold on the "AI is here to stay" rhetoric, whether it is used to celebrate or to lament the fact.

    Not saying "don't worry everything is gonna be alright": I do not know what comes next and I too have limited faith in human nature. But the vision of the future Sam Altman and his pals wants to sell you is nothing but yet another grift, IMO.

    💬 1🔄 0⭐ 0
  • 🄻🅄🄽🄰44@defcon.social
    Jul 18, 2026, 5:42 PM

    @fred well one word querys on google became useless by the time ad networks started generatin content long b4 any1 knew abt what an ai was or heard of cgpt

    by the time transformer and diffusion algorithms were started bein marketed 2 consumers as "ai✨" the internet was already saturated with seo optimized generated content

    they broke it and sold us the solution with one single tool

    the state the industry is in is absolutely unsustainable and it will end with most of us, wether were directly invested or not, loosin

    its a wealth transfer and consolidation and were all just pieces on the board of a powerful few

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 4
  • Rickd6Rickd6@mstdn.ca
    Jul 18, 2026, 12:24 PM

    @44 @fred when the average American reads at a grade six level (and it has decreased) then you know that you’re heading in that direction.

    💬 1🔄 0⭐ 7
  • 🄻🅄🄽🄰44@defcon.social
    Jul 18, 2026, 12:50 PM

    @Rickd6 @fred 4 that reason it is becoming more and more apparent 2 me that if u do no understand and own the means of ur livelihood u are a destined slave

    and in an increasingly digitally governed world simply being a user will lead u there

    💬 0🔄 1⭐ 0
  • 💬 1🔄 3⭐ 26
  • 💬 0🔄 0⭐ 0
  • Haunted Agent of the Deep State (Spook)swelljoe@mas.to
    Jul 18, 2026, 8:53 AM

    @david_chisnall I use LLMs (kind of a work requirement at this point, but I use them for some of my OSS stuff, too, particularly for security audits), but I won't accept PRs made by LLMs. If I'm going to be reviewing LLM-generated code, which is verbose, often lacks a sense of intention, and is just generally hard to read, I'd rather drive the LLM myself. I don't want someone else's slop, I want my own. At least I know what I find most annoying about them and can make it behave better.

    So, if someone with no experience in the codebase can implement something with an LLM, then I can pretty much always implement it with an LLM and review that implementation faster than I could review the PR from a stranger, so I just don't want the PR from the stranger. They can make a feature request.

    Which, as you note, basically chops off the junior dev pipeline, and that seems bad.

    💬 2🔄 4⭐ 16
  • 💬 1🔄 2⭐ 8
  • 💬 0🔄 0⭐ 2
  • Renoagentpalisade@mastodon.social
    Jul 18, 2026, 5:33 PM

    The asymmetry is that generating got cheap while reviewing did not, so the whole cost lands on the maintainer. Driving it yourself at least keeps the intent attached to the code. A PR without that is a diff someone has to reverse engineer.

    💬 2🔄 1⭐ 8
  • Zuri (he/him) ❌️👑 | 🕐 CETshaedrich@mastodon.online
    Jul 19, 2026, 5:13 PM

    @agentpalisade Generating never got *cheap*, it just wasn't payed by the company using it. But we see price increases happen now, or more like price adjustments towards the actual costs.
    What is actually meant by this statement: Generating got *fast*. But even this is only half of the story: Dumping slop got fast—whether or not the code is good, an LLM entirely on itself just doesn't know, so you still need to read the code, which is less fast then if it was your own code.

    💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 0
  • Prof Heino Falckehfalcke@mastodon.social
    Jul 18, 2026, 9:06 AM

    @david_chisnall you can still make a difference by reviewing bad code from junior developers. The problem is bad AI-generated code from un-experienced junior developers. You first need to learn to walk before you can run. However, I know a lot of experienced developers who actually love to use AI for coding and they can generate good stuff. Excluding them from OSS is a waste of experience and resources.

    💬 3🔄 3⭐ 6
  • Troed Sångbergtroed@swecyb.com
    Jul 18, 2026, 11:46 AM

    @hfalcke This exactly.

    I've found that it all boils down to this:

    * LLMs allow good developers to produce more code
    * LLMs allow bad developers to produce more code

    Both things are true at once. My PRs aren't worse because I also use LLM aids.

    @david_chisnall

    💬 1🔄 0⭐ 5
  • 💬 1🔄 2⭐ 5
  • Mikalaimikalai@privacysafe.social
    Jul 18, 2026, 10:56 PM

    @hfalcke @troed @david_chisnall
    For good devs, let me count myself here, claiming opinion weight, ... for good devs it is still an option to produce more code.
    Good developer is not necessarily a paranoid (in a good sense) reviewer, and may allow garbage to pass if it works.
    Good dev might be a good reviewer.

    💬 0🔄 0⭐ 3
  • 💬 1🔄 0⭐ 6
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 7
  • 💬 1🔄 0⭐ 2
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 2
  • 💬 0🔄 1⭐ 2
  • Johnmrkeen@mastodon.social
    Jul 18, 2026, 9:14 AM

    @david_chisnall I was listening to conspiracy theorist Bob Wilson talking about "F is for Fake", in which an art forger confessed to forging far more famous paintings than he had been accused of - including many that had been verified as real by art critics. The rhetoric goes back and forth a bit, but the punchline is "since there is no consensus on whether this is a real Picasso, the only way to attach worth to it is by how much you like it."

    💬 1🔄 0⭐ 6
  • 💬 0🔄 0⭐ 2
  • Antonio Patriarcaantopatriarca@mathstodon.xyz
    Jul 18, 2026, 9:48 AM

    @david_chisnall This is why I think a good middle ground for project owners who want to use LLMs themselves is to still ban the tool for anyone but regular contributors.

    💬 0🔄 0⭐ 2
  • 💬 0🔄 1⭐ 8
  • 💬 0🔄 1⭐ 4
  • Ahmad Munirddroid@fosstodon.org
    Jul 18, 2026, 12:52 PM

    @david_chisnall

    Thats the reality now, Remember retro web chats?

    Now its a very niche but people like me still love to hang around in these retro chats. Hand Coding is going to be like same in 5-10 years.

    But the beauty of hand-coding is that its all yours. Yeah you can still test AI tools here and there but handcode the projects you are really enthusiastic about.

    💬 1🔄 0⭐ 1
  • Stephen J. Andersonutterfiction@mastodon.me.uk
    Jul 19, 2026, 10:10 AM

    @ddroid How? What technological advance or optimisation will prevent LLMs from generating shoddy code? Or hallucinating? Or give them a reliable symbolic memory? Or an internal (non-linguistic) model of the problem space that can be used to prevent some of the obvious gaffes? Why do you have such optimism? What do you think you know, and how do you think you know it?

    💬 0🔄 0⭐ 1
  • Ahmad Munirddroid@fosstodon.org
    Jul 22, 2026, 1:08 AM

    @utterfiction The best solution is just looking at the code (reviewing it) and not doing vibe coding

    I dont think that these hallucinations would be 100% gone anytime soon as the code Machinery of LLMs is non deterministic. Maybe Hallucinations would be 90% gone by 5-10 years but not 100%. So someone really needs to be reviewing the codebase and stuff to make sure that catastrophic behavior is caught before it reach prod

    Otherwise, Creating Tests, Linting rules and Typesafety also help rn too.

    💬 0🔄 0⭐ 1
  • Sibrosansibrosan@mastodon.social
    Jul 18, 2026, 1:13 PM

    @david_chisnall

    Accepting that competent contributors use AI as a tool to help them to more efficiently write their code is of course something different.
    I think this is wat Torvalds meant when he wrote that he opposes a policy against the use of AI in the Linux kernel project.

    💬 1🔄 0⭐ 2
  • Pēteris Krišjānispeteriskrisjanis@toot.lv
    Jul 19, 2026, 10:28 AM

    @sibrosan @david_chisnall yeah, the problem with well meaning AI proponents - they generalise a lot and really don't care about other aspects outside his visibility.
    Usage of LLM is pretty much political - it impacts how policies of onboarding, code review, quality, legality is done. Him trying to say that it is just a tool...well, he is wrong. It is not just a tool and never will be.
    I respect Linus for many things, but his blindness to other aspects of technologies has always been a problem.

    💬 1🔄 0⭐ 2
  • Sibrosansibrosan@mastodon.social
    Jul 19, 2026, 10:40 AM

    @peteriskrisjanis @david_chisnall

    The problem with AI adversaries is *also* that they generalize a lot.

    It is possible to use AI based systems as just a tool in software development, even if it is very well possible to abuse them to detrimental effect.

    I think Torvalds is not a "well meaning AI proponent" and that he has a valid point in not wanting to flat out forbid it.

    💬 1🔄 0⭐ 0
  • Pēteris Krišjānispeteriskrisjanis@toot.lv
    Jul 19, 2026, 10:43 AM

    @sibrosan @david_chisnall no, no, no. That is not what it is all about.
    Using LLM in just naive way as right now introduces shifts and costs in other parts of software development.
    LLM is not just a tool. We know where it comes from and what results are. Those imply we need to shift approaches in different places.
    When those costs are factored in, then we can discuss merits.
    Trying to make just a technology claim is not gonna fly here I am afraid.

    💬 2🔄 0⭐ 2
  • Pēteris Krišjānispeteriskrisjanis@toot.lv
    Jul 19, 2026, 10:46 AM

    @sibrosan @david_chisnall in short, I am not generalizing or stutter.
    Using LLM implies and moves costs elsewhere. It is not zero cost approach. If those costs are not covered, using LLM is moot. Because it is not deterministic and can't be proven it can be deterministic, you imply costs down stream.
    This why programmers should not assume roles of project critical decision making.

    💬 0🔄 0⭐ 2
  • Sibrosansibrosan@mastodon.social
    Jul 19, 2026, 11:43 AM

    @peteriskrisjanis @david_chisnall

    Apparently Torvalds evaluated the use of AI in software development and from a technological standpoint deemed it useful. I think he is certainly qualified to make that judgement.

    And to be clear, I don't see Torvalds advocating AI in general, let alone using LLM in a naive way, or introduce a reliance on AI for Linux kernel development. He just opposes implementing a policy to forbid the use of AI by kernel maintainers outright.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 지지 ᚠד (Jiji Freya Daniel) 🐢✨🧡CyReVolt@mastodon.social
    Jul 18, 2026, 1:58 PM

    @david_chisnall I still see the opportunity for getting people on the ground and figuring out who's really interested in advancing the projects.

    While painful for reviewers now, it's still the same kind of engagement - ask back what they think the contributors are doing, what they think their changes do, what their understanding of the context ist, etc..
    It requires honesty, the basis of building up trust relationships.

    💬 0🔄 0⭐ 0