Login
You're viewing the slime.global public feed.
  • [object Object]zzt@mas.to
    Oct 19, 2026, 1:48 PM

    huh, somehow I missed this while I was distracted with netbsd (thank you @techokami)

    awful.systems/post/10050264/12

    > Thanks for linking my toot, here’s another update since then: the same person that found the race condition also found that there was a performance regression on non-synthetic benchmarks. It went super fast on his little test but is complete dogshit when used in the real world! lwn.net/ml/all/atI0edrmeiRuRtN

    the remarkable kernel patch that remarkably convinced linus torvalds to go all-in on LLMs, remarkably running into all of the reasons why this was a horrible fucking idea

    anyone with an ounce of familiarity with how LLMs function could’ve seen this coming, but not kernel boy. why wouldn’t you slop up a feature and all of its tests and its documentation and then just go on vacation so someone else can try to clean up your mess?

    💬 9🔄 80⭐ 77

Replies

  • 💬 1🔄 0⭐ 0
  • zivir'karihsza@social.tudbut.de
    Oct 19, 2026, 2:12 PM

    @zzt @techokami when was it that he last contributed anything useful whatsoever. anything at all to warrant his and others’ high opinion of himself? hes basically thinly veiled dead weight no

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 8
  • 💬 0🔄 0⭐ 0
  • Taeltael@yiff.life
    Oct 19, 2026, 2:00 PM

    @zzt "Bottom line is, they do not want to have those labels because that would enable someone to actually measure how good or bad LLMs are at code.

    This is another thing that is extremely familiar to most commercial developers: there’s an LLM-powered code review bot, but unlike any tool ever before, nobody’s even trying to capture statistics whether it works or not. Why? Because whoever rolls out AI tooling isn’t sure that the tooling actually works for its stated purpose, and wants to for sure get their recognition and promotion for rolling out AI tooling, without jeopardizing it with the risk of adverse KPIs."

    This pretty much sums the whole thing up. If these tools worked people would be gathering statistics on them to show it. They're not because they don't. The only thing that's ever measured is what % of people like/use them. That's simply peer pressure.

    💬 1🔄 15⭐ 26
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:11 PM

    @tael @zzt

    We're seeing full on attribution to the bot at my work and... Apparently the other devs don't even feel bad about it.

    Also... The commit messages have become _paragraphs_ long.

    How the fuck is that helpful?

    Comments in the code are _paragraphs_ long.

    You know what that means, right?

    💬 3🔄 3⭐ 14
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:12 PM

    @zzt

    Anytime someone wants to know what comments were added for given commit or ticket... They're just going to ask the robot to summarize it for them because it's fucking _overwhelming_ to read.

    The writing is absolutely inundated with noise. If there's signal in there _at all_ it's buried. The cognitive overhead just to filter the bullshit is so high... No one is going to bother.

    Nice.

    The effect would be the same if we didn't _have_ comments or commit messages.

    💬 0🔄 1⭐ 4
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:15 PM

    @zzt

    And the only conclusion I can come to is... The people I work with and... Half (more?) of the dev community has forgotten that high-level programming languages were created so a _human_ could read them or (and this is my fear) they _never fucking knew that_.

    Commit messages are for _humans_.
    Objects in OOO code are for _humans_ to reason about. Otherwise we'd _all_ just write non-stop data transformations all fucking day.

    💬 0🔄 4⭐ 12
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:17 PM

    @zzt

    "Code" is for humans. The compiler, pre-compiler, parser, etc turn the _human_ readable stuff into _machine_ readable stuff.

    The shit we _can't_ read is for the machine.

    These concepts aren't _hard_ but they are fucking fundamental.

    I didn't get a fuck'n CompSci degree and I still learned these things early.

    Why the fuck doesn't anyone else seem to know this or care?

    🤪 I'm fuck'n losin' over it.

    💬 0🔄 7⭐ 24
  • 💬 2🔄 0⭐ 5
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:23 PM

    @maaneeack @zzt

    I'm _positive_ I'm the only human doing code review at my job.

    I'm constantly encouraging my co-workers to assign their CRs to me so at least _someone_ has _tried_ to read and understand their changes.

    I ask questions. I push back on weird implementations. You know... The stuff you fuckin' do for a code review, almost no one else ever did this at my work. Which tells you what they think of code review, in general. My belief is they simply don't understand it's purpose.

    💬 3🔄 3⭐ 21
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:25 PM

    @maaneeack

    Even our f'ing VP thinks a code review should only take a few minutes.

    Why fucking bother if that's the case?

    Apparently they basically agreement with my sentiment and think that Claude(tm) can handle the task.

    Doesn't bother a soul that the non-deterministic robot that wrote the fucking code is also reading the code and making random word-salad comments about the code, answer itself, resolving any issues itself...

    ...

    For fuck's sake what did I we come to work for today?

    @zzt

    💬 1🔄 3⭐ 19
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:28 PM

    @maaneeack

    I spent somewhere around 6 hours doing code review last sprint. I'm not patting myself on the back, I'm saying... I did _a lot_ of fucking code reviews.

    I had questions. I had _suggestions_ for simplifications and clarity improvements. I _talked_ to the devs who did the work to find out what their thought process was.

    Why the fuck am I not getting manager pay? I'm already doing all the work. GAH!

    Anyway, no one else thinks there's anything wrong with how we work.

    @zzt

    💬 2🔄 1⭐ 11
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:30 PM

    @maaneeack

    But _everyone_ is completely surprised that the application is failing in strange and novel ways.

    They cannot correlate the results to the behavior changes, _at all_.

    I sit in meetings and I'm constantly saying to myself, "I cannot be the smartest person in the room. I cannot be the most cognizant person in the room. I cannot be the only person that cares in the fucking room."

    That all probably sounds like arrogance, it's not. It's empirically observed. And it's insane.

    @zzt

    💬 1🔄 0⭐ 10
  • 💬 0🔄 0⭐ 3
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:49 PM

    @shellsort

    Where are they? Where are they going? I wanna go too. I want to work with people who care, for a company who cares.

    This wild indifference and C-level constant recapitalization grifting scheme bullshit has to end, at least for me.

    I'm constantly on the verge of telling senior folks about themselves and... That's not going to genuinely help anyone. It'll be _briefly_ cathartic and then I'll be one those devs on the market.

    @maaneeack @zzt

    💬 1🔄 0⭐ 3
  • shellsortshellsort@mastodon.social
    Oct 19, 2026, 3:52 PM

    @401matthall @maaneeack @zzt A few weeks ago I funded my own business. Quite a few competent people in my circle are in the same position as you and have asked me to hire them away from their current place. I really wish I could, but I dont have the money unless they bring some kind of customer with them. :/

    💬 1🔄 0⭐ 3
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:28 PM

    @shellsort

    OMG, heard. My previous gig kinda went like that. I left the place I'm at _now_ went and worked on some really cool Elixir projects and the company I was with grew a little too fast and the market contracted a little too fast at the exact same time. That was that.

    So, I came back _here_ and asked for more money.

    Somehow, they took me back and gave me more money.

    I still don't want to be here.

    💬 1🔄 0⭐ 1
  • 💬 1🔄 0⭐ 1
  • 💬 0🔄 0⭐ 1
  • maaneeackmaaneeack@noc.social
    Oct 19, 2026, 3:34 PM

    @401matthall
    it definitely shouldn't be this way. I'm not even a programmer but leaving the understanding to the computer is stupid. unless you can get c-suite to understand it's just gonna continue and get worse.

    I've been looking for work since the end of March last year and I can't even get a remote tech support/product support anymore. All the listings are fake or llms talking to llms
    @zzt

    💬 0🔄 1⭐ 2
  • Sharp Cheddar Goblinsharpcheddargoblin@reclusive.blog
    Oct 19, 2026, 3:47 PM

    @401matthall @maaneeack @zzt I remember code reviews for a relatively minor refactor that had two senior devs and the project architect in a room watching as I walked them through the changes. Lots of questions, clarifications, a few suggestions. Everyone walked out with a complete understanding of the changes and impacts for the system. That's a code review. I don't know what the fuck people are doing now, but whatever it is it's not engineering.

    💬 1🔄 0⭐ 0
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:56 PM

    @sharpcheddargoblin

    Whew, no, it's not. You're absolutely right. I'm able to see about 20% of what's going on at my company and I have to intuit the other 80% but I fairly convinced we're simply driving metrics to make management look good, or at least promotion/accolade ready, ad nauseam.

    Managers and up get MUs (stock equivalents with caveats) and bonuses, the MUs get significant the further up the chain you go, when the company is re-capitalized you can cash out your MUs.

    @maaneeack @zzt

    💬 1🔄 0⭐ 2
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:59 PM

    @sharpcheddargoblin

    So, the incentive isn't _on_ quality. It's on performance for the quarter, always the quarter.

    And performance is _always_ a loaded metric. It _never_ measures anything tangible or meaningful.

    Like `velocity`... We need to improve our velocity.

    OK. Um... What _is_ it?

    "Oh, it's the rate at which defined work is completed in relation to the estimated effort the work requires in a given sprint."

    Oh.

    @maaneeack @zzt

    💬 1🔄 0⭐ 4
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:01 PM

    @sharpcheddargoblin

    And estimated effort is tracked... How?

    "Story points."

    Cool. What're those?

    "Measurements of effort but we just use hours."

    Uh, alright. So... We're just guessing how long work will take? And hoping that the scope is defined well enough to not completely blow the estimate?

    "We groom our backlog weekly to ensure the tickets are properly estimated."

    I've _never_ hit an estimate yet. Most of the team hasn't either. They either finish early or late.

    @maaneeack @zzt

    💬 1🔄 0⭐ 4
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:03 PM

    @sharpcheddargoblin

    "Right, so our velocity is improved when you, as a team, accomplish _more_ story points in a given sprint."

    OK... So... Just increase all the estimates by some factor and then finish most tickets early. Velocity goes up.

    "Well, that wouldn't be an accurate measurement of our velocity."

    ...

    Yeah. I know. But you don't _actually_ care or you wouldn't encourage us to take note of our velocity every two weeks at the review.

    @maaneeack @zzt

    💬 1🔄 0⭐ 2
  • 💬 1🔄 0⭐ 3
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:06 PM

    @sharpcheddargoblin

    And that shit will eat entire _days_. And it's worse than doing nothing because you're still exhausted at the end of it and all you did was change the color of a cell on somebody's stupid-ass Sales Force plugin.

    You wanna fucking talk to me about value-add?! Why don't we discuss how much time you spend reviewing TPS reports each week, mother fucker?

    (I'm fine. Really. I'm fine. This is fine.)

    @maaneeack @zzt

    💬 1🔄 0⭐ 3
  • 💬 1🔄 0⭐ 2
  • 💬 0🔄 0⭐ 3
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:31 PM

    @Aut

    Yep. I don't feel like there's anything I can do to stop my co-workers from using it the encouragement from above is _strong_.

    I _do_, however, remind them _constantly_ this thing is _ltierally_ non-deterministic. It can, and will, give you a different answer to the same prompt multiple times in a row. Everything about AI/LLMs is a _cheat_.

    Since it's non-deterministic to _give_ the robot a sense of continuity that it _fundamentally_ lacks they created contexts.

    💬 1🔄 0⭐ 1
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:35 PM

    @Aut

    Then bundles of context became Skills, with a capital S.

    These things are _all_ work-arounds to the fundamental nature of LLMs. They're explicitly incongruent with the tasks they're given.

    This is all scripts, work-arounds, and an insane amount of compute scaling to _hide_ the absolute fail-state of "AI".

    Anyway. I share your pain and frustration.

    💬 0🔄 0⭐ 1
  • Emelia/Emibecomethewaifu@tech.lgbt
    Oct 19, 2026, 3:50 PM

    @401matthall @maaneeack @zzt Same with me here at [redacted]. Others mostly skim it looking for obvious issues, I'm the only one who actually questions things consistently. Like recently, a "test" that one of my coworkers "wrote" (it was extruded by the slop-bot, I could tell just by the fist) that "tested" a config file "to make sure it was set correctly"...

    (Though previously we did have someone who was the exact opposite of my coworkers: nitpicking every goddamn line and "requesting changes" over what are the architectural equivalent of a typo, so PRs couldn't be merged until he approved it...)

    💬 1🔄 0⭐ 3
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:16 PM

    @becomethewaifu

    Heard. There's _definitely_ balance to be had. Typos, code style, formatting guidelines... Linters take care of that. Humans should be looking for genuine inconsistencies, lack of clarity or simplicity in the code. Shit that's genuinely going to increase the cognitive overhead when somebody has to change this file at 3am in the morning because a production server went down in Roanoke, VA and now we're showing all our customers an `Error[error, error]` message.

    @maaneeack @zzt

    💬 1🔄 0⭐ 3
  • Emelia/Emibecomethewaifu@tech.lgbt
    Oct 19, 2026, 4:35 PM

    @401matthall @maaneeack @zzt Oh this wasn't a "balance" thing, many of his comments were his personal opinion on how code should be structured, aka "how he would have written it" or him wanting us to do things in more convoluted ways for no real benefit. His comments weren't "a human linter" so much as comparable to LLM "reviews"

    And the convolutions he suggested weren't "optimization", it was almost always "taking a simple straightforward thing and turning it into an over-engineered nightmare." He was on the same side as the "architect" that required we use a network API "microservice" just to shuffle some JSON fields around. For something that would have otherwise been entirely self-contained.

    💬 1🔄 0⭐ 2
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:53 PM

    @becomethewaifu

    👀

    M'kay. That's clearer.
    Good thing I wasn't in any of those reviews or meetings.

    The number of times I've just said in meeting, "Oh, I'm not doing that," is _way_ too high. At some point someone probably _should've_ disciplined me or fired me but... That hasn't happened so... I continue pushing back... Assertively.

    Them: "Oh, you're gonna do it."

    @maaneeack @zzt

    💬 1🔄 0⭐ 2
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:55 PM

    @becomethewaifu

    Me: "I'm not writing a single line of code for your newly distributed Rube Goldberg machine until _you_ write the unit, integration, and acceptance tests for each change."

    Any further pushback just gets a, "Fuck you. You do it. You maintain it. You document it."

    Again... I probably shouldn't still be employed but... Fuck _all_ that noise.

    💬 0🔄 0⭐ 1
  • CM Harringtonoctothorpe@mastodon.online
    Oct 19, 2026, 3:24 PM

    @maaneeack @401matthall @zzt which of course is infuriating and confounding… but let’s also be real. I used to work in corpo land, and the *bulk* of code never got truly read/ understood/debated in anything but the most cursory trivial ways…

    Because the incentives are there to ensure that happens.

    So in a very real way, folks are just scaling out their already performative/perfunctory actions.

    💬 1🔄 0⭐ 3
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:34 PM

    @octothorpe

    This. 💯 So fucking much this. It's exhausting.

    Maybe your experience was different but we don't have meaningful tests to avoid breaking changes. No meaningful unit tests (honestly we don't have _units_ to test. It's legacy code written consistently by the youngest, cheapest devs they could find), no acceptance tests, no end-to-end tests.

    We _have_ a QA team but they don't even understand the application.

    @maaneeack @zzt

    💬 2🔄 0⭐ 2
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:36 PM

    @octothorpe

    We make unbounded changes to the code, somebody "signs" off on it, and if it's urgent enough we'll even cut a git.patch so we can update production with out actually going through our meaningless release process.

    It's a fucking miracle anything works, at all.

    I've been at three or four different companies in my _almost_ 15 years. Only one place I was at did meaningful testing/review/processes and it was a small shop.

    So... Is this just how the world operates?

    @maaneeack @zzt

    💬 1🔄 0⭐ 2
  • 💬 1🔄 0⭐ 2
  • 💬 1🔄 0⭐ 2
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 3:51 PM

    @octothorpe

    I _deeply_ empathize with this. This is the time in my life where I look back and _wish_ fiscal responsibility had been higher on my list. So I would have _choices_ now.

    I carry enough debt that I can't manage a pay cut and I'm pretty-okishly-compensated where I'm at... With the exception of all the bullshit.

    But I _feel_ all of that, man.

    @maaneeack @zzt

    💬 1🔄 0⭐ 2
  • CM Harringtonoctothorpe@mastodon.online
    Oct 19, 2026, 4:52 PM

    @401matthall @maaneeack @zzt ooof. I feel you though. And yeah, I got lucky enough to be in the right places at the right times. I may work again (I retired to take care of my husband full-time), but if I do, Imma take a long look at what the tech landscape looks like, and see if I want to do that again, or something radically different like intern at a residential architecture firm.

    💬 1🔄 0⭐ 2
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:58 PM

    @octothorpe

    <3

    Right back to the Office Space question, man, "What would you do if you had a million dollars?"

    I'd solve _meaningful_ problems for a given value of meaningful. <3

    💬 0🔄 0⭐ 1
  • CM Harringtonoctothorpe@mastodon.online
    Oct 19, 2026, 3:40 PM

    @401matthall @maaneeack @zzt Sadly, my experience (for the most part, save two orgs I’ve worked with) mirrors yours.

    Codified/automated testing is a joke as time to create and test those is always deprioritised, and QA doesn’t actually have the power to prevent a build from shipping, and items they do log are never prioritised. Only after a major customer gets on the phone does that one thing that broke get attention, and even then in the most emergency slapdash reactive panicked way.

    💬 2🔄 0⭐ 2
  • 💬 1🔄 0⭐ 2
  • 💬 0🔄 0⭐ 2
  • maaneeackmaaneeack@noc.social
    Oct 19, 2026, 3:47 PM

    @octothorpe @401matthall
    Same for me, I found an OpenSSL deployment bug on a release of a big update and I had to make a KB for an urgent issue + workaround while devs updated it and patched it

    our support team often asked each other if devs tested any of the updates they submitted or not

    @zzt

    💬 1🔄 0⭐ 3
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:10 PM

    @maaneeack

    I used to. Religiously and consistently. Then I was told I spend too much time testing and I should let QA do it.

    👀

    "I'm not _QA_ing. I'm making sure the code does what I fucking _think_ it does you psychos. Settle the fuck down."

    My work tended, in general, to be high quality and didn't _usually_ need urgent fixes right after deployment (there's _always_ exceptions to this; you ever drop a production database in the middle of tax season?).

    @octothorpe @zzt

    💬 1🔄 0⭐ 4
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 4:12 PM

    @maaneeack

    My co-workers were faster to the _initial_ release that might be buggy or break production but they still _said_ they were done and that _counted_.

    They'd have to fix it on production, edit live databases, manually update SQL schemas instead of writing clean migrations.

    My work would have all that upfront but I was slow. I wasn't a _closer_. 👀 (Like I was asking for coffee in Glengary Glen Ross, or something.)

    Anyway, they want code? They can have it now.

    @octothorpe @zzt

    💬 2🔄 0⭐ 3
  • 💬 0🔄 0⭐ 2
  • 💬 1🔄 0⭐ 3
  • 💬 0🔄 0⭐ 2
  • Mark Levisonmlevison@hachyderm.io
    Oct 19, 2026, 5:25 PM

    @401matthall @tael @zzt

    Funny detail. Longer comments and longer files cost more tokens to process. In addition the model can handle fewer files due to the bloat.

    (This on top of your other points)

    💬 1🔄 0⭐ 1
  • 💬 1🔄 0⭐ 0
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 5:28 PM

    @mlevison

    Almost like the arrived at solution is inherently inefficient and a poor choice for the marketed task.

    👀

    💬 0🔄 0⭐ 1
  • 💬 1🔄 0⭐ 1
  • Matt Hall401matthall@mastodon.xyz
    Oct 19, 2026, 8:30 PM

    @tael

    I'm not sure I follow the _intent_ of the term. I've heard it before but I've never used it in context.

    So... In this case tokenmaxxing is a _bad_ thing right? Because I'm consuming all available tokens while trying get the slopbot to process a prompt, right?

    Or is more different?

    💬 2🔄 0⭐ 0
  • 💬 1🔄 0⭐ 1