[–] 58 points 2 months ago* (last edited 2 months ago) (7 children)

I think it’s telling that llama.cpp, the interface engine for LLMs on desktops, has blanket banned vibe coded PRs, as they simply cannot deal with them with the manpower they have.

…An LLM inference project. Banning LLM vibe coding.

I remember a collective groan from the devs when Google’s TurboQuant paper trended on Twitter, and they knew there’d be a flood of Claude PRs to “implement” the feature, even though most of it already existed in the engine.


vllm (the premiere enterprise server inference engine) is quite careful about them too, though they use bots to filter them and have more manpower to cull what slips through.

  • source
  • [–] -1 points 2 months ago

    I think people should try to start thinking about this stuff in terms of what assumptions need to change, rather than how it contributes to pro-AI or anti-AI narratives. If LLMs can automate discovery of exploits, that means knowledge of them is easily accessible and that whole group of exploits is going to just be off the table unless the software maintainer really doesn't give a shit about fixing them.

    Making this about the moral failings of "slop kiddies" misses the point, if there is little to no special sauce on top of the LLM itself then some Sony employee could have done this and probably will in the future.

  • source
  • parent
  • context
  • [–] 8 points 2 months ago* (1 child)

    Especially

    While Euro-Office champions European independence, it has been under scrutiny over its ties.

    The software is based on the code of OnlyOffice, an open-source productivity suite owned by the Latvia-registered company, Ascensio System SIA, whose roots trace back to Russia.

  • source
  • parent
  • context
  • [–] 27 points 2 months ago (3 children)

    Same brother, same. I'm retiring the end of this year. I transitioned from a dev, to contract work, to now a contract code reviewer. I'm sick of it. I get paid to go to random startups or other companies to review code, explain to them why it doesn't work because they don't know WHY it's not working, and when I tell them they thank me for my time and keep doing it. And they all lie about it. They're all using AI for end to end builds but they simply continue to lie that THEY built it until after you consistently prod them they admit "yeah I had Claude build it, I don't know how it works". It's like dealing with a drug addict like you said. You know they're doing it, everyone around them knows, and yet they won't admit it.

    So i'm done. Even with the FOSS projects I work on I'm done because now you're getting PRs from randos using omarchy they think they can now contribute to projects and when you review their shit they can't explain what they did, why they did it, or why they're even submitting it.

    I'm totally done. I'm done with any type of code. the past 2 years have killed the past 25+ years of passion I had for this stuff. It's not fun anymore and I feel like I wasted half my life.

  • source
  • parent
  • context
  • [–] 25 points 2 months ago (2 children)

    I wouldn't be surprised if that's the goal.

    To be honest though, it's not gonna happen. It may kill GitHub eventually, but AI slop isn't really much of a problem on platforms that don't actively promote and sell it (e.g. Codeberg/Forgejo).

    Devs will just migrate their projects off of GH; some already have. FOSS is too strong a movement to be killed by the corporate bullshit machine.

  • source
  • parent
  • context
  • [–] 57 points 2 months ago* (last edited 2 months ago) (10 children)

    I use AI at work. Not "vibe coding" or anything: I use it locally on a DGX Spark machine to answer simple technical questions and spew out boilerplate. I admit it's really useful, and I work quite a bit faster.

    But here's the thing:

    There's no effort required to do anything anymore. It makes all achievements bland.

    It means my 40 years of experience are pretty much worthless, when anybody can ask an LLM what it took me decades to master.

    We go through the motions at work: someone makes a PR and asks for code review. Someone else reviews the code and then it's merged.

    But here's what really happens: more often than not, the guy requesting a PR asked AI to make the code. The guy reviewing the code asks AI to review the code. And if those two talk to one another, in the background, they pass each other's questions to their individual LLMs.

    And you can tell nobody really understands what the LLMs have done when you start asking them detailed questions about the code: they start blurting out nonsense, vague explanations of what they "did", before eventually admitting that they don't fully understand it and they asked AI to do it.

    In other words, the (shoddy) work is done by machines, and the human workers are just proxies. It's a simulacrum of humans working together. It's really bizarre, it's utterly absurd and it's insulting to those of us still trying to do the work ourselves. I don't want to review machine-made code, especially when some guy asks me to do it without telling me he didn't do the work himself. Fuck that!

    And the worst is, I'm the first to recognize that I can't continue to work at the pre-AI pace I used to work, even though I am a fast human worker: I'm completely outclassed by the machines. They create bad code but they do it fast, and most of my colleagues are completely happy with good-enough shit they don't understand.

    Hell, one even asked me the other day why I keep bitching about the code being commented and readable when nobody reads it anymore, and AI can read the code and produce summaries. That's... bleak.

    And it's so easy it's like cocaine: you know AI is bad for code quality, but you can't help doing it, because it's quick and lazy.

    I'm not built for this. I hate it with a passion. It feels like being a great horse-drawn buggy driver in the age of automobiles: I know my skills have been obsoleted and I'm about to be completely washed out. I could transition and start driving cars - and to an extent, I have already. But I just don't wanna. Driving cars is not fun, it's not exciting, it's not challenging, it's boring and it's utterly depressing.

    So I don't code for myself anymore. What's the point... I could do it pre-AI stylee, but then I'd be very slow, and I'd have to refuse contributions - because all contributions are AI slop, or very nearly so nowadays. And eventually I'd be tempted to use AI in my projects anyway, and I have too much pride in the quality of my code to do that - not to mention, I have to use the damn technology enough during work hours as it is.

    I've informed my boss that I would like to transition to another position in the company. I love my company, I love the folks I work with, and I used to love what I do. But AI ruined everything in the space of less than 2 years. Now I fucking hate dragging my ass to the office every morning, and I relish the days when I have to spend time in the workshop machining parts, away from fucking computers.

    Luckily, I'm senior enough and respected enough that I can request a radical change in role without being thrown out for being unreasonable. And my boss understands why I hate AI. But at my age, it's not gonna be fun to change job completely once more. I was really hoping to retire quietly doing what I do now.

    Anyway, long story short, and to answer your question: it's not a matter of trusted devs or the tools they use: I want nothing to do with computers anymore. If it was up to me, I'd be machining parts in a blue overalls all day every day until I retire. Fuck computers.

  • source
  • parent
  • context
  • [+] -16 points 2 months ago (4 children)

    “Slop kiddies found the only hypervisor bug left, which I had also found a while ago, and decided to report to Sony,” the modder said on social media. “I asked them to at least wait for GTA 6 to come out so that people would have the opportunity to legally purchase the game and also enjoy linux. They agreed to wait, but not a day passed and they decided to waste it instead.”

    isnt this proof that AI security scans AREN'T useless? They found a legit exploit that a skilled human hacker found but didnt disclose

    This is all more of a human problem than an LLM problem.

    You wouldnt reject a LLM assisted patch sent by Linus Torvalds, or label him a skop kiddie

    Idiots using LLMs and submitting slop is a human problem, not proof that LLM coding tools suck

  • source
  • [–] 9 points 2 months ago

    If I have ai fix something I just make a bug report telling them what I had ai fix for myself, I don't push it for anyone else and update to the actual version when they fix the bug correctly

  • source
  • parent
  • context
  • [–] 72 points 2 months ago (3 children)

    “used AI to detect a bug that they were using to crack consoles, and reported it to Sony for a bounty.”

    Wow, that's literally "McDonalds burger flipper who narc'd on Luigi" level of stupid. And they are likely about to obtain the same amount of prize...

  • source
  • [–] 34 points 2 months ago (4 children)

    Yeah, I can't imagine being an open source developer for a popular program these days. They get absolutely flooded with slop, constantly. I wouldn't be surprised if it somehow makes open source die, or at least go into the shadows.

  • source
  • view more: ‹ prev