579
Open source game engine Godot will no longer accept AI-authored code contributions: 'We can't trust heavy users of AI to understand their code enough to fix it' (www.pcgamer.com)
submitted 4 months ago by to c/indiegamedev@lemmy.world
57 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] 39 points 4 months ago (14 children)

Very reasonable, in my opinion. PRs for features that will go unmaintained, have unexpected side effects due to AI not being capable of comprehending the entire codebase, or be able to guarantee their code isn’t ripped from a non-MIT licensed source, is a recipe for future disaster.

Better to put up reasonable barriers to entry so that committed community members can still get involved (who could possibly become the future maintainers of the project) while denying risky, low value contributions from statistical models trained on goodness knows what.

  • permalink
  • fedilink
  • source
  • hideshow 14 child comments
    •
  • [+] -26 points 4 months ago (13 children)

    due to AI not being capable of comprehending the entire codebase, or be able to guarantee their code isn’t ripped from a non-MIT licensed source

    As if a human contributor isn't like that.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 13 child comments
    •
  • [–] 20 points 4 months ago* (last edited 4 months ago)

    The reviewers can catch a bad non-AI programmer by simply looking at the pull request. AI-written code on the other hand usually looks fine and the cracks will only show when testing or building upon that code.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 3 points 4 months ago* (11 children)

    The human contributing human-made code knows where they themself were trained.

    The human "contributing" slop has no idea where it was trained.

    It also helps that humand are capable of thought and have an understanding of contexts whereas statistical models are just guessing the next word.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 11 child comments
    •
  • [+] -8 points 4 months ago* (10 children)

    The human contributing human-made code knows where they themself were trained.

    A human also can copy-paste code from a non-MIT licensed source.

    And you didn't address, that a human too is not guaranteed to comprehend the entire codebase of the project.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 10 child comments
    •
  • [–] 2 points 4 months ago (9 children)

    And if the reviewer catches that shady and dishonest practice then they can reject the submission and blacklist the user.

    Just like with AI submissions.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 9 child comments
    •
  • [–] -2 points 4 months ago* (8 children)

    Just like with AI submissions.

    That's my point.
    AI slop is hardly worse than human slop.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 8 child comments
    •
  • [–] 5 points 4 months ago (1 child)

    Human slop is less likely to get off the ground.

    For example: https://github.com/TuncorReUnion/TLAC-MODERN-LOCAL-ANTI-CHEAT-REUNIONED/blob/main/kernel/tlac_kernel.c

    If a human were capable of crafting that much of a kernel module, they absolutely would know better than "if you are suspicious, you have to name yourself suspicious" logic on display. That lack of understanding usually precludes getting to an even vaguely coherent code for a kernel module.

    Even as human slop manages to compile, it tends to be more obviously bad looking. The AI slop looks more like decent coding practice, just wrapping dumb concepts.

    Finally, there's the volume. Human slop takes someone fumbling a long time to deliver relatively little code. AI slop vomits up gigantic changes in moments. So with the door open to AI slop, you get overwhelmed with the volume of crap.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 0 points 4 months ago* (5 children)

    No, AI is hardly worse than civil malpractice which is why it should always be against the rules to slop.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 5 child comments
    •
  • [–] -2 points 4 months ago* (4 children)

    civil malpractice

    You forget there's also honest civil incompetence

    And that not every AI output is a slop.

    I'm saying that as a person, who doesn't like AI (LLM) and consider it almost dead end technology.

    P.S. It's quite petty of you to downvote me.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 1 point 4 months ago* (3 children)

    It's 100% knowing and willfull malpractice to go to some source, steal code, and submit it as your own.

    Also, I will downvote you every time, fuckin slopper. You lost all my respect instantaneously while advocating for slop.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •