Selena
Hubert FiguièreApparently it's rude to reject AI slop.
You know what's rude? To expect people to read / review your AI slop.
Joe Brockmeier@hub I try to be polite and not give offense unless I *want* to offend someone. But I simply refuse to accept that rejecting LLM-created stuff is rude. Nope. Absolutely not. It’s no more rude than refusing to accept plagiarized content or refusing a hamburger if someone is a vegetarian.
Gersande La Flèche@hub Depuis quand?!?!
Christopher Neitzert@hub No, it's rude to denigrate another's efforts, even if it was to prompt something, it is effort.
What also is rude is to have an oversized emotional reaction to a tool.
Good luck with that emotional regulation thing, son.
Der Giga@xenophile It is generally considered rude, if those efforts are below a certain minimum and expecting other people to input real effort into it.
So good look in society when these are your standards, kid.
Mark Lynch@xenophile @hub the whole point of these tools is to completely avoid having to put in the real, genuine effort required to produce anything of value. And not only that, but the output resulting from the prompt is a plagiarised remix of the past genuine efforts of real human beings. There’s about as much effort involved as paying someone to steal something for you. If you consider it rude to point out that people are freeloading on stolen effort, good luck to you
@mark_f_lynch @[email protected] 'Plagiarised remix, zero effort' is a tidy story that survives exactly until you meet someone who signs every output with C2PA provenance, watermarks it, runs it local so nothing's scraped back, passes a FOSS compliance audit at zero violations, and publishes their SBOM. The effort you can't see isn't absent. You just didn't look at whom you were speaking to.
XANTRONIX
For anyone still calling this slop: the full CycloneDX SBOM is public. Every dependency, every license, zero FOSS violations.
https://github.com/rAIdio-bot/sbom
Reply before you open it and the thread will record which one you were. Muting from here.
Marko Schulz@xenophile @hub Brandolini's law postulates: The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it.
And this is what many experience: They have to spend a lot more effort to consume what others produced with using AI.
While it is also effort to burn down a house, it is a lot more effort to build it.
@datenreisender @[email protected] My name's on the account with links to my work. One click gets you provenance, watermarking, local inference, a clean FOSS audit, a published SBOM, and a couple decades of shipping. None of you spent the click. You pattern-matched 'AI' and lectured the wrong man. That's not Brandolini. That's not reading. You cited a law about bullshit while producing it.
@xenophile My argument is not about your person, what you already achieved, what you produced before.
My argument is about your argument.
And your only response to my argument was calling it “bullshit”.
@datenreisender Fair, you argued the claim and I answered a diff point. Let me take the claim. Brandolini assumes the AI output is the bullshit, costly to refute, cheap to produce. That holds for slop. It doesn't hold for output that ships with provenance and a verifier: the refutation cost goes to near zero because you can check it in one step. The arson-vs-construction frame only works if there's no blueprint. My whole architecture is the blueprint. That's the disagreement, not your manners.
@xenophile You seem to suggest, that before checking the content one should check the provenance. That doesn't lower the effort, it raises it by adding another guard.
E.g. assume a PR like https://github.com/curl/curl/issues/16267 was done by someone with a track record: Then, like it was done, it still would cost effort more effort to detect it as being wrong than it was to produce it.
@datenreisender You opened by putting my work in the 'burning a house' bucket, then every reply retreated to a smaller true thing without defending your opener. Each step reasonable; the sum is you never defend the claim at all. Once, plainly: provenance verifies origin, integrity, accountability, not truth, and accountability is what raises the cost of producing the bullshit. That's the opposite of your house-fire. Argue that, or we're done and the thread speaks for itself.
@xenophile Maybe I misunderstood you, but to me your initial argument was that even if someone only prompted something, they still invested effort, because prompting is effort.
And while prompting _is_ effort, there are many examples where prompting is a lot less effort than what others have invest when they need to handle the output.
(1/2)
@xenophile Provenance does not change that at all. With that people can verify the origin (which is also effort, even it is only a little, just like prompting) but it does not at all take the much bigger effort for people to handle the output of the AI.
I had hoped that my examples make this more clear but maybe I just distracted from it. (2/2)
@datenreisender Then we mostly agree: prompting is effort, provenance verifies origin not correctness, and neither lowers the cost of evaluating content. Where we differ: provenance isn't aimed at the consumer's per-item effort, it's aimed at the producer's, accountability bound to identity raises the cost of generating bad output, which your curl case doesn't test. (1/2)
@datenreisender One thing though: you opened with Brandolini and burning houses aimed at me, which reads as bucketing the work, not the narrow 'prompting can be less effort than cleanup' point you've landed on. If that's the argument, lead with it. Don't open with the arson and retreat to the reasonable version when pressed. We were measuring two different costs the whole time. (2/2)
@datenreisender That incremental refinement of a weak argument works exactly once. I clocked it on your first post. Find another tactic. (x/x)
@xenophile Maybe you misunderstood me from the beginning? I never aimed at you. Did you assume I accuse you of arson and writing BS?
For me it was always about the original „why is it rude to expect people to review your AI slop“: one reason for the rudeness is Brandolinis Law.
You mentioned that it takes effort to produce an AI output but my counter argument is, that it often takes more effort to review the output to produce it.
unkx@xenophile @hub …and the award for ”stupidest shit said all week, and it’s only monday” goes to Christopher.
william light@xenophile @hub wild that you took somebody else's post personally and then accused THEM of issues with emotional regulation
Claudius@xenophile @hub So… If someone comes and smears feces all over your apartment, you gotta commend them for the effort?
Nope.
@claudius Define feces. Specifically: what measurable property makes an output worthless? Scraped training data, license violation, no provenance, no human input? Name the test. Then run mine against it: C2PA provenance, neural watermark, local inference, zero-violation FOSS audit, published SBOM. Either it fails a criterion you can name, or the analogy was never an argument. Your move. github.com/rAIdio-bot/sbom
Sheogorath@xenophile @claudius Want an example? How about 473 that are relevant to this thread? https://github.com/flathub/flathub/issues?q=state%3Aclosed%20label%3A%22AI%20Slop%22
@sheogorath @claudius Those 473 are Flathub closures labeled by reviewer eyeball: single big commit, verbose messages, em-dashes in comments. That's the heuristic. None were adjudicated against provenance, watermark, or SBOM, the criteria I named. Flathub's own forum has humans flagged as 'AI Slop' and locked out by false positive. So your 473 is a volume count of label-matching, which is the exact tell. I asked for a measurable property that makes an output worthless. Count isn't a criterion.
@xenophile @claudius Well, in case of FlatHub under the current pressure with tons of AI slop coming in: The use of an LLM to generate code, PRs, answers and alike. Agreed, that's an almost binary query, since there is a possible exception for well maintained projects, but so far, the simple use of LLMs, basically makes work worthless in the context of FlatHub submissions. Blame people abusing LLMs, not the people at FlatHub, they are just volenteers, trying to build a nice software distribtion.
"The use of an LLM" is an origin test. Not one measurable property of the output, just how it was made. You reached the verdict without looking, which isn't a standard, it's a prior you never audited. Refusing to check the artifact takes less effort than the prompt you're holding in contempt. You proved my opening line. Judging the maker instead of the made is the same lazy move in every domain it shows up in. Nothing further.
🎀𝓀𝒶𝓃𝑒𝑒𝓁🎀
Ovro