-8
What is slop? (jeferson.me)
submitted 1 week ago by to c/fuck_ai@lemmy.world
25 comments fedilink hide all child comments

all 25 comments

sorted by: hot top controversial new old
[–] 18 points 1 week ago (1 child)

Fuck_ai

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
    • [–] 10 points 1 week ago (3 children)

    🤣 1
  • permalink
  • fedilink
  • source
  • hideshow 3 child comments
    •
  • [–] [S] -2 points 1 week ago (2 children)

    Totally agree!

    But what when it isn't slop?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 2 points 1 week ago* (last edited 1 week ago) (1 child)

    LLM's are statistically average by design, if one wants something nice then effort and/or expertise is required. To exaggerate: "A broken clock has the correct time twice a day", that does not mean I will use it.

    Even if the output is correct then one could say that the particular case is not slop, but I think that is not the case because the LLM's will never capture our mistakes, biases and emotions. That to me is a no-go in most cases. To me slop is something lazy and soulless, not necessarily correct/incorrect.

    As for technical output like programming, to me is also a no-go because these companies do not respect licenses and overall have bad practices. Like monopolizing open knowledge for example.

    I'm in fact not technically anti LLM's, Neural Networks are a fantastic piece of technology and LLM's are a fantastic implementation. Advancements like chess engines, image recognition, language translation are the most well known benefits of it. In the end I'm anti whatever these big companies with their debt hiding shit are trying to accomplish.

    I hope this answer your question.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] [S] 2 points 1 week ago

    In the end I’m anti whatever these big companies with their debt hiding shit are trying to accomplish.

    Holy mother, tears in my eyes reading this.

    I hope this answer your question.

    Yeah, thanks for the comment and reading the post.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 7 points 1 week ago (1 child)

    So we need to stop calling everything generated by automation slop. Because there is good automation and bad automation, and they don’t live in the same room.

    The issue is that the majority relying on LLMs for their code are lazy with it or just want what they’re creating to work, so naturally some people will associate LLM usage with slop.

    I’ve found it quite difficult to balance light or hard usage of LLMs. Either I end up having to correct and adjust the model’s output all the time, which costs me more time than just writing by hand, or I have to let it go loose on the codebase and end up with code that’s hard or even impossible to maintain without an LLM.

    So I decided to stop using LLMs to write code for me a long time ago and I find it hard to believe there are many people out there who can get this right (not saying they don’t exist).

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
  • [–] [S] -4 points 1 week ago

    The issue is that the majority relying on LLMs for their code are lazy with it or just want what they’re creating to work, so naturally some people will associate LLM usage with slop.

    Lazy is a strong word... But I see that. Most of the companies have pushed so hard that any push back put you in the next layoff list. And in the capitalism, you either work or starve... So the choice is simple.

    [...] difficult to balance light or hard usage [...]

    I imagined that green field projects would be an easier path for LLM to help, I was wrong. Multiple times by the way. Continuing the work isn't also a good thing. My current workflow for coding I can see me using the tool to spot reviews before it happens. Naming variables that in my little head makes a lot of sense, but don't in a general sense.

    But again, my focus is in quality over quantity. Specially for my pet projects. So this is something that I care.

    Regarding time, if you don't look at the code it's a 1.000.000x productivity boost like the Theo’s online want to say, and if you push back is either skill issue or wrong model. But in my view, if you care about the code, and it's quality... the tool barely move the speed dial up.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 5 points 1 week ago (1 child)

    “Automation is slop” is sort of a strawman here, no?

    Slop is pretty clear to me: prompt -> LLM -> post (without careful review and editing).

    Appreciate that you’re writing and sharing, hopefully folks in this comm will not judge you too harshly by headline alone.

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
  • [–] [S] 1 point 1 week ago

    Fair point, and this is what I think we need to have a better judgment on the topic. The tool is bad in a lot of sense, heavy use will have impact that we can't measure since we don't have the data. Some points to psychosis, other to 100x productivity.

    The problem that I'm facing is, if someone says that have "AI assisted" is immediately is shite. The binary view (even on piefed) is a problem, and there is few people gaining from that, and isn't the piefed.

    The same with the AI assisted in the Steam page, there is a shit ton of Slop, but there is use of LLM that isn't a problem. And Steam (and Piefed too) using the "AI" instead of the LLM.

    So names are important, level of usage is important, and quality is also important. Gosh, this comment itself already made me realize there is even more to discuss and think about it.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 5 points 1 week ago (3 children)

    Being neck-deep in shit is terrible, but what about just knee-deep?

  • permalink
  • fedilink
  • source
  • hideshow 3 child comments
    •
  • [–] [S] 1 point 1 week ago (2 children)

    Never saw this problem in this perspective. But considering the world we are, can I be knee-deep then?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
    • [–] 4 points 1 week ago (1 child)

    What about moving the goalposts like, not as much?

    No.

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
  • [–] [S] 1 point 1 week ago

    This is not moving the goalposts. This is identifying who is gaining from this binary view. And it isn't the worker, it isn't the web... So who is gaining from it?

    Which would be somewhat clear if people would read instead of jumping into conclusions.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 3 points 1 week ago* (8 children)

    Completely unclear what "light usage" is supposed to mean. Either you type in a question, let it generate an "answer", read it... or you don't?! What's the middle ground here?

  • permalink
  • fedilink
  • source
  • hideshow 8 child comments
    •
  • [–] [S] 2 points 1 week ago (7 children)

    That's the point of the blog post. Discuss that LLM (or any automation tool in this case) have different level of quality/usage and having a binary view on the topic makes only a very specific class profit from it.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 7 child comments
    •
  • [–] 1 point 1 week ago* (last edited 1 week ago) (6 children)

    The blog article says it's using Photoshop's background removal tool and autocomplete... But that has nothing to do with LLMs, does it?

    (Edit: And they make a point about letting an LLM come up with a first draft... Though I'd argue letting AI come up with your structure / architecture / library choices is kinda slop times 10?!? I mean the basis of something is the most important aspect. And requires most of the effort. The rest is just writing it down.)

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 6 child comments
    •
  • [–] [S] 2 points 1 week ago (5 children)

    But this is automation. And some tools use AI, not LLM.

    Another point that we need to be better. We need to call the tool LLM, not AI. Like I said in the post, spellchecker is a kind of AI, a decision tree is a kind of AI, and so on.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 5 child comments
    •
  • [–] 0 points 1 week ago* (last edited 1 week ago) (4 children)

    Kinda wrong title, then? So don't use chatbots, do use background removal, markov chains and text to speech and speech to text? I'd agree with that...

    Still, none of that has anything to do with large language models. And automation is a different thing as well. It's your webshop automatically validating payment and preparing and invoice and shipping label... Managing your webservers. Or an Amazon fulfillment center which automatically fetches the products from the shelves... That's automation. And it may or might not contain 'AI'. That term only means a programmer didn't program in every step to do, but we give the machine a goal and let it figure out how to do it.

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

    Which title? Because, from what I can see I did use LLM on this. And in the text I was trying to avoid the usage of the LLM as AI and AI as LLM, but I may have left a few wrong uses....
    Edited since the original comment was edited.

    Kinda wrong title, then? So don’t use chatbots, do use background removal, markov chains and text to speech and speech to text? I’d agree with that…

    Not sure, some of the tools fits in the AI bucket, some may not. Still I did try my best in the post to avoid the usage of LLM as AI and also AI as LLM. But I may have left some wrong usage.

    Still, none of that has anything to do with large language models. [...]

    But they have with automation and AI.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 1 point 1 week ago* (last edited 1 week ago) (2 children)

    I meant the title here on Lemmy... But I mean the entire article is talking about everything from image filters, autocomplete, to drafting code and LLMs. It also contrives some percentages, tries to differentiate between the different forms of AI... And the ends in a conclusion which again lumps everything together. We just talked about how background removal tools are acceptable and now we're generating scaffolding with LLMs. Those use-cases of AI aren't the same! We just tried to establish that...

    (Also, I don't think inventing definitions for words is of any help. LLMs are LLMs. Automation is automation. And machines are machines, less so entities with agency or "intelligence".)

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] [S] 2 points 1 week ago (1 child)

    I meant the title here on Lemmy [...]

    Here on the piefed (or lemmy how you prefer), I'm using the LLM usage and asking the quesiton of: "it's ok to call anything LLM touched as slop", and more important, who benefits from this view.

    (Also, I don’t think inventing definitions for words is of any help. LLMs are LLMs. Automation is automation. And machines are machines, less so entities with agency or “intelligence”.)

    Inventing? I think slop right now is binary and means a low level of quality. The problem is, if LLM touches anything, it's getting the tag of slop. And (ignoring the other problems of the LLM), I don't think that just by touching it, it get slop immediately.

    I'm not denying the other issues with LLM, and I'm fairly critical and vocal about it.

    But calling everything slop isn't helping, worse it's benefiting a very small group of people.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 2 points 1 week ago* (last edited 1 week ago)

    Sure. I can agree with that. Though I think it is a bit besides the point. The real problematic part isn't how people call something or if they're agitated and oversimplify things. I mean that's not great either. But I think far more interesting issues are de-skilling, changing the industry for the worse, open-source projects now hooked on AI subscriptions by nasty companies... People letting their coding-agent draft something and they don't understand it's now not them anymore who constructed the basis of the project, chose the tooling and made the important design decisions... Or the same with drafting a book or letter. It seems like that's a simple and tiny step... But it's the most important one. Some people are afraid the next generation of people will sound like ChatGPT, start every dialogue with a reaffirmation and say lots of "it's not X, it's Y" because that's what they grew up with...

    Also: I see far too much slop these days. We get several people each day advertising for shiny new projects they "wrote" (or they let their bot do it) and turns out a few of them actually compile and 1 in 10 doesn't get abandoned after a month... It's just way too much spam to me. And this isn't some rare exception... A lot of stuff currently gravitates towards that. Everything gets flooded with more stuff and everything is about attention. And it regularly isn't following the open-source spirit either. I can't contribute because the codebase is a huge mess, garnished with a massive amount of redundant documentation but all the important stuff is missing. And even if I do there's no one reading my PRs anyway... Which is kind of a letdown.

    And then every company tries to replace quality with quantity. I get half-assed products with weird quirks, art which isn't art, "articles" with just incoherent ramblings...

    I mean sure, I know there's like 10 people out there who do useful stuff with AI. I happen to use some of that. But still... That's far and in between. If you call something slop if it reeks of slop, you're kinda right 99% of the time. At least in my experience. And it usually bears consequences as well.

    But calling everything slop isn’t helping, worse it’s benefiting a very small group of people.

    Yeah, I think pretty much none of what we do is helping in any way. It's a decision of some big companies and powerful people to flood the world with "AI", fake stuff and whatever they like. And it's a trillion dollar business. Only thing that'd help is for example if the European Union mandates watermarking or forces people like Elon Musk to stop their nude generators... It's not really something we can do. And even in the realm of computer programming, it's really, really hard to convince management to adopt any kind of sane approach towards LLMs or quality of the product.

    I guess what I'm trying to say is: I understand names are important and how we call stuff. But it's even more important what's actually happening out there, right now. And most of it isn't really concerned with how I for example handle it. All the bad stuff is pretty much forced upon me. And there's just an uncomfortable amount of it. And it doesn't even matter if I call your 22% coding-agent drafted project slop or not... What difference does it make? It's still the same thing.

  • permalink
  • fedilink
  • source
  • parent