I’m kind of partial to the brain on drugs metaphor with an egg and a frying pan. Brain, drugs, brain on drugs.
This is IF, this is AI, this is your IF on AI. Maybe an image of a yolky egg cracked and opened over some classic piece of fiction in a frying pan.
Not many people are yet in the camp that AI usage has all the hallmarks of drug usage with all the dopamine delivering power where more is needed to get the same effect. I wish we had a topic for that. I predict eventually there will be a 12 step program for people to get off of AI.
LLMs, in particular Anthropic’s Mythos model family, is currently significantly revolutionizing the life sciences industry, from the discovery of new medicine to cancer research and rare diseases.
Got any links? I haven’t heard about this and a quick search didn’t turn up anything. (It should go without saying that I’m looking for news from people other than Anthropic, because they’re a bunch of liars.)
Having AI generate prose and an author saying, “This looks good!” is still AI-generated content.
This. Something being good doesn’t stop it from being slop.
You can take offal and grind it up and turn it into a meaty slop, add some spices and call it a meal. Some people will find it delicious — lots of Hawaiians love actual SPAM. That doesn’t stop it from being a reconstituted meat product, meat slop in a can.
Similarly with the email kind of spam, whether you’re interested in the product, whether the email is attractively laid out, whether the sender is legitimate, none of that stops it from being spam.
All those slop AI images on social media are something that someone looked at, and thought “this is good, I’ll post it”. They’re still slop.
If it’s content produced by extruding chewed up stolen human creativity using a generative AI system, it’s slop. That’s what people want labeled. If you want to build a game using slop, go ahead and build it, all we’re asking is that you label it, just like you might label a game with adult content, references to self-harm, and so on.
There’s possibly a conversation worth having about whether people should additionally disclose use of AI to generate ideas or plot. I quite like the “levels of AI use” idea.
But let’s be honest here, the continued “but what about” attempts to justify not labeling slop are because the people generating it know that the majority of people don’t want to consume it. You think the majority are wrong and it’s perfectly good? Fine, label it and let people make their own decisions. It’s the refusal to label it that has people proposing “No slop inside” labels.
If I had spent my youth writing assembly code by hand, I suppose I’d be saying the same thing about automating the assembly-writing part as I am about automating the identifier-typing, parser-writing, and code-formatting parts.
I did and I’m not. (When I started I didn’t even have an assembler.)
It’s muddying the waters to try to claim that using current generative AI tools is like using a compiler. Compilers are deterministic. They don’t use insane amounts of energy. Nobody is being forced to use them. People don’t try to hide using them. You don’t have to pay an unethical corporation every time you compile a piece of code. And in particular, compilers don’t use stolen code in their output. So compiler use is very different from the way AI code generators are currently being pushed.
An 18-wheel truck is not a bicycle, even though both are wheeled vehicles used to get stuff from point A to point B. It’s perfectly valid to say that people shouldn’t use an 18-wheeler to go to the grocery store and buy a carton of milk, but that it’s fine to use a bike to make the journey. Similarly, it’s perfectly valid to think it’s OK to use a compiler but not OK to use gen AI as currently exists.
Now, if you can find me a deterministic LLM trained on licensed code that I can reasonably run locally, at that point I might conclude that it’s like using a compiler or assembler. If it’s useful I might even use it.
If I could locally run a LLM trained 100% on just my code and only my code, I would play around with it. The danger for me is the assumption that my training data, my code, is bug free in all aspects: design and implementation. I wonder what it could make. I don’t know the first thing about training an AI model and then somehow using it. I don’t really want to know and prefer this head in the sand moment while it lasts. The future will come for me eventually but I can resist it today.
Pro-AI people are like “I use it. I like it. Shut up.” (At least the people who are using it very uncritically.) Then why do we have to argue and defend AI-Freedom endlessly?
Yep. I use local LLMs for search. I’ve trained neural networks and played with them for fun. I’ve used generative AI to remove an annoying child who photobombed a picture I took. It’s not the technology that’s the problem.
AI will not be going away, nor do I want it to. So we challenge it’s usage constantly, to ensure a message perseveres that: reliance on AI can reduce cognitive ability and AI is not a replacement for creativity. For me, it’s about a message for the next generation.
Middle-aged people using AI doesn’t bother me. They have fully formed frontal lobes and have graduated (with honours) from the school of hard knocks. Younger people need to recognize the developmental and societal dangers associated with AI. AI is a powerful tool… and with great power comes great responsibility. We’ve already seen the damage extended screen time has created in the current crop of young minds and attention spans.
I repeat this “cautionary use of AI” sentiment because doing something the hard way is the best way to learn. AI is the path of least resistance and it’s seductive even to me. It’s important that young people develop skills (true skill takes years to develop), then they can augment their abilities with AI, if they want to. Young minds (under 25 years old) require a strong cognitive foundation, not a strong cognitive dependency.
I wonder what worthwhile skills a young person could develop by relying on AI for most of what they do? If someone can shed a positive light on that concern, I’d love to hear it.
Related to this convo: Wired just ran an article about the death of mid-level budget games – which is more about the current financial environment than it is about AI – but this quote leapt out at me (emphasis mine).
…gaming platforms like Steam—which indie and double-A games once depended on for discovery—have been flooded with low-quality dreck. According to the AI Transparency Index, more than 18,000 games on Steam now feature AI usage disclosures. “User trust in platform stores as a curatorial layer has been all but erased,” says North Cook. “When you feed people enough ads and slop, they just check out.”
I’m sure genAI has accelerated the process, but I’m pretty sure Steam was already full of dreck in the late 2010s, before the genAI boom. I think as soon as any storefront moves beyond human curation, this is pretty inevitable.
Yes. Steam and other platforms like it “have been flooded with low-quality dreck” long before generative AI became a thing: shovelware, asset flips, and so on.
I agree, though, I’m sure the problem is even worse now. In general I feel like the main problem with AI isn’t that it actually creates a lot of new problems but that it makes a lot of existing problems worse.
Not just Steam or gaming, going all the way back to the beginning of the web, the dark side of the digital revolution has always been that for every maverick genius that traditional publishing would never take a risk on who found an audience online there’s a hundred talentless hacks with overinflated, fragile egos posting stuff anyone with a shred of self reflection would be embarrassed to show off. Human-made slop has been a problem for over a quarter of a century… Of course, AI makes the slop problem magnitudes worse, with human-made slop, it still takes an appreciable amount of time and effort to produce, limiting the output of those who keep at it, providing a minimal barrier to even try, some of those fragile egos will give up if no one bites, and some of the folks who do keep at it will actually improve to something of decent quality… meanwhile, generating and posting AI slop is practically zero cost and effort, removing limits on rate of output, removing barriers to entry, models do get better, but the slop peddler isn’t the one driving model improvement, and the thousandth uncurated AI image from a given model is no better then the first where as even the most talentless human artist is probably going to see some improvement across a thousand drawings, and if one piece of slop gets no bites, the next piece is just a few clicks away… And unfortunately, the AI is good enough to trick some people, and you don’t need many bites that lead to people giving you money to profit on what is essentially zero cost, zero effort, unlimited output… If there wasn’t that financial incentive to profit off AI slop, I suspect much of the bad faith concealment of AI usage would evaporate… Of course, some of the anti-AI crusaders have folks who supplement their human effort with AI(such as the essayist that uses AI diagrams because their graphics skills suck and they have no budget for a illustrator or the non-native English speaker using AI to clean up their grammar) scared that being honest will get them digitally tarred-and-feathered, though of course, that’s talking the folks who see AI or something they think is AI and start spouting hate in the comments, not the folks ho just filter out anything tagged AI and move along.
And yeah, it sucks that so many people have such a low tolerance for nuance that granular, informative labeling systems often get shot down in favor of systems that summarize to the point of being effectively devoid of information.
Going all the way back to the beginning of forever. This is just Sturgeon’s Law: “ninety percent of everything is crap”, exacerbated by rapid fire automation. Now probably 99%.
This perspective strikes me as a somewhat adjacent to moral panic. The introduction of calculators didn’t end the study of mathematics, nor did spell checkers destroy writing. We all use those tools on a daily basis, but I can remember when parents used to lose their minds over students using calculators in math class.
There’s a time and a place to use tools. You could maybe make an argument that over-use of a tool (as with anything else) is not profitable.
I have some points against what you wrote. But, hey, do what you have to do. But how does this discussion fit in here? This thread is for people what are critical about slop, and who would like a label. This thread should not be the zillionth discussion pro and contra AI.
Apologies for taking it one step beyond. I do routinely see people talk about the perspective but it has no place in a forum talking about icons. I do like the idea of a frying pan, egg, and book as the start of a logo. As far as the tooling metaphor goes, there was an excellent counter point to it. Or that somehow the idea that people who scratched images on cave walls with burnt ends of sticks would incite the mob over the introduction of spray paint.
Yes, clearly we’re different people with different attitudes toward automation.
You’ve come up with a list of distinctions (well, sort of – most of those are dubious at best) but not explained why you think those distinctions are dispositive or relevant. Talk about muddying the waters!
The purpose of an analogy is not to imply that two things are identical in every respect, but to highlight the respects in which they are similar. In this case, the respect in which they’re similar is that they’re ways that machines produce code instead of humans. Your list of purported differences doesn’t change the fact that it’s dishonest to claim your code is “human-made” when it was in fact produced by a machine.
If you disagree, then by all means, please explain why “nobody … being forced to use” a compiler, or your opinion about the ethics of the corporations that make compilers vs. AI, or the amount of electricity used in datacenters, makes it acceptable to pass a compiler’s output off as your own work.
If you read my post more carefully, I think you’ll notice that I wasn’t talking about whether it’s “OK to use gen AI as currently exists”, I was specifically talking about using a label like “Human Made” to describe machine-written code.
I consider the C code I write, the assembly produced from it by Clang, and the machine code produced from that by the assembler and linker, to be broadly equivalent things. That’s why I’m willing to check only my C code into version control instead of the assembly or the binaries. I trust the compiler to produce equivalent machine code again whenever I need it. There’s no need to store the assembly or the machine code because it can be regenerated on demand in a reliable, trustworthy way.
I believe even most LLM advocates check the code produced by the LLM into version control instead of only the prompts. Case in point. To me, that indicates that the code produced by the LLM is not equivalent to the prompts fed into it; if it were, you would only need one form of that information, not two.
Since I consider the C code and the machine code produced from it to be broadly equivalent, I see no contradiction in referring to the binaries as “human-written” when it’s actually the C code that was written by a human. And I don’t think anyone else has expressed confusion about this point, either? When I see an effective, well-regarded repository for a ZILF-level project that only contains the prompts, with the author trusting the LLM to regenerate the entire codebase every time they want to use it, then I’ll believe that LLMs are in fact equivalent to compilers. But not until then.