Substack implemented Pangram. It allows user to scan notes and articles for AI. It's absurdly accurate and has caused a lot of debate on the platform. Moreover, the tool has exposed salient accounts heavily AI-assisted, and in some cases, completely AI-generated. Did they go too far, or not far enough?
>>25419598>It's absurdly accurate[Doubt]
>>25419598>It's absurdly accurateHow did they measure this?
Nobody outside of substack cares about substack. Please stay in your containment site.
Why does /lit/ shill Substack?
>>25419650very little engagement here is of any value or depthI don't like little Substack fiefdoms either, but at least it's possible to find some writing about modern topics and commentary by people who can at least occasionally stay on-topic, rather than those who engage in endless bad-faith digressions
>>25419613>>25419625It is accurate, because most of the people using it are conmen, frauds, and losers. Meaning they have no writing ability at all and publish exactly what AI spits out. There's a difference if you edit with AI. Pic related.>>25419630>>25419650You will graduate from /lit/ one day.
>>25419692That's great, but that doesn't tell me how they measured it. I'm genuinely curious, as I've written my own neural networks in grad school.
>>25419598I read through the entire thread before realizing you were talking about substack and not stack exchange. For stack exchange it's a pure positive (though I don't know if they've implemented something like pangram). For substack I'd say it's the same. Only issue I can see is that AI will continue to get better at imitating humans, so these tools will get less and less effective. AI is such shiny new technology at that the arms races involved haven't even really begun to play out, but they will in the future.
>>25419598So say it shows your essay as AI assisted, meaning you wrote it and then stuffed it into a robot for it to line edit. What then? Is this supposed to be for "transparency"? Because it's obvious then that, like every other platform that generously lets you label your own work as "AI", anything that is labeled as such will be seen as inferior, regardless of the actual output of the work itself. This is incredibly common online, especially in certain sub sections of communities, to the point where you could show someone two writing pieces of equal caliber, fairly even in quality, with the only meaningful difference being a label that says "AI", and they will immediately shrivel their nose at the sight, as though someone had just rolled out a shit on their desk.So what is this going to do? For a small group, it might make them just not use AI altogether, which may be of some benefit to those who have certain ideological opinions on AI. But for the majority, they'll likely just work harder to conceal the AI, whether it be for editing or outright fabrication. And with the incremental improvement of AI, this becomes more and more feasible, as others have already said here.So I ask again, what is the point of this? In other words, if you read a piece and can't tell that it's AI without an algorithm labeling it for you, does it matter that it's AI? Especially for editing, where it isn't actually making anything, rather just refining what is already there into a more standard house style. I personally think not, especially when most users of AI are already very obviously using it, in images and in other unrefined ways. Let the reader decide instead of a machine
>>25419692>>25420586And just to quickly expand on it, look at the color choice on the detector. You have three categories: AI, AI-Assisted, and Human. Human is coated in a bright green, while AI is darkened into a shadowy grey. AI-assisted, oddly enough, leans more onto the AI side of the coloring, merely being a lighter grey instead of, say, a yellow, or something else. I'm not saying that Pangram is withholding a subconscious conspiracy based on three colors, but the choices reflect a decision made somewhere.And not to dig too deep into it, but what value does knowing something is "15% AI" have? Is that too much AI? Just enough? How does it compare to 19% Ai-Assisted? It doesn't say, merely giving the numbers, a quick blurb, and leaving the reader to decide. Except this isn't REALLY giving the reader a neutral decision. Most factors point one way here, and again, the label(s?) itself makes the decision tilted.
>>25420586>Is this supposed to be for "transparency"?Yes.>anything that is labeled as such will be seen as inferior, regardless of the actual output of the work itself. Exactly what it's intended to do. Actual human effort, using your own words, should be valued over a machine's. >But for the majority, they'll likely just work harder to conceal the AI,I don't think that's a problem. They're already frauds. They will have to find workarounds when investigative tools sniff them out. It will require more effort to keep up the ruse. >So I ask again, what is the point of this?The point is to draw a clear line in the sand. Substack is a platform for human connection. AI writing threatens that. You seem upset that there is a preference given to human writing over AI. I don't know why this is baffling to you, it's a recent phenomenon that we haven't fully work through, though tech overlords are pushing to make this our everyday. I'll ask you, what amount of AI do you use in your writing? In my research, the only ones complaining about it are the ones who would be nothing without it.
>>25419598I'll care more if this makes kicking foid newsletters off the platform easier.
>>25420667To answer the bottom up front, I do use it for editing when it's effective. It's useful for brainstorming or debating against, or fixing grammar mistakes, though I don't think you really care about that.So you present it as transparently functioning in favor of preserving human authenticity on the platform. That's certainly a coherent stance, but it's a pointedly ideological/opinionated one, and I don't think you'd disagree on that.But above all else I think you're taking your idea of substack and in a sense mandating it for others to follow. Substack, at it's core, is a platform for sharing writing, and a platform for people to read writing from people they like. Note that it isn't a platform specifically for, say, people who write English, or for people who write poetry, and so on. It's an intentionally broad platform for everyone. So why force others to have to concede to your specific vision? If someone wants to read AI slop from their favorite writer, let them. And before you say "well they can, nothing has changed", note that we both have already agreed on Pangram functioning as a values machine that is inherently not neutral.>I don't think that's a problem. They're already frauds. They will have to find workarounds when investigative tools sniff them out. It will require more effort to keep up the ruse.This is oddly antagonistic. But apart from that, you still haven't resolved the issue at hand. It's already harder for these systems to "sniff out frauds" then it was 2 years ago. What will it look like in another 2 years? Is it sustainable?Ultimately AI writing, in all it's forms, ranging from the most braindead to the most eloquent, are not going away, and we indeed haven't worked through the problem completely.>though tech overlords are pushing to make this our everyday.It'll be increasingly hard to shrug against the tide, and I have no doubt you're aware of that. Writing will have to have this conversation properly, but we can't get there if half of the writers are terrified of the concept and the other half are livid at it
The last thing I want is to actually research, write, and edit an article with the help of actual friends/family in editing and have it get mislabeled as AI/AI edited. Is there no way to mitigate this or know otherwise?
>>25419598that's fine however, its more about content than form, I'd argue. if a five paragraph essay dismantles nominalism and someone discovers its AI generated it more or less says more about humankind's nominalist counterarguments than AI understanding that substances exist outside mental perception.
I just tried it on my own bs and it wont scan it. lame
>>25421060>Substack, at it's core, is a platform for sharing writing, and a platform for people to read writing from people they like. Note that it isn't a platform specifically for, say, people who write English, or for people who write poetry, and so on. It's an intentionally broad platform for everyone.Evidently the substack staff disagree. That's why they have an AI detector and ban smut. https://substack.com/content
>>25420603>but the choices reflect a decision made somewherethe more categories included the less accurate it will be
>>25421060>To answer the bottom up front, I do use it for editing when it's effective. It's useful for brainstorming or debating against, or fixing grammar mistakes, though I don't think you really care about that.Better off hiring an editor or finding another writer to collaborate. There are also non-AI editing tools like https://www.grammarcheck.net/editor for spelling and grammar check. If you are making substantial edits to your work to the point a program such as Pangram notices, then you've not only ceded your voice and your intellectual property to a machine, you've made slop. >So you present it as transparently functioning in favor of preserving human authenticity on the platform. That's certainly a coherent stance, but it's a pointedly ideological/opinionated one, and I don't think you'd disagree on that.Let's get back to first principles, because we're arguing like certain things have been decided and agreed upon, when they have not.Concerning the creative process, there's the idea, the act of creation, and the result. The idea always has a human origin. I argue the idea is the catalyst, and ideas that do not go through the creative process are not ideas at all. I digress. Up until the last few years, the creation process and result have also been exclusively human. For the writer, the result, the language, that final written word, is the idea.They do not exist independently from another one.When using AI, the creative process ends before creation. The only thing the non-writer has done, and is truly responsible for, is the idea and the prompt. Everything that comes out of the machine, the language, belongs to the machine, therefore it is no longer the non-writer's idea. That non-writer may take that text and edit it and choose what to publish, but they can only call themselves a publisher. Any attempts to obfuscate the origin of the language denigrates them to a fraud. You will find this is extraordinarily common among non-writers, and bad writers turned non-writers. I can go farther and argue it is an entirely separate process from writing, as the idea culminates in a prompt and not the act of writing, but I think the above is sufficient. >>25421060>It'll be increasingly hard to shrug against the tide, and I have no doubt you're aware of that. Writers will adapt. There will be an increased demand for transparency, access, experience, courage, and iconoclasm, likely in all fields, even mathematics. These are things machines cannot provide, and after a few years of unoriginal recursive thinking we'll be back to humans. Machines will then again try to catch up by plagiarizing human work, and the process repeats. Humanity is essential for AI. AI is not essential for humanity.
man i hope the ai bubble pops soon and we can forgot about this shit like crypto, yeah scammers and druggies still use crypto, but no one else even remembers it was a thing. hopefully google getting punished by the market for going cash flow negative to build absurd data centers no one wants and have no path to roi is the beginning of the end.
>>25421837Probably because it's full of false positives and labels pre-AI content as AI
bump
>>25423207You're dumb and you don't know the first thing about AI.
>>25419692>It is accurate, because most of the people using it are conmen, frauds, and losers.The latter does not logically entail the former here, assuming that you are even correct, and even then you add the "most" qualifier for some reason. Man, so many OPs on /lit/ are just straight-up retarded
I hate it when dumb subhumans come on here to scream at us whenever they lose their shit in an argument on twitter or whatever. These people should be gelded in a just and decent society.
>>25425121Brother, you had all this time and that's the response you managed to drum up? I'm feeling good about my competition. You lack mental fortitude and it's only going to get worse.
