site banner
PaperclipPerfector  1yr ago (text post)   3515 thread views

Small-Scale Question Sunday for June 18, 2023

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

2
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

https://meta.stackexchange.com/questions/389834/statement-from-so-june-5-2023-moderator-action

https://meta.stackexchange.com/questions/390106/moderation-strike-update-data-dumps-choosing-representatives-gpt-data-and-wh

There's some sort of mod strike happening on stackexchange too. Does anyone know the details beyond the histrionic statements?

“AI detection tools” really don’t work. There is no tool that can consistently detect GPT-4 generated text. The text that looks ‘obvious’ does so because it defaults to the lowest common denominator of professional writing, overuses adjectives etc, like a high school essay. This isn’t actually ‘obviously AI’ per se, it’s just that few people write in a high school essay style online.

But even one or two descriptive terms (‘in the style of a Vanity Fair article’, ‘concise; few adjectives’) in the prompt is usually enough to mitigate this entirely. The only possibility is that OpenAI and the other major providers hide markers that ‘AI detection’ tools can read. Even then, people would just run the output through text rewording tools or use other LLMs.

The smart approach is to ban low quality and low effort content, whether AI or human generated.

The smart approach is to ban low quality and low effort content, whether AI or human generated.

Speaking for StackOverflow specifically: when it comes to technical questions, there's no easy heuristic that a non-expert can use to distinguish between a low effort post and a good post.

Posts that look good on the surface can still be bad. You can supply a 50 line block of code that compiles and seems to work, but it can have subtle problems that won't present themselves until later on. Or you can write a post that would be perfectly good in another context, but it's the wrong approach to this particular problem because of X Y Z non-obvious reasons. And of course LLM-generated answers can have these sorts of problems.

The site is based on a relationship of trust: if the post looks good, and it's upvoted, and the user has a high amount of karma, then the post was probably written by a human expert who is aware of the sorts of pitfalls I mentioned and knows how to avoid them. If you just have a blanket policy of "yep, AI is 100% allowed, go nuts", then it starts to erode that relationship of trust. More non-experts come to the site who start posting AI-generated answers without understanding them, they accrue karma, it gets harder to distinguish the signal from the noise.

In the limit case, human experts get disincentivized from posting because, well why go through the effort of spending 30 minutes writing a detailed answer when the AI posters will post 5 different (equally long and detailed, but perhaps subtly incorrect) answers in that time and the OP will probably just accept one of those answers anyway. That's the fear.

Whether a total AI ban is actually enforceable is another question. But this is the reason why people would want a full AI ban, instead of just a "low effort post" ban.

More non-experts come to the site who start posting AI-generated answers without understanding them, they accrue karma, it gets harder to distinguish the signal from the noise.

How is this different than the classic StackExchange / Quora regular who knows nothing about the topic but spends 20 minutes on Wikipedia and then writes out an answer as an expert?

If it takes 1 minute with AI instead of 20 minutes manually, that seems significant enough to affect volume.

Obviously we don’t want people to do that either.

Exactly, they're both bad and both deserve to get the banhammer, the only difference is in volume/prolificacy.

I think it's as simple as: SE is positioned as a high-quality Q and A site above all else. In particular high-volume users and mods are particularly prideful about this and really, really hate low-effort stuff (plus, again, pride). SE rightfully banned chatGPT and AI content both in answering questions as well as (AFAIK) posing questions, both in generation as well as clarification. This lead to the site more or less continuing as before as well as a lot of bans and mod action to preserve this, coordinated with the site/network as a whole. They then did an about-face, threw the gates open to AI content, and have forbidden mods for banning people (or even doing more regular moderation) for AI content-related stuff. The manner in which this was done feels to some like shoving it down their throat.

So the mods and hyper-active users are unhappy. It should be mentioned that although SE feels like a Wikipedia-type site, and operates similarly in some ways, it's NOT. It's owned by a for-profit ed-tech company.

I think a blanket ban on AI content is stupid.

As long as it's clearly disclosed, all that matters is whether or not the answer is correct, which can be independently verified and better quality answers encouraged through voting.

I've found GPT-4 to be more than competent in medicine, and to the limited extent of my ability to judge, the same in coding and maths when I use it for autodidactic purposes.

The mods are choosing the wrong hill to die on.

As long as it's clearly disclosed, all that matters is whether or not the answer is correct, which can be independently verified and better quality answers encouraged through voting.

The problem with AI-generated text is that we can no longer use eloquence and details as a proxy for correctness. If I write a thousand words to answer a SE question, I probably know what I'm doing. If I write ten words, I can either be correct or not, but it's easier to verify. Now anyone can write a thousand-word reply to a question that will sound very authoritative, but might be as wrong as any of the short ones.

Fair enough, but at the end of the day I still think that GPT-4 is more competent than the median user, and that excluding it outright is a mistake. I think clear disclaimers more than suffice, and worst comes to worse we'll just be asking other AI assistants to spot-check them for us sooner rather than later.

The catch is that top stackoverflow posts probably aren’t the median user.

Wow, I wasn't aware of any of this. Despite being a developer, I don't really engage much with SE. If I find an answer to something there on a Google search, I'll usually prefer it over random blogs, but I don't have much interest in writing questions or answers there.

It seems to me that SE is designed to maximize karma given for fast answers to simple newbie questions, so that's where the great majority of the energy of it goes to. IME, it's very rare to get a real answer to an actual hard question. If I manage to find an unanswered hard question that I know the answer to, I'm willing to post mine, but I don't see much point in asking questions there.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats