-21

We’ve got a dedicated team working on adding GenAI to Stack Overflow and Stack Overflow for Teams and will have some exciting news to share this summer. --- Prashanth Chandrasekar, April 2023.

The introduction of GenAI to Stack Exchange is not going away, but we may still be able to influence how it's developed. There are many posts about GenAI at various sites, and my understanding is that the primary objection is to mindless copy/pasted answers which could be totally wrong, are difficult to moderate, are plagiarized, and waste everyone's time, and not Stack Exchange using (gen)AI per se. There potentially are uses for (gen)AI that facilitate, and don't obstruct, Q&A at Stack Exchange.

Question: How can Stack Exchange use AI, and especially GenAI, to facilitate Q&A?

I posted some ideas in a previous answer, but I think we should brainstorm this more systematically as a kind of "feature request request":

  1. One answer = one proposal (so people can vote).
  2. Please don't answer "I don't want GenAI" here (there are plenty of other posts where you can do this).

I'm not looking for well-polished proposals here, but rather constructive ideas we can consider in depth in subsequent posts, along with some basic "sanity checks".

Improve this question
13
  • 3
    define "facilitate Q&A". do you mean content moderation in general?... everything?
    – starball
    Jul 24 at 0:38
  • 1
    Literally any AI-based Stack Exchange feature that would help us Q&A. I could imagine genAI helping with moderation, such as by reading two potential duplicate questions and summarizing what they have in common, and how they differ.
    – Rebecca J. Stones
    Jul 24 at 0:44
  • 17
    Why do you believe that GenAI - generative AI, or the use of algorithms to synthesize new content - has any place in Q&A here? I'm not dismissing the application of learning or analysis algorithms in general, but the correct answer is "it doesn't belong here" since it is impossible for AI to generate new content that is anywhere near appropriate for the network.
    – Thomas Owens
    Jul 24 at 1:10
  • 6
    The quote is from April, and it talks about a big announcement come summer. They've been working at something for a minimumn of four months. "... the alternative is for Stack Exchange to implement GenAI features with limited community input." They're already doing that.
    – JonathanZ
    Jul 24 at 1:19
  • 2
    @ThomasOwens Strawman? I'm looking for ideas for AI-based features that help human users be more productive.
    – Rebecca J. Stones
    Jul 24 at 1:23
  • 8
    GenAI is just a stochastic parrot where no matter the size of the data set that feeds it will always be a parrot. The recurring mistake is to want to treat GenAI like an Artificial General Intelligence (AGI).
    – Augusto Vasques
    Jul 24 at 6:11
  • 23
    I don't understand why are people trying to bend over backwards just to find possible use cases of GenAI on the platform? While there are some possibilities, mostly around searching and finding duplicates, even in those areas there are more appropriate technologies. Just because company is hell bent on using it, we don't have to indulge them in their misplaced endeavors. Currently they are in "we have a hammer and everything looks like an nail" stage and until they snap out of it, it will be impossible to have any reasonable discussion around the topic.
    – Resistance Is Futile
    Jul 24 at 8:26
  • 3
    It is not that we cannot see possible uses or we just plainly hate GenAI, but because the second we mention its possible use cases, the company will stop looking for other, better solutions to the problem.
    – Resistance Is Futile
    Jul 24 at 8:28
  • 1
    @ThomasOwens "it is impossible for AI to generate new content that is anywhere near appropriate for the network" incorrect, see meta.stackexchange.com/a/387577/178179
    – Franck Dernoncourt
    Jul 24 at 17:22
  • 10
    @FranckDernoncourt It seems like you're under the mistaken impression that because an answer is correct it is somewhere near appropriate for the network. That is incorrect. Other people come to the network because they trust that they are getting content written by humans that share human knowledge and human experiences relevant to the problem at hand. Algorithmically generated content has no place here. Even if someone claimed to review it, there's no way to determine if that is true. So the only acceptable recourse is to prohibit such content to ensure that only human-written content stays.
    – Thomas Owens
    Jul 24 at 17:46
  • 1
    @ThomasOwens We have different views on appropriateness. Yours made your quoted statement a tautology.
    – Franck Dernoncourt
    Jul 24 at 17:50
  • @ThomasOwens "Other people come to the network ..." We should investigate this more. I'm here for the answers. I do not really care much about them being produced by a machine or a human. So far I assumed that humans did the work but if I could only get an answer by a machine, I would probably take it too. If I had the choice ... I don't know which one I would prefer, I have to think about that.
    – NoDataDumpNoContribution
    Jul 24 at 20:30
  • 2
    Honestly, I don't think there's any point in discussing AI answers until we take a large chunk out of the blindly copy/pasting genAI problem.
    – Rebecca J. Stones
    Jul 25 at 12:33

13 Answers 13

Reset to default
16

Given what I've seen of GPTs so far, I think it's best to set aside trying to find ways to integrate AI into the site entirely for the time being, at least until the strike is over and SE begins to engage with the community on its GPT initiatives in an honest manner.

If AI was as good as the hype and had so much to offer, surely the benefits would be apparent by now, since there's no shortage of people trying to find uses for it.

GPTs have been shown to be unreliable across the board at writing accurate answers, detecting spam and improving questions. It's even been suggested that GPTs replace moderators and a prompt engineering site has been forced open without precedent. Outside of SE, OpenAI has shut down its GPT classifier tool due to poor accuracy, indicating that that use case won't be possible on SO for some time.

What value would the SE network offer if answers, questions and moderation are all powered by AI? The point of the site is exposing access to and organizing human expertise, not exposing access to and organizing AI "expertise" (https://chat.openai.com already achieves this nicely).

Why is there such a rush to integrate GPTs into the site? We should be adopting initiatives that are proven to work, uncontroversial and stable--it's hard enough to get the basics solid, like proper onboarding and guidance for new users, or enabling moderators to do their job uninhibited. The fundamental SE model works fine, which is why we're here. SE should be working to get back to that and refine it, rather than shoehorning in a novel technology that undermines the trust that represents the basic value proposition of the network.

Improve this answer
6
  • 3
    Not an answer...
    – Rebecca J. Stones
    Jul 24 at 23:37
  • 14
    @RebeccaJ.Stones Sure it is. Your question is "How can Stack Exchange use AI, and especially GenAI, to facilitate Q&A?" and my answer is "they shouldn't". You're also looking for basic sanity checks, and this is one. The network is already pushing AI plenty hard at the expense of the community--no need to give them more ideas (especially since everything in this thread so far has been suggested before).
    – ggorlen
    Jul 24 at 23:39
  • 12
    @RebeccaJ.Stones Seems like the rest of that sentence would read "that I wanted to see."
    – Dan Mašek
    Jul 24 at 23:53
  • 4
    Relevant discussion: Does Stack Exchange allow for answers which question the validity or stance of the original question? (cc @RebeccaJ.Stones)
    – Sonic the Anonymous Hedgehog
    Jul 25 at 0:20
  • 8
    @RebeccaJ.Stones It's a frame challenge answer. And there is something to it. The recent "code formatting" assistant experiment was a complete disaster, not because there weren't enough guardrails but because it's an inherently bad idea to completely transform a whole post by GenAI. This results in unpredictable behavior. How could they even think about it is beyond me. However, not giving the company ideas sounds weak. The freedom of ideas is one of the corner stones of meta. We should be free to think about it and the company will come up with plenty ideas on their own anyway very likely.
    – NoDataDumpNoContribution
    Jul 25 at 6:03
  • 1
    @DanMašek Seems like the rest of that sentence should read "for the 1000th time".
    – Rebecca J. Stones
    Jul 25 at 12:15
13

I think there's potential for ML to be useful in improving the quality of the knowledge base, but I don't think focusing on GenAI is useful. And I'd hope to see experimentation being done carefully, and without being muddied by primary goals of increasing engagement or getting more content.

  • Detecting noisy phrases in the post drafting / editing tools and giving tips about what is considered as noise and should be removed would be nice.

  • There's already been work done to try and improve related posts using ML. Work to improve the possible duplicate section of the Ask Question process could be useful.

  • Detecting comments that should never have been posted would be nice. Ex. giving the commenter a "This comment might be considered noise. Are you sure it is not noise according to /help/privileges/comment?" could be useful.

  • I've been flagging tons of non-answers on SO recently- pretty regularly running out of my 100 daily post flags on weekdays. Dharman's bot and Natty could be inspirations for some ML work...

But basic pattern matching could also go a long way, and in some places, already does, and I'd prefer to see that explored first for simplicity and cost reasons.

I'm still personally convinced that doing better promotion of the Help Center and trying to give better JIT guidance is the way to go.

Improve this answer
0

AI can be used to do translation. I've used it many times already on SE, typically to translate non-English into English. That could also help avoid fragmenting SO into different languages (ru, jp, es, etc). LLM have some abilities to do translation.

Improve this answer
0

Creating a kind of duplicate report.

E.g., a user could input two URLs into a form, and an LLM could summarize the differences between the questions, and give its opinion as to whether or not their duplicates. These reports could be attached to the posts' timelines.

Just now, I gave Assistant the links to two questions on the Chinese Language Stack Exchange (the second was closed as a duplicate of the first):

I asked for a summary of their differences, and if it thought they were duplicates. To my surprise, it wrote...

Based on the URLs you provided, the two links are not duplicates. The first link discusses why the verb "有" (yǒu) requires "没" (méi) instead of "不" (bù) to form negative sentences, while the second link discusses when to use "没" (méi) and "不" (bù) before verbs in sentences.

The AI was right---they're not duplicates! The first is an advanced question about etymology, the second is a beginner question about grammar. It turns out the second one should have been closed as a duplicate of another question. So I went and fixed it as best I could.

Improve this answer
2
  • 2
    Summarizing is something GenAI does fairly well. It might be interesting to generate summaries of long content threads or meta discussions. I'm a bit reluctant to have it summarize answers for questions, because it might suppress voting or have trouble capturing all the nuance. Maybe it could generate an "index" of answers that would allow a reader to jump to them based on some "teaser" of their content.
    – ColleenV
    Jul 25 at 15:15
  • related: Preview of Search and Question-Asking Powered by GenAI
    – starball
    Jul 29 at 9:02
0

A genAI TL;DR summarizer.

Some questions (especially here on meta.SE) get a formidable number of answers and comments, and the majority of users may simply not have the time nor interest to read them all. Moreover, not all contributions are as friendly as they could be; a fair number of comments are aimed at winning arguments and making people look foolish, rather than bringing new ideas to the table. Thus:

  • Users may drop out of discussions, leading to survivorship bias; it's quite possible more extreme, unfriendly viewpoints ("this idea will totally ruin Stack Exchange") lead to users with moderate, nuanced viewpoints ("here are the pros and cons of this idea from my perspective") choosing not to post.

  • Users may repeat discussions because they haven't read everything, thereby adding even more work to someone who wants to read everything, and wasting time with via redundancy.

Thus, genAI could be used for summarizing posts with many answers and comments, to save time and remove the unpleasantness.

Improve this answer
1
  • 1
    I would upvote save for the “remove unpleasantness”. AI can’t be trusted to sift out what one user may find to be an unpleasant tone. If something is disruptive to the site, it should be flagged for removal not filtered out.
    – ColleenV
    Jul 31 at 10:32
-1

AI can check for plagiarism, both for human and AI-generated answers. GenAI gets blamed for plagiarism, but humans do it too. But few people cared about plagiarism before GenAI started to get popular here.

Improve this answer
3
  • 8
    "few people cared about plagiarism" [citation needed] ? | There were solutions for this before GenAI became a thing, right? In what sense does it do the job better?
    – Dan Mašek
    Jul 25 at 21:59
  • 2
    @DanMašek "There were solutions for this before GenAI became a thing, right?" yes. Not specific to GenAI
    – Franck Dernoncourt
    Jul 25 at 22:02
  • 5
    Mods took plagiarism very seriously way before the GenAI plague, and there is a dedicated bot called Guttenberg (named after a German politician who was involved in a plagiarism scandal) to help look for plagiarism. Even with the bot, the tooling for finding plagiarism is rather inadequate; I'm sure better solutions would be very welcome.
    – tripleee
    Jul 29 at 8:45
-2

I'll get the ball rolling...

Automatically replace the placeholder enter image description here with useful text.

This feature already exists at Facebook:

Auto-generating descriptions of images

Scalable annotations have also helped us build automatic alt text (AAT) within the core Facebook app. The technology generates text descriptions of photos that help the visually impaired. With AAT, people can use screen readers to hear descriptions of content images. We used Halo to post-process categories generated from our classification models. Creating this labeled data for AAT models helps us make Facebook more accessible.

It may also help Stack Exchange posts be more searchable (via Google).

Here's an example of GPT-4 from this Imgur:

enter image description here

Improve this answer
13
  • 1
    Didn't DV, but would that generation technique turn images of code into code? If so, that's definitely a feature we don't want.
    – Daedalus
    Jul 24 at 0:27
  • 6
    I'm just suggesting the concept; the details would need fleshing out. But I feel even "screenshot of code" would be more useful than "enter image description here".
    – Rebecca J. Stones
    Jul 24 at 0:32
  • 4
    As someone who occasionally uses image descriptions (though doesn't rely on them): I have yet to see an AI tool that produces even half-way passable alt text. (Here's how it should be done: WebAIM: Alternative Text – see how much of that is just outside the ability of current AI‽) All this proposal accomplishes is preventing actual competent people from finding posts without alt text with SE search and fixing them properly. (The context of Stack Exchange is very different to Facebook, where that is better than nothing.)
    – wizzwizz4
    Jul 24 at 1:07
  • 2
    No, that text is not supposed to be a description of what’s in the image. It’s supposed to be the same information that image would convey to a sighted person. AI can’t write good descriptions of images because it has no idea why the author included them. “A line graph with a y axis from 0 to 100 and an x axis of years 2009 to 2016” may be accurate, but it’s not any more helpful than “enter image description here”. Faux accessibility is an obstacle to actual accessibility.
    – ColleenV
    Jul 24 at 1:18
  • 1
    My screenreader also automatically provides image descriptions and it has one important feature: the ability to turn that crap off. It was extremely confusing (even though I could see the image with my own eyes!). There's also apparently a similar feature in Chrome but I've never used it (except Google Lens to transcribe). Do you really think SE can make a better feature than those big hitters? (Of course, the best image descriptions are written by humans, who understand the context it appears in and what it needs to convey.)
    – Laurel
    Jul 24 at 1:31
  • 3
    That would require some form of image recognition which is a whole other can of worms that is too expensive to even touch on a SE-scale.
    – Cerbrus
    Jul 24 at 5:46
  • 12
    Likely scenario: this suggestion is in place, SE is proud of it. Someone uploads picture of dark skinned person, e.g. movie screenshot on Movies.SE and the AI decided to put "Picture of an ape" as the default description. Author doesn't notice/don't care, and the next day, media is full with "Stack Exchange racist" shouts. And yes, that would totally be SE fault, for using that terrible feature.
    – Shadow Wizard Is Sad And Angry
    Jul 24 at 6:36
  • 1
    "Automatically replace the placeholder enter image description here with useful text." aka image captioning
    – Franck Dernoncourt
    Jul 24 at 20:27
  • 2
    A verbose general observation of everything about a graph is not good ALT text. The example still doesn’t explain why the author included that graph in their post. Maybe it’s because the line is blue. You’ve said “here is an example” and then not filled out the image description for it. If I can’t see that image, I have no idea how good or bad your example is… how would ChatGPT caption that example image in the context of your post? Would it write a paragraph explaining “This is an example of AI captioning an image” or would it explain the point you tried to make by including it?
    – ColleenV
    Jul 25 at 9:19
  • 2
    After thinking about it a bit, I also want to point out that the prompt is “analyze this graph” not “explain this picture”. That’s cheating a little bit. An auto-captioning system wouldn’t get that nudge. GenAI is exciting technology, but I don’t think trying to make everything into a nail because we’ve got a really cool hammer is productive.
    – ColleenV
    Jul 25 at 10:50
  • 1
    I just meant it as a kind of "proof of concept"---AI can do much better than it used to "dog (80% confidence)". In any case, the constructive criticisms expressed in the comments here are quite valid, and if Stack Exchange were to ever consider this option, I think the feedback would be quite valuable. Or maybe it's already enough to shelve the idea.
    – Rebecca J. Stones
    Jul 25 at 12:42
  • 4
    I didn't downvote the discussion, because I do think it is a valuable one to have, but I think that the technology isn't mature enough yet to do anything automated for users who don't understand it. GenAI has tremendous potential, but right now, it's only useful as an assistant for users who understand its limitations and a bit about how it works. It could be trained to help someone knowledgeable about GenAI to write ALT text, but it would probably be better just to prompt the post's author to answer some form of the question "What does this image tell readers?"
    – ColleenV
    Jul 25 at 15:05
  • I would be very curious how ChatGPT would caption that image if given your entire answer as context and prompted to write ALT text for the image. One chat bot "summarizer" explains Alt text is meant to convey the "why" of the image as it relates to the content of a document or webpage. so I believe the knowledge of ALT text for accessibility is in the model somewhere, I'm just skeptical a genAI-based tool can put all the pieces together. I bet it could explain a meme GIF easily though. The captioning might not be useful for SO, but could be for social media.
    – ColleenV
    Jul 26 at 18:10
-2

Many sites are currently undergoing site design. AI-generated images could potentially be used to generate mock site designs. And indeed, with very little human effort, generate multiple mock site designs so that users can point out elements they like and dislike (rather than think of them off the top of their head, or wait until Stack Exchange designers have posted their mockups before saying "nah, don't do that").

For Chinese.SE, I whipped up these mock site designs using leonardo.ai:

Obviously these can't be used as is, but they could help promote discussion, and get those creative juices flowing.

It would be better if the genAI had some idea of what is and isn't modifiable. Or better yet, if the genAI could be used in a "sandbox site" to modify site elements one by one.

Improve this answer
-2

I think you could help a lot the answerers to find the questions where they could create the best possible content. As an answerer, it was a long problem for me. Ultimately I invented some filters (in the top search bar) and queries (on the SEDE). I think, none of them were really useful. An AI-supported thing would be the most useful.

And you have the input for that: you have the vote stats, the various parameters of the OPs and of the answerers. And probably you have both the resources and the intention to improve your site with AI features. So, give these inputs to an AI and make it to a site feature.

For example, a link like "Show me questions to answer", I think I had clicked that link a lot in the past decade. This link should show questions which

  • I can answer
  • I would get the possible highest count of upvotes for the answer

So, here is a way.

(P.s. the foreseeable community mass down-vote is a no-entity here. They just hate people collecting rep. You want a community creating good content.)

(P.s.2. Actually, it could be done also independently from you and I am thinking on that; however as an official site feature it would be likely yet more effective.)

Improve this answer
-3

AI can be used for OCR, e.g. when people post screenshots of code, error message,
table, math formula or a slide. Not long ago, table markdown wasn't supported, so many tables are images.

Improve this answer
5
  • 2
    I would be worried about errors there and limited use cases. Posting images of code is against all rules and doesn't exist in any decent posts. Image to table conversion may be helpful in a few cases.
    – NoDataDumpNoContribution
    Jul 24 at 20:24
  • 3
    I really don't see any good reason why people should post images of code in the first place. So not only is this a waste of limited resources (both developing and then running the system), but it also encourages and supports undesirable behaviour (and creating a positive feedback loop that's only going to result in more and more resources being wasted on this).
    – Dan Mašek
    Jul 24 at 23:42
  • 1
    @DanMašek agreed, for the code, in most cases it's bad behavior.
    – Franck Dernoncourt
    Jul 24 at 23:45
  • 3
    There is a nontrivial cohort of new users who genuinely think that images of text are accepted or even preferred. Some of this superstition probably comes from stupid web forms which strip all text formatting, some of it probably from Twitter et al. where an image of text is the only way to post more than a paragraph or two, especially when quoting another source.
    – tripleee
    Jul 29 at 8:35
  • 2
    Merely detecting a screen shot with code might be sufficient, and could offer either "the screen shot illustrates a problem which is unrelated to the actual code in it" or "continue editing to replace the screen shot with a copy/paste of the actual code, as text"
    – tripleee
    Jul 29 at 9:04
-3

A Data Explorer query assistant

Honestly, I don't know if this is feasible (I'm just brainstorming), but... Often I find myself able to express in words what I want the Data Explorer to show me, but unable to write the code.

I asked GPT-4:

Can you write a Stack Exchange Data Explorer query which can list the first hot network questions at GenAI Stack Exchange?

and ran the code it gave me here. It didn't work for two reasons: the sorting is wrong, and it totally ignored the hot network question requirement. It just gave me the most viewed questions.

Turns out this query works (or at least it would have worked if GenAI.SE data was included in the Data Explorer).

Improve this answer
3
  • 2
    I think exploring a data set with natural language queries would be a great use of generative AI if the model were trained properly. The query language has very structured syntax that should help it make the right choices (or at least get you there after a couple iterations).
    – ColleenV
    Jul 27 at 12:48
  • 2
    There's also another issue of searching for queries, because people use all sorts of synonyms and abbreviations.
    – Rebecca J. Stones
    Jul 27 at 13:07
  • 4
    related: Can GPT-4 be used to do SEDE queries?
    – starball
    Jul 29 at 9:01
-4

GenAI can be used to draft an answer from scratch, which the human will improve. Authors can learn when GenAI is more likely to give a decent draft. FYI: stats on ChatGPT plagiarism.

Improve this answer
7
  • 2
    First thing would be a fact check before any improvement can be done. In the end, it might even be more work than starting from scratch.
    – NoDataDumpNoContribution
    Jul 24 at 20:22
  • Maybe if the IA draft was not editable, not copyable and not publishable like a grayed <input placeholder>. It could be a solution to discourage an user from posting a ChatGPT answer.
    – Augusto Vasques
    Jul 25 at 4:58
  • 1
    @NoDataDumpNoContribution thanks, yes improving includes fixing incorrect statements.
    – Franck Dernoncourt
    Jul 25 at 5:00
  • 1
    @FranckDernoncourt If the AI result is incorrect it's like starting at -1 instead of starting at 0, more work instead of less. If only someone knew how often AI gets it right and wrong? Oh wait, you asked for that already.
    – NoDataDumpNoContribution
    Jul 25 at 6:07
  • 1
    @NoDataDumpNoContribution depends on how incorrect it is, the author's typing speed, etc.
    – Franck Dernoncourt
    Jul 25 at 14:49
  • 3
    The answer you link to has no "stats" whatsoever, just a statement that the amount of direct quotes correlates with the accuracy of the genAI model, and conflates "plagiarism" with "verbatim quoting".
    – terdon
    Jul 26 at 11:25
  • @terdononstrike it's a question looking for stats. Agreed for the conflation but adding another question may be closed as a duplicate. Maybe I'll try though.
    – Franck Dernoncourt
    Jul 26 at 11:27
-4

Now bear with me on this one... if someone can turn a Reddit post about Minecraft into news, then why can't Stack Exchange turn Q&A into news? Hence...

A genAI Stack Exchange journalist

I asked Assistant to turn this Chinese.SE post into news:

You Won't Believe Why "雨下了" Is Not Used in Chinese!

Have you ever wondered why the phrase "it's raining" in Chinese is "下雨了" instead of "雨下了"? Well, buckle up because we've got the answer, and it's going to blow your mind!

First of all, forget everything you know about subject-verb-object (SVO) word order. According to Chinese grammar rules, the fundamental sentence structure is actually topic-comment, and SVO is only the default word order when it fits into the topic-comment structure. Mind-boggling, right?

Now, let's talk about presentative sentences. These are sentences where the subject is new information, and they have to be structured differently from regular SVO sentences. In Chinese, presentative sentences involving a new subject use a semi-pleonastic existential verb or a verb of motion, and the subject occurs after the verb.

So, what does all of this have to do with the phrase "it's raining"? Well, weather expressions in Chinese are essentially presentative in nature, meaning that the default position of a weather expression is that the weather is treated as new information. In other words, the phrase "下雨" on its own would be a presentative sentence consisting of a motion verb and a presentative subject, equivalent to "rain is falling" or "there is rain falling" in English.

But what about the "了" at the end of "下雨了"? This is a particle that indicates a currently relevant situation or a change of state, and it's often used with weather expressions to show that the currently relevant situation is rain/wind/whatever. When used with "下雨", it indicates that it's currently raining. However, if we use "雨下了" instead, it would be grammatically incorrect because "雨" would be treated as the object, which clashes with the indefinite nature of the phrase.

In conclusion, the reason why "雨下了" is not used in Chinese is because it violates the grammar rules of presentative sentences and weather expressions. So, there you have it, folks. The shocking truth behind why "雨下了" is not used in Chinese. Don't say we didn't warn you!

(Okay so this news has bugs and clickbait, but it's to illustrate the concept.)

This (perhaps with some manual intervention) can be used to promote the sites, draw attention to detailed answers, and perhaps get our contributions into Google News or other news aggregators.

Improve this answer
3
  • 4
    You'd need to make sure your AI is extra careful about which posts it does this with, you don't want to end up with a repeat of the Glorbo incident.
    – F1Krazy
    Jul 27 at 13:29
  • 2
    and the point of this would be... ?
    – starball
    Jul 29 at 9:04
  • 1
    I'm envisaging it being used for those high-effort answers, which contain a lot of detail, that showcase how awesome Stack Exchange can be. The idea is convert such answers into a news report, press release, or similar promotional material, to share on other sites like Reddit, Twitter, and so on, and attract users to Stack Exchange. (I'm assuming there'll be a human involved in the final draft, at the very least.)
    – Rebecca J. Stones
    Jul 30 at 4:44

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .