[stackoverflow.se] — InSync 19 secs ago

@tchrist To be honest, I came away from that post more confused than before. — user3840170 22 secs ago

@Sasha - You'll have to excuse my hasty mock up - but something like this would be ideal: i.sstatic.net/GscF2hmQ.png. Obviously some thought would have to go into who gets to see what links but most of that logic already exists for the links on the top bar. Plus, moving this functionality to the left-nav would allow you to remove it from the top bar in some contexts (eg mobile), making the top bar a bit cleaner. — Robotnik 23 secs ago

Hi Anish Raut, welcome to the Stack Exchange Network Meta site! I'm not sure which search brought you here but the problem you describe will not be answered on this specific site. To get an expert's answer for the topic of your question you'll have to find and then re-post on the proper site. Check How do I ask a good question and What is on topic on the target site to make sure your post is in good shape. Your question is definitely off-topic on Meta and is better deleted here. — Journeyman Geek 7 secs ago

Hi sbv, welcome to the Stack Exchange Network Meta site! I'm not sure which search brought you here but the problem you describe will not be answered on this specific site. To get an expert's answer for the topic of your question you'll have to find and then re-post on the proper site. Check How do I ask a good question and What is on topic on the target site to make sure your post is in good shape. Your question is definitely off-topic on Meta and is better deleted here. — rene 48 secs ago

Hi LJ S.A., welcome to the Stack Exchange Network Meta site! I'm not sure which search brought you here but the problem you describe will not be answered on this specific site. To get an expert's answer for the topic of your question you'll have to find and then re-post on the proper site. Check How do I ask a good question and What is on topic on the target site to make sure your post is in good shape. Your question is definitely off-topic on Meta and is better deleted here. — rene 23 secs ago

This does not really answer the question. If you have a different question, you can ask it by clicking Ask Question. To get notified when this question gets new answers, you can follow this question. You can also add a bounty to draw more attention to this question. - From Review — Rory Alsop 44 secs ago

Still present in 2024 on EE.SE. English + Russian. — winny just now

Spam. Please flag as spam. — greg-449 40 secs ago

It is worse than just recordings. I received an invite and it asked me to sign an NDA and another form, both seemed like complicated legal documents. I am not a laywer and was concerned about the implications upon me, hence I declined the invite. — AdrianHHH 12 secs ago

You've posed a few questions, but I don't follow your argument. What would be the purpose of automatic acceptance - if an the top answer isn't the best solution, but the best solution isn't complete - there wouldn't be any reason to have bounties to attract better answers. Can you clarify your argument. — W.O. 27 secs ago

I have the problem in Chrome Chrome is up to date Version 124.0.6367.119 (Official Build) (64-bit) but the image feature works fine in Firefox Firefox is up to date 125.0.3 (64-bit) I am using Windows 11 version 23H2 22631.3527. — Michael Harvey 22 secs ago

No. This is entirely against the point of acceptance, which was already a junk function anyway. — Nij 14 secs ago

"One of these information is in contradiction with another." or the "4 mins ago" is just wrong. The timestamps don't update in real time. — VLAZ 39 secs ago

@VLAZ I saw your comment as it said it was 4 mins ago — The Empty String Photographer 35 secs ago

But then it doesn't! Seems to be intermittent — Michael Harvey 5 secs ago

Hi Frankie, welcome to the Stack Exchange Network Meta site! I'm not sure which search brought you here but the problem you describe will not be answered on this specific site. To get an expert's answer for the topic of your question you'll have to find and then re-post on the proper site. Check How do I ask a good question and What is on topic on the target site to make sure your post is in good shape. Your question is definitely off-topic on Meta and is better deleted here. — samcarter_is_at_topanswers.xyz 18 secs ago

... which does not change the fact that the two pieces of information are in contradiction. Rather, it explains it. — Emil Jeřábek 42 secs ago

Please consider including more information directly in this post instead of just linking to the blog post. It would also be worth including here preemptive responses to the most obvious concerns that users are going to have in response to this announcement. — Mithical 40 secs ago

This post is on Stack Exchange Meta, but it sounds like the partnership is only for Stack Overflow and not the rest of the network. Can you clarify? — Adám 24 secs ago

I second @Mithical. I'd also recommend keeping in mind that the audience of the blog and the audience of Meta SE are very different. The blog, based on the past posts, tends to be the general public, including various stakeholders. It's a marketing tool. Meta SE should be focused on the community. It may take some more work, but I'd strongly recommend drafting two versions of these announcements and making sure the Meta SE version doesn't have business/marketing speak, but targets our concerns as active participants in the communities. — Thomas Owens 13 secs ago

@Adám How I read it, this is more about the company and less about the community sites. In which case it's best to post it where most people would read it, which is here. — Mast 28 secs ago

@Adám They're probably feeding the companies data from the entire network — Zoe 58 secs ago

Did this partnership get delayed because of the recent (humongous) LLM abuse case over on SO? The timing raises some eyebrows. — Cerbrus 24 secs ago

@Cerbrus - no, not at all. The two are unrelated, and the situation on SO was never a factor in the timing of this release. — Philippe ♦ just now

@Adám The [featured] tag for MSE goes network-wide. I believe this is a network-wide feature. Remember, "Stack Overflow" is both a website AND the collective company. It gets confusing, but I believe the press release is referring to the latter. — Machavity 18 secs ago

LLMs, by their very nature, don't have a concept of "source". Attribution is pretty much impossible. Attribution only really works if you use language models as "search engine". The moment you start generating output, the source is lost. — Cerbrus 58 secs ago

@Cerbrus Enterprise Copilot would like a word with you. LLMs do what they are trained, if you train them on attributing, they will attribute. — tomdemuyt 6 secs ago

@tomdemuyt No. GenAIs are not capable of citing stuff. Even if it did, there's no guarantee that the source either has anything to do with the topic in question, nor that it states the same as the content of the message. Citing stuff is trivial if you don't have to care if the citation is relevant to the content, or if it says the same as you. (source) — Zoe 23 secs ago

The most important part is the last question, if OverflowAPI can disrupt the answer quality, I definitely will left this rubbish site. — user1176409 25 secs ago

I will be keeping tabs on this post as is @Philippe for the next week or so and we will answer what we’re able to. 2/2 — Rosie ♦ 33 secs ago

The details are in the press release that we linked. I understand there is a desire for more specifics, but we are just announcing the partnership today. As work begins in the future, we will have more to share about how integrations with our partners will work. In terms of project phases, we’re at the very beginning - there’s a great deal of discovery work and actual coding to do, and during that time, anything specific promised about how this will work here could change, so we want to be sure that we aren’t inadvertently misleading folks in these early announcements. 1/2 — Rosie ♦ 59 secs ago

There are plenty of cases where genAI cites stuff incorrectly, that says something different, or citations that simply doesn't exist at all. Guaranteeing citations are included is easy, but guaranteeing correctness is an unsolved problem — Zoe 56 secs ago

@Zoe If you ask ChatGPT to cite it will provide random citations. That's different from actually training a model to cite (e.g. use supervised finetuning on citations with human raters whether sources match, which would also allow you to verify how accurately a model cites). This is something OpenAI could do, it just doesn't. — Erik A 8 secs ago

@ErikA I've seen people use models that generate citations, and they still manage to incorrectly cite sources. Those models are certainly better than ChatGPT, but they still screw up correctness of the content they cite regularly. Overflow AI had this very same problem while it was still running. — Zoe 16 secs ago

Do the people who actually contributed the answers get anything out of this deal (attribution? access to the trained model? part of the profit?) — Aykhan Hagverdili 20 secs ago

Guys, you seem to misunderstand how LLMs generate text. They don't copy text verbatim. They generate it word by word. When generating, LLMs don't even have any knowledge about their original training data. They just have the model that was built from the training data. — Cerbrus 51 secs ago

Hmmm. Seems like there is an opportunity to allow for further preferences within the research bucket to be selected. — Piper 35 secs ago

Re "speed is a core goal of the Stack Exchange model": It was the original killer feature of Stack Overflow (also encouraged by FGITW): Comprehensive answers to all answerable non-duplicate questions within a few minutes (but things changed). — This_is_NOT_a_forum 58 secs ago

The first jobs to be done is capturing why some users choose Stack Overflow and what one of their pain points is. Although this is a wider problem users are facing then what Stack Overflow is positioned to solve, it is in fact a reason why a majority of our (think anon) users visit the site(s). — Piper 20 secs ago

Our accessibility research is not a part of that pattern. It is about taking a team who is well positioned to solve certain accessibility issues and is not well positioned to work on core product features. — Piper 36 secs ago

Where is it broken? In the revision views, but not in the main view (that most users frequent)? — This_is_NOT_a_forum 13 secs ago

@AykhanHagverdili we're allowed to pay to access it through the 3rd party's services — Kevin B 25 secs ago

There are also additional financial aspects here; especially regarding SO: If OpenAI decides that only their Paid Tier gets the newest SE/SO data, you essentially need to pay, for efficient GenAI use cases. This makes a unique selling point for ChatGPT, where basically another company is profiting off of the contributors works. Apart from that, there needs to be a consideration of why the knowledge was given. SE thrives from people providing knowledge for free. Not really sure how I should feel about that yet... — A7exSchin just now

Related: Our partnership with Google and commitment to socially responsible AI — This_is_NOT_a_forum 58 secs ago

Is there a difference between conciliated and reconciled? — This_is_NOT_a_forum 39 secs ago

@Robotnik thanks! I shared that with someone on our UX/UI team. — Sasha ♦ 20 secs ago

Is this a question for SE or about LLM in general? — Mast 29 secs ago

SE's current writeup is very thin on the details - and I'm not entirely sure if the gen AI companies are buying access to data retrieval on demand, a package of data or something else. If its the former - and its live data, it would be simpler than if SE was selling training data with attribution baked in — Journeyman Geek 42 secs ago

@Rosie That's the kind of content that should be in the body of the post on Meta SE. Information about what phase the work is in, when we should expect more information (days, weeks, months), what the next things that the community will see are. This is the important stuff. Also an explicit statement that someone or some people are watching and will be addressing questions (and then, of course, following through by updating the post with answers to questions). I think if you do these things, you may alleviate some down votes that result in these posts being hidden from the front page. — Thomas Owens 34 secs ago

SE specifically, edit the title if you can improve it. To be clear, this is not a generic question for "GenAI" — SPArcheon 59 secs ago

What action by SE specifically do you think is a problematic with the GDPR's right to be forgotten? They haven't build an LLM (yet). — Mast 10 secs ago

I would guess that if you trained an llm with attribution - how you would revoke it? It would be like an old data dump in that SE no longer controls it — Journeyman Geek 48 secs ago

@Mast short version: I was reading this users concerns about missing attribution in regards to the announced OpenAi partnership and that made me wonder if there is a symmetrical issue with the inability to keep the attribution "up to date" — SPArcheon 34 secs ago

@JourneymanGeek true. But the dump is just a snapshot of data at a given moment, so no one will probably require you to update an old version. Can a LLM be considered a snapshot? Or is it a product that is required to be updated in some way? Really, as I said, I am mostly brainstorming here. Feel free to post your views as answers, it will be appreciated. — SPArcheon 12 secs ago

if people assume AI generated content is "quality, accurate, sourced data" (it's not), then they will think it's fine.; we'll it doesn't help if one of the network's most active/known user has posted more then 1000 questions based based on generative AI content. — 0stone0 20 secs ago

Why put forth effort into a post noone here wants to exist that's just gonna get downvoted to oblivion like the rest of such posts — Kevin B 24 secs ago

@JourneymanGeek Didn't have the time to read it thru yet, but have a look at the link I just added. — SPArcheon 24 secs ago

"...I love using SE because SE is not a forum" --> post edited by @This_is_NOT_a_forum. Amazing! — SilentCloud 22 secs ago

@SilentCloud What is the purpose of this comment? — user1176409 just now

"We’re pleased that OpenAI shares our commitment to socially responsible AI." This can never be true so long as Sam Altman is involved in OpenAI's leadership. — TylerH 46 secs ago

"If the "attribution" contains the actual user name, have we just created an uncontrolled persistence of PII? [...] If that is the case, I can't see a way to maintain it, nor a way to satisfy the requirement of most countries data protection laws." It depends on how the AI model is trained on SO data, I guess. But regardless, once data has been copied/transmitted somewhere else from Stack Overflow, it's no longer Stack Overflow's responsibility to deal with it as part of a GDPR request. You'd need to issue a GDPR request to the company that builds the AI model to do that. — TylerH 41 secs ago

This is just an initial test of a new version of Stack Overflow Jobs, during which we are focused on validating the potential of this feature (based on the questions I listed in the “What's being tested” section). So for now, it is basically a dynamic, searchable list of technical job openings, which will be linked from SO and technical SE sites. — Sasha ♦ just now

In the future we hope to build out more unique functionality that will increase the site’s value to users who are seeking new job opportunities. With that in mind, we appreciate all the feedback on this post so far about what specifically users would find the most useful. — Sasha ♦ 1 min ago

Thanks for this input. I just want to point you to a few of my prior responses that address some of these concerns, in terms of wanting to complete applications within SO, and about the possibility of bringing back dev story. — Sasha ♦ 13 secs ago

Thanks for this response! We are keeping track of the feedback on this post and will keep it in mind as we continue to iterate on this feature. — Sasha ♦ 20 secs ago

This makes me more comfortable about the liability accrued by OpenAI's actions (which, unless copyright law is changed, Stack Exchange can not afford to take on). Unless they have secret technology they haven't released (not unlikely), OpenAI does not have the capacity to provide attribution in their AI systems. Are all parties comfortable with a partnership where one of the partners isn't allowed to do anything? — wizzwizz4 35 secs ago

@Zoe It is more likely they will use some kind of RAG i.e. search engine + summarization . — pcpthm 1 min ago

@user1176409 Silent probably forgot NO FUN ALLOWED ON STACK EXCHANGE. — Azor Ahai -him- 44 secs ago

@TylerH Scraping human-generating writing that was not licensed to OpenAI is socially irresponsible and any company that does this can fundamentally not be doing "socially responsible AI." — Azor Ahai -him- 56 secs ago

OK, it's May, I checked back, and it's not looking much better to me. Best of luck trying to ride out the AI hype-fueled data grabs. I've got to figure out how to be a tech optimist and luddite simultaneously lol. I still want my flying car but not if it's going to steal all my know-how for some giant corporation's AI model so they can sell it back to me in a tool that will make flying my car easier. — ColleenV 45 secs ago

@JoshZhang How can you say (in my face!) that others on my ISP don't have any issues? when I'm observing the issues also at my workplace and this information has been there in the question from the revision 1. Please explain this to me. "We have found no evidence that others have issues" may be true; "others don't have issues" is false. — Kamil Maciorowski 29 secs ago

"Will the users who actually spent time writing and contributing their content be attributed in any way when their content is fed into OpenAI's products?" I don't attribute where i got my code from when i conjure it out of my brain. Maybe it was 31 years of documentation, or open-source projects, or online respositories, or textbooks from university. If a carbon-brain can't supply attribution, we will not require a silicon brain to supply any. Nor do i require attribution of any of my code contributions. — Ian Boyd 16 secs ago

Why isn't this in the body of the question post? — zcoop98 32 secs ago

Trusting OpenAI to be socially responsible, is like trusting Facebook to respect your privacy. — S.L. Barth is on codidact.com 26 secs ago

No elected moderators were involved in closing your question: math.stackexchange.com/posts/4910819/timeline — rene 40 secs ago

Will this really be attribution, or will the output of ChatGPT just be used to find random posts that happen to resemble the answer? I suspect the latter. — S.L. Barth is on codidact.com 9 secs ago

@Sasha - I hope thats the case but lets remember - "The effort it would take us to truly differentiate in this space is not one we could justify" — Sayse 35 secs ago

Not everyone can access external sites, and the links may break over time. — Sayse 14 secs ago

See also: Names of close-voters are not visible in the post notice but in the revision history and timeline — Spevacus 14 secs ago

I asked if moderators had authority to do this. — user253751 9 secs ago

"an exciting new partnership" -- oh, please, stick to the facts, you're not addressing a bunch of kindergartners. And judging from the vote count, the "excitement" is not exactly mutual. — Dan Mašek 34 secs ago

Folks, let's not close this. While more details in the post would have been nice, it's not productive to close staff announcements, especially featured ones. — cocomac 53 secs ago

I know stuff like Copilot will link back to things that it used in a web search, but yeah, I haven't heard how attribution works practically right now for stuff that goes into the training of the model (and not just its web searches after training is over). I've heard of some research stuff, but not something that has gone into the big models today. — super-starball-ultra 19 secs ago

And still: Our Partnership with OpenAI (with a score of -81 at the time of writing). — Joachim 17 secs ago

@This_is_NOT_a_forum, the edit and revision views (to my knowledge, these are the sole alternative views). I've listed all that I know of. This indeed excepts the primary answer view. — RokeJulianLockhart 11 secs ago

This statement is very weak, especially the "in a license-compliant way". The attribution clauses in the CC licenses have a lot of ambiguity and the "reasonable manner based on the medium, means, and context" leaves outs for the person receiving the content. It's very possible to be fully compliant with the license with respect to attribution and have a very weak form of it. — Thomas Owens 14 secs ago

As a moderator on Writing.SE, where writing critiques are disallowed, I'm not sure I'd accept such a loophole. Answering a question by just dumping a piece of writing isn't really allowed. — F1Krazy 41 secs ago

Why is the default link from the Stack Overflow Featured tab sorting answers by newest first? Could be just me, but the link I'm being shown contains ?cb=1 at the end. — Unchain2644 6 secs ago

Guess my contributions to the community have reached an end. Godspeed, stack overflow. — stateMachine 53 secs ago

While the social responsibility of AI is important, what about the environmental impact of AI? — McN 21 secs ago

I mean... I knew all my answers were being stolen before, but having it cemented like this is repugnant. My contributions will be made elsewhere from now on. — Caspar Harmer just now

It's a bit unclear what your answer is saying. You seem to be talking about another platform entirely. While that's perfectly reasonable, how does that relate to the announcement here? Could you edit your answer to give details about how this relates to Stack Overflow Jobs? — Catija 31 secs ago

I will be expecting my username to be attributed to every output made by this AI model (which itself will be licensed CC-BY-SA) assuming any content I have contributed is used to train it, as dictated by CC-BY-SA. A note giving attribution to SO/SE as a whole is not enough, it has to be my username, otherwise it is a breach of license and I will take appropriate action and issue takedown notices to stop any parties violating my copyright. CC-BY-SA is very explicit about this. I doubt I'm the only one going to do this. Are you going to comply with this — Kryomaani 31 secs ago

@TylerH Whilst not EU GDPR, the UK ICO guidance is that because SO was responsible for the initial consent, they are also responsible for getting any third partys they sold the data to (like google) to delete any personal data on request: ico.org.uk/for-organisations/law-enforcement/… — user1937198 32 secs ago

Trivia: What does Stack Overflow have in common with Radio Shack and Blockbuster? — mickmackusa 11 secs ago

SE seems to have found a route forward in the form of exchanging data for something useful. Whether that's lipstick on a pig or a wodge of cash is anyone's guess. techcrunch.com/2024/05/06/… — Andy Gee 27 secs ago

How does it cause "a lot of unneeded traffic"? It could be updated in real time purely client side. — Luna 7 secs ago

I agree with "inconclusive". The other two are far too vague, especially "handled", which just sounds like a synonym for "helpful". — Luna 42 secs ago

I don't know about the current devs (mainly cause their quiet) but I'd seen Nick literally talk about shaving milliseconds off on twitter. As for the actual technical details, as opposed to historical context and memes, I think rene's answer covers that better and is worth an upvote ;) — Journeyman Geek 54 secs ago

Hi Seydina Sonko, welcome to the Stack Exchange Network Meta site! I'm not sure which search brought you here but the problem you describe will not be answered on this specific site. To get an expert's answer for the topic of your question you'll have to find and then re-post on the proper site. Check How do I ask a good question and What is on topic on the target site to make sure your post is in good shape. Your question is definitely off-topic on Meta and is better deleted here. — Journeyman Geek 32 secs ago

The company that's running an un-monitored experiment on billions of people is socially responsible? What? Though I mean, this is also the platform that people compare to the Stanford prison experiment, so maybe these two companies are a good match. — Rei Miyasaka 23 secs ago

"anything specific promised about how this will work here could change" is the most hilarious oxymoron I've heard since the last time I tried gaslighting ChatGPT. — Rei Miyasaka 29 secs ago