1
$\begingroup$

Recently, I’ve seen a number of CV users plugging questions into ChatGPT to help generate answers/content. Some of these users self-report to be educators. So, here’s the question:

When we turn inquisitive minds to ChatGPT, are we not participating in our own obsolescence?

I worry that because ChatGPT is “free” (for now) and most of us use it (or have tried it because of its novelty), we’re actively abetting the OpenAI developers in our own labor displacement.

The “we” in my question is a bit broad, but may apply to professors, web developers, and investment bankers.

$\endgroup$
24
  • 1
    $\begingroup$ Hi Thomas. I can understand your concerns. To be frank, the community is not really that much delighted either with the recent plague of half baked ChatGPT advised posts. But I don't know how to answer your query and for that matter how it is related to the site, if not irrelevant. We have a detailed meta thread though: Temporary policy: ChatGPT is banned and What to do about answers written by ChatGPT. $\endgroup$
    – User1865345
    Apr 22 at 5:32
  • 1
    $\begingroup$ These others posts are quite intriguing. I was promoted primarily by the irony of educators to turn students to ChatGPT. $\endgroup$
    – Thomas Bilach
    Apr 22 at 5:51
  • 3
    $\begingroup$ Never used it, never tried. That’s manifestly not a refutation of “most of us” but how do you know? $\endgroup$
    – Nick Cox
    Apr 22 at 8:54
  • $\begingroup$ I don’t. But I do predict OpenAI’s ChatGPT will have more monthly visitors than Amazon by the end of the year. I’m trying not to catastrophize, but I am more worried than hopeful. $\endgroup$
    – Thomas Bilach
    Apr 22 at 9:10
  • 4
    $\begingroup$ @User1865345, this Q does seem to me to be about the site, its users, & their behavior, broadly construed. Cross Validated Meta (or really any meta SE site) has a pretty loose standard for what is 'about' the site & thus on topic. This Q seems fine to me. $\endgroup$
    – gung - Reinstate Monica
    Apr 22 at 12:13
  • $\begingroup$ I interpreted the post as an open discussion of the present fate of the practice. Indeed it affects the site. But somehow I couldn't see any possible robust answer to it, for that matter. Again, to reiterate, serious discussions have already been done in those posts. This is my humble opinion only. At the end of the day, we all want to get rid of this impending menace. $\endgroup$
    – User1865345
    Apr 22 at 12:23
  • 1
    $\begingroup$ The other posts are helpful, but they don’t quite address my concerns. I’m not asking what the community (moderators) should do about ChatGPT-derived answers or whether ChatGPT should be banned (although it’s not). My question is, should we encourage participation in a process that may displace us? $\endgroup$
    – Thomas Bilach
    Apr 23 at 8:11
  • 3
    $\begingroup$ Didn't everyone that has produced text (well, text that has been digitized in some form) end up doing free labor for OpenAI? That's a lot of good company (and also some bad company). $\endgroup$
    – dipetkov
    Apr 23 at 20:25
  • 1
    $\begingroup$ This does not seem like a duplicate of the previous ChatGPT question, but I don't really understand the level of concern here. If LLMs were that much of a worry, we'd have to stop contributing any content on the internet - not just text. Killing the internet because of plagiarism worries seems like an overreaction. If you think this contributes to our obsolescence, surely you'd want to avoid even writing an answer here? $\endgroup$
    – mkt
    Apr 24 at 7:45
  • 1
    $\begingroup$ Also, I would absolutely welcome a future in which most basic questions get addressed by LLMs instead of getting posted here. It would be easier and more fun to participate. Though unfortunately given the amount of bad statistical information online, I expect that LLMs will be regurgitating some lousy advice for a while. $\endgroup$
    – mkt
    Apr 24 at 7:46
  • 1
    $\begingroup$ @ThomasBilach Too late for what? I think it's worth integrating in our own workflows - I am, especially for coding - but the role it should play in CV is a separate question. At present LLMs are unsound in that they (perhaps unavoidably) produce highly convincing nonsense in addition to helpful text. It can be easy to filter out the bad stuff if you understand the topic reasonably well but very misleading if you don't. Which is why it's presently banned here, as the linked thread discusses. $\endgroup$
    – mkt
    Apr 24 at 9:11
  • 3
    $\begingroup$ @ThomasBilach: I'd expect the question would be closed if it provided no issue more specific than wanting to learn more about something. And what people may suggest in comments isn't strictly policed, but "try that" wouldn't constitute a valid answer (regardless of where the code came from.) $\endgroup$
    – Scortchi - Reinstate Monica Mod
    Apr 24 at 17:40
  • 4
    $\begingroup$ I do pay to use it; about $20 per month to get access to GPT-4, the latest version of their model. I'm not worried about automating myself away. Contributing to LLMs doesn't seem that different to me than contributing to Wikipedia or CrossValidated or StackOverflow - nobody gets paid for that either. I think raising the general standard of human knowledge & skill (setting aside weaknesses of LLMs) is valuable enough to me to do for free, at least part time. And it's not obvious that it will automate us away. Raising general standards might create more demand for expertise. $\endgroup$
    – mkt
    Apr 25 at 7:40
  • 2
    $\begingroup$ So it might even expand our professional opportunities! noahpinion.blog/p/nobody-knows-how-many-jobs-will-be $\endgroup$
    – mkt
    Apr 25 at 7:41
  • 2
    $\begingroup$ I have three concerns, one highly relevant here. 1. Do I want to use it? 2. What does this imply for me and my University and other universities? Important questions personally but consider 3. Is use of Open AI messing up ,Stack Exchange with lousy questions and answers? I am already seeing postings of bizarre code that "doesn't work" and that goes beyond what learners would guess for themselves. So, while I can ignore what I don't want to use, the pollution problem is serious. $\endgroup$
    – Nick Cox
    Apr 25 at 10:09

3 Answers 3

Reset to default
11
$\begingroup$

I think there's a confusing mix of different questions and concerns here.

Are we free labor for OpenAI?

Yes, we are. As is nearly everyone who has ever written any form of public (and some private) text that survives - in books, documents, websites, etc. I understand Sextus Empiricus' argument but am willing to accept a broader definition of 'labour' - though I think it ultimately does not matter, for the reasons that follow.

When we turn inquisitive minds to ChatGPT, are we not participating in our own obsolescence?

Regarding the first half - nobody here is actually advocating telling people to go to ChatGPT for answers. This community is arguably more hostile to it than most (perhaps understandably so).

As for the obsolescence part, I have multiple disagreements.

  1. Most questions here are basic and many are entirely misguided. A tool that could address even the basic questions automatically (and LLMs cannot do this yet, but maybe in the future) would make this website much better and more fun to participate in.
  2. You are already contributing your expertise for free (with a CC BY-SA 4.0 license) on a website that allows anybody in the world to ask a question and get an answer. This is what makes CV great, as well as StackOverflow, and what to me is one of the greatest sources of knowledge in the world - Wikipedia. Wikipedia is run by a nonprofit but note that CV and StackOverflow are also making a private company money. Under what principle is contributing to CV/StackOverflow/Wikipedia fine and not an LLM?
  3. Maybe what bothers you is the lack of credit/status that accrues to the originator of the text? But many of us participate anonymously or pseudonymously, so it seems unlikely that this is a broadly shared concern. Providing such attribution may not even be possible given how LLMs work. Like a human-produced text, the output arises from a mishmash of many sources, at least some of which will be contradictory.
  4. Much of the world is arguably expertise-limited. If we raise the general standard of understanding and skill, the world could become a much better place! I would love it if LLMs can make this happen, even a little. And if they do rise in importance and change our societies, the people who created the large corpuses of useful training data - such as the anonymous wikipedia writers and editors - will have had a remarkably large role in shaping our future. I suspect many of them would be quite pleased by this.
  5. It's entirely possible that raising the general standard of knowledge & skill will lead to a greater demand for real expertise! Which could increase demand for our services. We really have no idea how this will play out: https://www.noahpinion.blog/p/nobody-knows-how-many-jobs-will-be
  6. I asked you about your proposed alternative and you say that we could "stop using it". If you mean stop using ChatGPT, well, that doesn't matter one way or the other - your writings will still be part of its training data. You just won't be able to take advantage of what it provides. If you instead mean to stop using CV, I don't know what that achieves. As I said - your previous participation here was already making a private company money. You could do so but many of us will continue for the reasons I mentioned above.

Instead of advocating that we all down our pens, you could consider directing your efforts towards policy/regulation. I don't claim to know the best path forward but there are ongoing debates about what we should allow AI companies to make use of for free and what they owe the rest of us. These are normally framed in the context of plagiarism of art (for models like DALL-E, Midjourney, etc) but the principles apply more broadly. For example, is it acceptable/legal to train an LLM on data with a CC BY-SA 4.0 licence and not provide specific attribution in any answer? I think this would be technically impossible for LLMs and I personally hope such concerns don't kill them, but it seems like a grey area.

$\endgroup$
5
  • 1
    $\begingroup$ (+1) I like "down our pens" as a metaphor for "keep our fingers away from the keyboard". Nicely put. $\endgroup$
    – Nick Cox
    Apr 27 at 13:09
  • $\begingroup$ @mkt Thank you for your comprehensive answer, in particular addressing the other points in the comments section. When I cautioned about not using it, I was referring to ChatGPT. Do you suggest LLMs could reach their full potential using just the available crawl data? Why not put ChatGPT behind a hefty paywall? $\endgroup$
    – Thomas Bilach
    May 1 at 8:36
  • 1
    $\begingroup$ @ThomasBilach Nobody knows their full potential, but I imagine more data would be better, especially high-quality data. And more data will continue to be available whether a few individuals decide to stop writing or not. As for putting ChatGPT behind a paywall, I don't really get your point. (1) It is already partly paywalled (access to GPT-4 requires a monthly subscription), and (2) we have zero ability to effect such a change, and there's no reasonable way to make that happen through laws or regulation - it's a decision by a private company about how to offer its services. $\endgroup$
    – mkt
    May 1 at 8:54
  • $\begingroup$ The point is someone like yourself will pay for it and integrate it well into your workflows. But while it was “free” (and still is) it attracted 10 million dim-witted questions. I don’t suspect more “high quality” data is on the horizon, but that’s just my prediction. I’m also well aware that most predictions are wrong. Let’s see where this goes :) $\endgroup$
    – Thomas Bilach
    May 1 at 9:30
  • $\begingroup$ @ThomasBilach Again, assuming LLMs get better at accuracy, attracting many 'dim-witted questions' is a feature, not a bug. The public gets basic questions addressed, improving their knowledge and skills, and experts don't have to spend their time answering them. To me, the main open question on this narrow topic is: how accurate will they become? And we don't know the answer to that yet. $\endgroup$
    – mkt
    May 1 at 9:55
3
$\begingroup$

The company plans to sell its data to large-scale AI projects. So while we can't answer the question of whether or not the data will be sold to OpenAI specifically, we do know that there is a plan to sell it.

Paresh Dave. "Stack Overflow Will Charge AI Giants for Training Data" Wired. April 20, 2023. https://www.wired.com/story/stack-overflow-will-charge-ai-giants-for-training-data/

OpenAI, Google, and other companies building large-scale AI projects have traditionally paid nothing for much of their training data, scraping it from the web. But Stack Overflow, a popular internet forum for computer programming help, plans to begin charging large AI developers as soon as the middle of this year for access to the 50 million questions and answers on its service, CEO Prashanth Chandrasekar says. The site has more than 20 million registered users.

Stack Overflow’s decision to seek compensation from companies tapping its data, part of a broader generative AI strategy, has not been previously reported. It follows an announcement by Reddit this week that it will begin charging some AI developers to access its own content starting in June.

SE appears to be undertaking steps to support this monetization. See: June 2023 Data Dump is missing

But now has has temporarily suspended those steps. See: https://meta.stackexchange.com/a/390200/320588

In case anyone is curious, there are some options for downloading all of your contributions to StackExchange. This is the a recent Q&A that I've found on our Meta. Is it safe to use Cross Validated as a note keeping software?

$\endgroup$
6
  • 1
    $\begingroup$ Are we talking about the content of answers and questions? Isn't this open source anyway and can it not be scraped without any costs or obtained from database dumps? How can SO charge any money for it? Or is this also about non-public meta-data? $\endgroup$
    – Sextus Empiricus
    Jun 9 at 19:56
  • $\begingroup$ I only know what I read in the news. 🤷 $\endgroup$
    – Sycorax Mod
    Jun 9 at 21:05
  • 1
    $\begingroup$ I am considering to sent an application letter to openAI. Scraping this website is super easy and I believe that this amateur SE can be much cheaper than whatever the company SE will be asking. $\endgroup$
    – Sextus Empiricus
    Jun 9 at 21:09
  • 1
    $\begingroup$ As I understand it, there is an intention to make the data inaccessible for LLM training outside of SE's framework. meta.stackexchange.com/a/390040/320588 In other words, SE appears to be attempting to find a way to obstruct the methods that you would undertake in pursuit of that task. $\endgroup$
    – Sycorax Mod
    Jun 10 at 18:14
  • $\begingroup$ Good luck with that to the company. It is practically impossible to block a site when you want it to be open at the same time. Or are they gonna block google's and bing's searchbots destructing their search engine optimization? $\endgroup$
    – Sextus Empiricus
    Jun 10 at 18:40
  • $\begingroup$ I have no idea what SE's plans are, aside from what they've announced publicly. $\endgroup$
    – Sycorax Mod
    Jun 10 at 18:45
0
$\begingroup$

Are we free labor for OpenAI?

No, 'labor' is a physical or mental activity/effort with the purpose of producing value in an economical system.

An AI like GPT is using the community for written texts, images, and actions, and produces an economic value out of it, but that might be better described as 'mining'. To OpenAI we might be considered a resource, but that is not yet the same as labor.

For example, people that are playing an amateur level game of chess or soccer are not performing labor. Their purpose is the fun of the game. Yet, their activities can be used by others to generate value. A chess game might end up in a database that is commercially used by a chess computer or a professional chess player. The soccer players might be photographed by a professional photographer who uses the photo for art or news and makes money out of it.

Additional discussion

This question is surrounded with several other statements/questions

  1. are we not participating in our own obsolescence?

  2. ChatGPT already takes advantage of 60–70% of the common crawl data, mostly from webpages. Isn't that enough?

  3. should we encourage participation in a process that may displace us?

  4. The contributor tells the OP to “try that” but I doubt the OP understands the nuance of the code or how the model works. Is this acceptable practice here?

And it seems to be motivated by some potential outcomes

  1. I worry that ... we’re actively abetting the OpenAI developers in our own labor displacement.

  2. I do predict OpenAI’s ChatGPT will have more monthly visitors than Amazon by the end of the year. I’m trying not to catastrophize, but I am more worried than hopeful.

This is a very broad range of questions and it might be good to bring the meta-question down to a single clear statement. Several of the statements/questions might be tackled on their own, but while people do this (and I will make a little attempt below) it might get lost in the large puddle of questions.

Answers to the individual questions might be

  1. This has a varying degree of answers as it is very broad.

    In the first place it is a loaded question and presupposes that the use of GPT or AI is leading to our own obsolescence. Even if it is used to our own obsolescence, then it is implied that this is a bad thing. This is a bit like the question whether an employee should strive to make themselves obsolete or not.

    Another approach to the question is to question whether or not AI leads to obsolescence.

  2. This relates to the comment

    By default it appears the user is the product.

    The user being the product is not restricted to OpenAI. It is more in general a matter in the information industry that we have intellectual property that is produced for free or at a cost, and commercially or open source. And, these two worlds are blended with each other.

    It is not only when I ask OpenAI to translate my texts, it is also when I interact on a forum providing help about using R or linux that I am contributing to a free resource while indirectly helping commercial companies. Such companies might be for instance RedHat or Canonical who benefit from open source developments in Linux, or Posit (formerly Rstudio Inc.) who benefits from R and other open source development.

  3. like 1

  4. This is an issue independent from OpenAI specifically and answered in the comments by Scortchi

    And what people may suggest in comments isn't strictly policed, but "try that" wouldn't constitute a valid answer (regardless of where the code came from.)

The four arguments here are very straightforward and simple, but I believe that the issue is more delicate. The problem is not as much the technology itselve in principle, and instead how it is gonna be used in practice (similar to discussions around genetically modified organisms, GMOs).

This more delicate discussion is at the same time a very broad discussion, and might be better suited for a platform like economics or politics, rather than stats.meta, because in the end it is more generally about how we see the future of the order in our society and the distribution of the wealth that comes out of earth's resources like AI, mineral mines, fossil fuels, etcetera. For a long time we have been able to enslave workers into doing simple tasks in exchange of a bith of money/wealth. In the near future this model might need to be revisioned. Possibly not global since it is a big question mark how much impact the new technology is gonna have, but at least locally and short term the new technology is gonna stir up the labour market.

$\endgroup$
11
  • $\begingroup$ Thank you for your response. So you agree GPT produces economic value but it’s only refined through mining? I’m sure we can agree the pre-trained language model was not perfect. To improve, it requires some reward system to fine-tune the process. In general, it only works well if it’s tailored to the human preference. So “we” are needed in the feedback process. How do you get a billion visitors to voluntarily input prompts? Make it free. $\endgroup$
    – Thomas Bilach
    Apr 26 at 17:43
  • $\begingroup$ Sure, OpenAI uses people in all sorts of ways to train and fine tune the models. I am sure that OpenAI also lets models crawl stackexchange, including your question and comments here. Models can learn from all sorts of interactions. So do humans, they learn from interactions between other humans and they can be many times for free. $\endgroup$
    – Sextus Empiricus
    Apr 26 at 18:55
  • $\begingroup$ Good point. ChatGPT already takes advantage of 60–70% of the common crawl data, mostly from webpages. Isn't that enough? Doesn't it need additional training data such as what we feed it and even our conversation history? $\endgroup$
    – Thomas Bilach
    Apr 26 at 22:49
  • 1
    $\begingroup$ @ThomasBilach What exactly are you proposing as an alternative? That we abandon all online public activity? Because that's the only individual action I can see that would affect this. $\endgroup$
    – mkt
    Apr 27 at 5:05
  • $\begingroup$ @ThomasBilach I am not sure how to respond to "isn't that enough?". Your question can be interpreted in different ways. You have the purely factual question in the title. But there also seem to be suggestions to a question/discussion about whether or not this situation is good or not. So is the "isn't that enough?" about that discussion or about the technical point that GPT is not trained enough and needs fine tuning with user interactions to make it more practical in a wider setting? In both cases it is not clear to me why this is meta about the website here. You might explain that as well $\endgroup$
    – Sextus Empiricus
    Apr 27 at 5:56
  • $\begingroup$ @mkt The alternative is to stop using it. I found it interesting that the new “GPT Business” plan will allow users to opt out of the “model training” process. By default it appears the user is the product. $\endgroup$
    – Thomas Bilach
    Apr 27 at 6:17
  • $\begingroup$ @SextusEmpiricus I think the question is clear enough. It was reopened after some consideration. I think this is a conversation worth having, or maybe not. $\endgroup$
    – Thomas Bilach
    Apr 27 at 6:19
  • 1
    $\begingroup$ @ThomasBilach my answer was purely about the title question "Are we free labor for OpenAI?" which I interpret as a question about a fact. Regarding the additional discussion I am unsure what it is actually about and how it connects to cross validated. If there is any additional statement that you want to discuss, then I guess that it might be better to spell it out more precisely. $\endgroup$
    – Sextus Empiricus
    Apr 27 at 7:11
  • $\begingroup$ @SextusEmpiricus I would argue it’s spelled out quite precisely. And I believe my comments offer enough explication on the apparent dangers. Regardless, I still learned something. I appreciate your feedback. $\endgroup$
    – Thomas Bilach
    Apr 27 at 8:50
  • 2
    $\begingroup$ @ThomasBilach You are only repeating yourself saying that you argue that the question is spelled out precisely, but you do not present those arguments. I would appreciate it if you spell out the exact question mere explicitly. Having a fuzzy discussion about different topics at the same time only benefits people who want to keep a status quo of disagreement, and have no discussion in order to resolve or clarify issues. A few minutes ago, I have edited my answer incorporating some of your comments. But I still feel like I am aiming at a moving target and hope that you can make it stop moving. $\endgroup$
    – Sextus Empiricus
    Apr 27 at 8:58
  • 1
    $\begingroup$ I’d be careful about the legal claims here. An amateur soccer player in the US generally has a right of publicity allowing them to prevent the unauthorized commercial use of their likeness. (And if the R Studio company had an ad saying “Sextus Empiricus says R is the best language for StackExchange”, you would have a right of publicity to prevent unauthorized commercial use of your name.) The EU has similar rights from GDPR, and maybe elsewhere. But I agree that our work here is more like fun and less like labor. $\endgroup$
    – Matt F.
    Jun 26 at 14:33

You must log in to answer this question.

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