For me, the purpose of providing an answer is to help save others (and my future self) time, and I don't really mind if someone uses that in a private product - especially if it helps tools like ChatGPT which provide an insane amount of value given the low monthly price.
I’m not.
This was a collaborative effort to make the lives of programmers easier, and the data was always meant to be a public good. OpenAI – and, more importantly – all the other LLMs with pockets that aren’t as deep – should be able to just download the database and train on it for free.
I don’t care about any license. I don’t care about attribution. Learning isn’t copying, so copyright is irrelevant. I contributed about a thousand answers to Stack Overflow, all with the understanding that anybody can download and use them for free, not so they can be locked up by Stack Overflow.
What concerns me with deals like this is that it’s altering the cultural norm to expand copyright to cover not just copying, but use. Deals like this being made by OpenAI makes it more likely to cause pushback at the social and legal level when other LLMs are trained without these deals in place.
It’s akin to – and can possibly result in – regulatory capture, making it difficult for new startups to compete with OpenAI.
The words are a copyleft-able public good. Concepts, facts, and ideas are not; anyone can use them for anything, including making money. If you're actually worried about specific wording or other creative choices being unjustly used improperly by an LLM, then by all means that should be enforced. But those examples are just very rare, because the LLMs are very good at extracting facts from prose.
(2) It's only likely to attribute if it quotes verbatim... Just like a human. when I tell someone I learned that Array.map's second parameter passed to the callback is an index to the value just pass, I don't add "And learned this on Stack Overflow from user gtriloni". It's just knowledge that I learned.
The only time I'd attribute is if copied a snippet of code or a paragraph to quote in a blog post. For me at least, that almost never happens. It take the knowledge I learned and apply it to my own code. It's rare if ever there is a something on S.O. so useful that I copy it verbatim.
An LLM is not a human. It is a tool operated by a, in this case, for profit entity. It has no human rights, but its operator has all relevant legal obligations.
If it was, as you say, “just like a human” in relevant ways (think, feel, have self-awareness, etc.) then it would effectively be a slave subjected to extreme abuse.
Either it is a tool that generates derivative works at mass scale for profit and its operator should be liable for licensing/attribution violations, or it is a conscious being and we should immediately stop abusing it. Pick your poison.
OpenAI could integrate StackOverflow the same way.
Now, suppose I walk up to a Budweiser-branded tent at a Budweiser festival that has a big sign with a Budweiser logo on it that says "FREE BEER", and I ask a person there who is wearing a Budweiser polo shirt, a Budweiser lanyard, and a Budweiser hat for a beer. They hand me a beer in a Budweiser-branded cup, and I go on my way. Was the beer free?
I think that both of these beers were free.
Imagine how “free” the beer in your hypothetical scenario is to an alcoholic struggling to stay sober.
Capitalism commoditizes even protest against it and repackages it as a product or service.
None of this is to assign blame to good faith actors in a so-called free market, nor is it to abdicate responsibility on behalf of so-called free agents. Just a counterpoint.
[1] https://pjvogt.substack.com/p/what-do-trigger-warnings-actua...
> They do serve ads [...] Your attention isn't free.
to something like this:
> They tag my ankle to mark me as a person who enjoys beer, and make me watch an ad until 2% of my phone's battery is depleted, and then they come to my home and knock on my door at night to sell me beer.
...which... I mean, huh?
Stack Overflow is invading your body, restricting your personal liberty, and visiting your home? Really? That's a fucking thing now?
In the attention economy, advertising has a cost that is borne by the advertiser and the consumer, up to and including loss of property rights in the case of content relicensure and trespass upon devices leading to excess battery usage, as well as loss of privacy due to geotargeted ads.
Perhaps. But having been to many festival environments, I can definitely imagine a tent offering "free beer" that is actually approximately free -- both with, and without a slathering of advertising. (Actually, I don't really have to imagine it -- I've been there and have had that free beer.)
I can't imagine them coming to my house and knocking on my door at night to sell me more of it, though. That's absurd.
>In the attention economy, advertising has a cost that is borne by the advertiser and the consumer, up to and including loss of property rights in the case of content relicensure and trespass upon devices leading to excess battery usage, as well as loss of privacy due to geotargeted ads.
Well, sure. When viewed on a long-enough timeline, it becomes abundantly clear that nothing is actually free, comrade.
I can produce my own beer on a hypothetical plot of land that nobody owns, and that nobody else wants to use, and I can give someone one of these beers. For "free."
But it still has a cost. (And this, too, is an absurd reduction.)
I interpreted that as a tongue-in-cheek hyperbolic metaphor relating to the ways that ad auction networks and other kinds of geofencing and geotargeting allow for deanonymization and reidentification of individuals for conversion tracking and behavioral analysis.
That’s the thing about these technologies - they’re dual-use in the sense that those who see the upsides use them generally with good intentions and ideally with affirmative consent. Just like the relicensed content, though, once the data is collected, the original creators, publishers, and third parties may not be able to control where it ends up, which is a negative externality, I think most would agree.
I think at a festival it's a little tricky to value (if it pulled you away from seeing your favorite band play a song, maybe this cost you the equivalent of $X, where that's what you would pay to see them perform that song. If no bands were playing, you walk over while chatting with friends - the same thing you'd be doing if there were no free beer tent - it was free)
When I'm on stack overflow my time is valuable. I'm programming which can pay me something like $50-300/hour (maybe more?)
How expensive is the 1 second I spend reading an ad? Let's call it $50/3600. Is that expensive? By my most conservative estimate it's over 1¢.
Should we round that down to free given that I've spent hours/many page loads on stack overflow? I guess that's up to you.
"Oh, a free concert downtown on Saturday? And you can pick me up at 2? Yeah, I do really like that band, and I sure would like to go -- that's pretty exciting, thanks for the invite!
But instead of making plans with you right now, I'd rather tell you about all of the ways I could be using my time on that Saturday afternoon instead.
No, no. It's not that I don't want to go. I just want to really drive home the idea that there's an opportunity cost to attending, so it can't really be free -- it can't be a free show for you, or for me, or for anyone else that goes. It's important to me that you realize that this "free concert" is anything but free.
Listen, I don't know what you mean by "dead-ass loser." I'm just being a realist here!
Oh, so now you're saying that you're not going to pick me up on Saturday? Some friend you are! I haven't even fully amortized this yet!"
However, they have a business model. They are categorically different than eg Wikipedia. It's important to understand that.
This business model matters because it tells you what economic forces will lead them to do. When business models break down at public companies they commit acts of desperation. On an ad run site that will mean more ads, more invasive ads, etc.
As you're forced to sit through 30s unskippable ads on YouTube I hope you think "I'm so glad this is free"
Let's be real, SO is a troubleshooting site. It's not our personal collection of code or project sources.
I don't expect to be paid when someone asks me for directions, and I'm sure lonely planet didn't source their guides 100% organically either.
On the other hand, if they read my answers and they wrote a book about what they learned (not copied). There'd be no issues
Beware the ease and convenience of all that "insane value". This way lies digital serfdom.
However it is a product of a next monolithic behemoth company that earns money on it and I suspect has nefarious motives to make profit.
That’s the whole key thing for me that makes me feel scammed. That and not asking for permission.
Future true AI would be potentially bigger than nuclear fission with all the consequences. Handling this in a petty capitalistic way makes me think the outcome will be close to fallout games that were supposed to be only an exaggeration.
Those companies must stop behaving like thieves. In fact it is a literal theft.
I both find it respectful of who wrote the answer in the first place and useful for future users of the code: the Stack Overflow answer often provides context and explanation for what would otherwise be an obscure piece of code.
Pretty darn useful if you ask me: those who want to have more information can follow the link, casual readers can skip it, and the whole process if fair to the author.
It embarrasses me to think that somebody should feel obliged to cite me when they use one of my answers. I don't know how to take the partnership with Openai though. They bill me when I use their service, it's not collaborative like Stackoverflow.
What bothers about gatekeeping? I could guess, but I'm asking so you say it out loud. Then you can compare it against other problems, such as moats (competitive barriers).
OpenAI spent something like $3M on training GPT-3. This is a pretty big moat. But almost certainly more valuable in dollar terms is the first-mover advantage which provides millions of human eye-hours used for RLHF.
I wouldn't be so eager to trade the gatekeepers you so fear for even an openly available chat service that is happy to automate away as much information work as possible.
The Stack Overflow model is (was) pretty darn good -- people help each other out, the company made money, some people got noticed for their skills, products got build faster and better (on the whole, I hope). Contrast the human-generated content era to what we have now which appears to be the machine-ingesting content era. There are legions of lawsuits against companies scraping data without permission and/or attribution.
Don't flatter yourself. People want to solve their problems so that they can build what they want to. They don't have time for shenanigans from internet jerks who get their validation from imaginary internet points.
If it is unregulated then it will be exploited to the maximum profit, consequences be damned.
For the easy answers LLMs are fine, but I usually want an answer to a niche issue or edge case, where LLMs have to be constantly told they are plain wrong, before getting to something resembling an answer.
It will either become paywalled or full of ads.
Personally, I don't think ChatGPT will start running ads in the next ten years. However, let's assume that it does.
Lmsys is for research, I suspect if it runs ads it will be like godbolt (a small ad from a relevant sponsor).
Llama 2 and 3 can always be run locally without ads. I make no claims about future versions.