cornel_io

Whether or not copyright applies at all to model training is an entirely open question, and where rulings have come down, it's likely closer to these situations being fair use (e.g. the Google Book's case, which was ruled transformative and not a direct replacement for the works in question).

The reality is, these models don't copy or distribute anything directly, which makes applying copyright a bit of a stretch. Many people feel like it is a use that should have some sort of IP law applying to it, which is why I think there's some chance that courts or legislators will decide to screw the letter of existing law and just wedge new interpretations in, but it's not super simple: they'd have to thread the needle and not make things like search illegal, and that's tricky. Besides that, these models are out there, they're useful, and if they're ruled infringing they'll just be distributed illegally anyways.

I don't envy the people who will have to decide these cases, I suspect what's better for the world overall is to leave the law as-is and clarify that fair use holds (nobody will stop publishing content or code just because AI is slurping it up, a few weirdos like the article author excepted), but there are going to be a lot of pissed off people either way...

That was true several years ago, yes, back when GitHub Actions was super new and pretty crappy...

They used to say that about email addresses. But hold my beer and I'll be back in like a month with some buggy half-assed crap that kinda does the job and only occasionally crashes the system!

I have literally never seen a restaurant anywhere in the world that charges for tap water, and I eat out constantly. Is this a thing now somewhere?

If you are selling a product to the public and get a legal threat from a big company, you need to get a lawyer, whether or not you have one to start with. Period, end of story. OP mentions it's their livelihood, so it's worth it.

It doesn't much matter because capitalism is here to stay, and people like it.

And now the average worldwide lifespan is 72, higher if you look at countries that are fully industrialized.

Transitions suck for the people left behind, but that doesn't mean that progress is bad, it can really help people overall.

No. Hiring like crazy when interest rates are low and cutting back when they are not is proper management.

The economy correctly drives company behavior, and that's not a CEO's fault.

I have yet to hear a copyright lawyer who actually understands how these tools work say anything other than that it's going to be fair use; the ones that I have seen comment on it as such seem to have extremely misunderstood how the tech works and how the images are used, they've just regurgitated the arguments that the artists opposed to it make (about how it copy and pasted images, etc).

Maybe I'm wrong, also not a lawyer, but I've heard from enough at this point (including ones that I've paid and are actually our lawyers as we evaluate internal use of the tools) that I definitely wouldn't bet on these suits succeeding.

And Google was sued many times back in the day for the same thing, because of their indexing of the web. They won out under fair use for the exact same reasons that all of these lawsuits will fail, the way these images are used falls very clearly within the dead center of fair use, and is probably even less questionable than Google's, since Google really does store and display direct copies of pieces of the content that it organizes.

If you have a GCP, AWS, or Azure account already and know Javascript, you should be able to whip up a serverless cron task real quick. IMO the simplest setup is probably GCP via Firebase (which makes deployment super simple once you install the CLI): https://firebase.google.com/docs/functions/schedule-function...

You won't spend much at all, it's fractions of a penny per call and I think there's a free tier.

I absolutely agree that game dev is some of the most nasty programming that's out there.

Mid-tier game programmers are worse than mid-tier webdevs, though, having bounced between the two subfields a lot. The most difficult stuff is more difficult in games, which is why the 99% people there are actually better than the 99%ers elsewhere, but the median work is meh, and that makes up the bulk of the work. And the devs...the median in games is not very good, the big companies employ a lot of extremely mediocre devs who desperately need the 99%ers closely managing them to get even passable work out of them.

The pay difference is just because it's a more desirable job than doing boring webdev, and young programmers think of it as awesome because they imagine they'll be doing the stuff that Carmack does. But that's what the top folks do. In reality most are going to be writing UI integration code and content-reviewing and fixing up busted PRs from artists.

It sounds like you have a deep (and IMO unjustified and motivated by wrong or weirdly interpreted statistics) hatred of law enforcement, probably from either personal experience or political activism. That's your prerogative, of course, but it's a really fringe position and most people would disagree.

The rich will get richer and the poor will get richer at a slightly slower pace, as always, but quality of life for all will increase.

AI will probably vastly increase both rates of change, more for the rich than the poor, but things will be better in 20 years for the poor with AI than without it. They will just be even further behind the super rich than they were before, even if they're better off.

Some people would prefer a more equal but worse-for-everyone world. I'm not one of them, but I understand that it's a POV, and I can even semi-understand some of the arguments (sim to Communist talking pts, there's a real effect where people care more about how they rate vs other people rather than their absolute level of wealth/comfort). I fully reject them, but to each their own.

I still hate it, because it's a unicorn UX that doesn't follow any pattern that other software uses, and IMO is not an improvement in any way over the other standard options (normal menus). There's no special problem that Word faces that other word processors do not, so there's no reason for a screwy and impossible to decipher unless you've already memorized it UI.

If they're done right, they are exactly as rigorous as we'd expect from science, since they are literally the same as randomized control trials. You can do even better with A/B testing because you have much tighter control over inclusion criteria, treatment compliance, and outcome analysis.

And this is sadly true about almost every good A/B testing system out there. They always involve tooling that has to exactly fit your stack, and it's really tough to make a general purpose product that would offer anywhere near the same value.

To some extent, you grow your company and codebase around your A/B testing system, not the other way around, because it has to slot into so many places: deployment, testing, front-end, back-end, analytics, monitoring, etc. Almost no two companies share the same stacks across all of those dimensions, and an A/B testing system that doesn't hook in tightly to every one of those systems is not complete. This is why I always get a bit scared when people think they can just use one of those "drop-in" services and be done with it: yeah, great, you can now fiddle your JavaScript from some third-party website, but you're going to have a big project ahead of you passing group assignments out to all of the different systems that will need to know about them, and almost guaranteed some part of your stack will not have a library available from the service you picked so you're going to be writing your own REST wrapper, and dammit they don't document that API very well and it seems to be responding differently since the last update, and man it'd be a lot easier if I could just pipe results straight from my own service into Big query rather than running a daily user export, and damn, their dashboard doesn't let me set exclusion criteria on my own metrics, I have to send activation events now, and etc, etc. By the end you've basically built your own A/B test system from scratch, you've just paid someone else to do the "int myGroup = Random.next()" call, which is the easiest part to build.

> Folks who grew up in affluent households will venmo for the salsa.

This is a broad generalization that's actually the inverse of what I've seen, and I'm willing to bet I've interacted with an order of magnitude more "grew up affluent" types than the author, having grown up in one of the richest suburbs in the US.

Maybe it depends on the kind of wealth, but in my social circles (both growing up and now) anyone who did something like ask for a $6 Venmo would be laughed out of the room, and if they were serious people would Start Talking about them. That is Just Not Done, people pick up small tabs and (sometimes) split large ones and don't think about it at all, to do otherwise is considered extremely crass. Any fights over tabs are usually of the alpha male variety where paying the tab is a sign of dominance (and I've seen those fights get weird, with people ninja-settling in secret before the check shows up just so they can say "it's already taken care of" and then angry fights erupting afterwards because someone else had expected to pay).

Yes, there is a careless and/or arrogant disregard for money, but it does not usually come across as penny pinching and counting what other people owe you, it's more the type of thing where they'll casually assume everyone in the party would be thrilled to go to a $400/person restaurant for a celebration, not for a moment realizing how much of a stretch that is for the younger/less rich in the group and what a hard position it puts them in.

I think the word the author is looking for to pick out the kind of person who sends a Venmo request for a coffee is "asshole", not "grew up wealthy". I feel like they have some axe to grind here and have ended up severely overgeneralizing as a result...

Meredith Whittaker was always an activist, and her being in charge of a big product doesn't change that, it just increases the risk. I don't believe for one second that if push comes to shove and she thinks it's an important cause she won't wield her power to cut through the supposed security of Signal and expose "undesirables" to the light of day and leak all their private info.

I would not trust this product anymore, especially if you are not a hard leftist like her.

And I don't know what you're even getting at with the "assembling working programs" vs "navigating license minutae" stuff. Copilot helps the former, and you're apparently trying to argue that it should be banned because some people are feeling angsty about the latter?

You may hope that a judge considers a really weird and cherry-picked weird edge case to be "normal use", but opposing counsel will argue the opposite, and have a lot of evidence if they actually look at real use.

"I can use a gun to shoot someone" doesn't make guns illegal, even if people do so with some regularity. "I shot someone to make the point that guns can kill people" is worth even less in the eye of the law, and that is literally what you're pointing to here.

The big problem is that while under current law it's pretty clear that this stuff should all be fair use, enough people want it not to be that interpretations and/or laws will possibly change. These laws pretty clearly did not anticipate this sort of thing, so it isn't possible to affirmatively say that anything will hold going forward.

Funneling free money into rent, any way you structure it, will indeed raise prices unless you also build more housing.

Ironically the people fighting the hardest to funnel free money into rent are the same ones fighting the hardest to prevent new housing at all costs, under the delusional theory that "induced demand" means that the law of supply and demand is wrong and the inverse applies.

SF's NIMBY progressives are a weird lot.

Sort of. The error at a square wave's "cliff" does shrink in width as you add higher and higher frequencies to the approximation, but the height of the error does not diminish, so you have to be explicit about what you mean by "converge". If you considered MAX(ABS(approximation(x) - square(x))) as your error metric it would not go to zero even in the limit; for most intents and purposes we don't care about that, though, and in any case the limiting behavior is much nicer for "normal" functions that don't have discontinuities like a square wave.

Yes, I mentioned that. And I think it would be great if GitHub added some sort of scan to see if it was accidentally returning some verbatim code and gave a link to the repo it came from, but the situation with Google was much more clearly harmful: by scraping and presenting verbatim quotes, they are directly siphoning web traffic away from the sites they scrape from, and many of those sites make money based on page views. That is direct economic harm, and the actual impact was significant for many companies, not merely theoretical.

With open source code, the harm is much less tangible, since negligibly few open source projects make money from people going to their GitHub pages because they're searching for code snippets (not zero, but almost). My guess is that an honest quantification would put the lost revenue due to Copilot's existence in the tens to maaaaaybe hundreds of dollars. Courts look at that type of thing, which is why I don't think this will end up being an issue, at least in the US. Europe is wild, who knows what they'll do there, and that's where activists on this topic should most wisely apply pressure, you can always convince someone in government there to throw a spear at a BigCo. You won't take them down, but you may get them to negotiate, and I don't even necessarily think that's a bad thing.

That said, even in the US, if enough people make noise then things could change, so I encourage you to speak to your congressperson (I will be as well, but arguing the other side, because I really do think this is fair use and I'd like to see it enshrined as such explicitly, because this fight is going to be extremely common over the next few decades).

How likely that situation is is disputable. People have cherry picked some cases where by specifying a very particular comment you can get it to (unreliably) reproduce well-known pieces of code, sure. But that is neither the intended use of the product nor the actual way that a single person using the product really employs it, and that really, really matters. Judges are not automatons, and the fact that when and if this goes to court the developers will be able to get up and honestly say "this is a tool developed to help developers create new code that is not merely recreating the functionality of the code we trained the tool on, and all our users use it to create entirely new things" is going to matter when they argue that it is fair use/transformative work.

I do completely understand that a lot of people disagree with me morally, and think that extracting insights from scraping the public web should be illegal. You're free to have that opinion, but I'd recommend you start lobbying your congressman to change the law, because though I'm not a lawyer I hang out with a few who do copyright stuff, and I don't think the law as it stands is on your side. That said, who can say, maybe this will end up bubbling up through many layers of appeals and end up at the Supreme Court someday, this stuff is all certainly wildly outside the bounds of what anyone writing the copyright laws was thinking about back in the 70s (which I think was the latest significant iteration?) so it's fair to say it's a complete gray area.

"Derivative work" is a very specific thing, and it's contrasted with "transformative work" in a way that matters a lot, and fair use intersects heavily with both.

Even when it comes to stuff that seems reaaaaally close to pure derivative: Googling "How long does it take to boil water?" => "If you're boiling water on the stovetop, in a standard sized saucepan, then it takes around 10 minutes for the correct temp of boiling water to be reached. In a kettle, the boiling point is reached in half this time."

That's a verbatim snippet pulled directly from https://unocasa.com/blogs/tips/how-long-to-boil-water, and yet Google exists and continues to do stuff like this under the fair use doctrine despite massive efforts to attack/monetize their service. [To be fair, Google does link results, which probably insulates them because it's less hurtful to the commercial interests of the source; that said, with open source there generally are no commercial interests to hurt (open source attribution will be a tough sell as an actual commercial interest), and that's specifically called out in the law as a factor]

Copilot is even less explicitly at risk IMO, in that it never even stores the text, nor can it reliably retrieve it. I have no idea what makes anyone think it should be more vulnerable than Google.

From the copyright.gov page on fair use (https://www.copyright.gov/fair-use/, worth reading in detail for anyone who cares about this stuff, also has links to a monumental number of cases with shockingly intelligible summaries): "Additionally, “transformative” uses are more likely to be considered fair. Transformative uses are those that add something new, with a further purpose or different character, and do not substitute for the original use of the work."

Copilot without any shadow of a doubt does add something new, with a further purpose, and does not substitute for the original use of any codebase on Github (it can't create any of the codebases in full, without manual guidance so extreme that you'd have to be using the actual original codebase as a reference, so it clearly cannot substitute for a single one of them, and that's what a lawyer will argue, likely successfully).

In the Google vs. Oracle case (see https://www.copyright.gov/fair-use/summaries/google-llc-orac...), a big piece of the fair use finding was that "its value in significant part derives from the value that those who do not hold copyrights, namely, computer programmers, invest of their own time and effort to learn." and "further[s] the development of computer programs”. It's hard to see where Copilot wouldn't fall into that category, as well, and that's precedent on (multiple) appeal.

By my reading this should be a slam-dunk fair use ruling, unless precedent gets really upended, and Butterick is wasting a ton of time and effort for absolutely zero potential gain other than some bragging rights, but to each his own...I guess we all have to grind our axes from time to time.

Absolutely incorrect, fair use applies to *reproducing* small volumes of work, not analyzing it. If I published an article gleaning some conclusion based on an analysis of 10,000 issues of the New York Times, that would still 100% be fair use; similarly, Google is absolutely allowed to publish word count metrics based on their scanned book repo, even though publishing the books themselves is not fair use. You are trying to read something into the fair use doctrine that is absolutely not there (to the extent that anything is there, which very little is other than "I'll know it when I see it" and prior case law, unfortunately).

And it seems pretty clear to me that it is fair use, because it's not merely reproducing or creating a derivative work, but actually extracting patterns and modeling the works in a way that is intended to be used to create new and unrelated works. The fact that an occasional piece of code here and there might be reproduced verbatim is no different than e.g. Cliffs Notes occasionally quoting a passage, and Cliffs Notes are a well established case of fair use that to me, at least, seems even closer to "the line" than Copilot or Stable Diffusion.

FTA: "On the other hand, maybe you’re a fan of Copi­lot who thinks that AI is the future and I’m just yelling at clouds. First, the objec­tion here is not to AI-assisted cod­ing tools gen­er­ally, but to Microsoft’s spe­cific choices with Copi­lot. We can eas­ily imag­ine a ver­sion of Copi­lot that’s friend­lier to open-source devel­op­ers—for instance, where par­tic­i­pa­tion is vol­un­tary, or where coders are paid to con­tribute to the train­ing cor­pus."

This is the same argument that people use about Stable Diffusion, and it's kinda meh to me...I guess it'd be nice to allow people to opt-out, like Stable Diffusion is doing with their next versions, especially since a negligible percentage of people will do so and it won't affect the models at all. But yes, it basically is yelling at clouds. Opt-in would cripple models, and some people would make them anyways and just keep them secret, which is worse for the world. And at the end of the day, this really does just seem to me like a fair use of stuff that you've published on the Internet for anyone with a browser to look at. The AI models of the future are going to gobble the whole net up, and if you don't want them ingesting your stuff and learning from it, then you just shouldn't make it freely available.

If OpenAI/GitHub/MS really wanted to get ahead of this and head off any potential legal conflict, they could always just open source the models and weights, which would be in line with the name "OpenAI"...it would be a minor project to scrape all the correct headers to add to a license file(s), but negligible compared to the many millions of dollars spent on training.

Computers are not as intelligent as humans right now at coding. So it's no surprise that they can't improve code (let alone their own).

If we ever get them there, then it's likely that the usual resourcing considerations will come into play, and refactoring/optimization/redesign will be viable if you throw hours at them. But unlike with human optimization, every hour spent there will increase the effectiveness of future optimizations.