[Hizonner]

You also might want to "clarify" that it is not open source (and neither are any of the other "open source" models). If you want to call it something, try "open weights", although the usage restrictions make even that a HUGE FUCKING STRETCH.

Also, everybody should remember that these models are not copyrightable and you should never agree to any license for them...

[DHolzer]

When I read "open source" i thought they actually are doing open source instead of "open weights" this time. Surely they would expect to be called out on hackernews if they label it incorrectly...

Thanks for pointing that out @Hizonener

[Morphiak]

A personal bugbear is the AI fascination with calling themselves open source, virtue signalling I guess. Open weights is exactly right. Source code and arguably more important datasets are both required to replicate the work, which is more in the spirit of open source (and science). I think Meta is especially egregious here, given their history.

Never underestimate the value of getting hordes of unpaid workers to refine your product. (See also React, others)

[andybak]

> virtue signalling

I'd prefer "false advertising" - it's more direct and without the culture war baggage.

[seanw444]

"Open source" is perceived as a virtue, and their claim is false. Thus false virtue claim. Or... virtue signaling.

[HPsquared]

Indeed. The data is the main "information source" from which the model is trained.

[astrange]

It's certainly not true that models are not copyrightable; databases have copyright protection if creativity was involved in creating them.

That said, I don't think outputs of the model are derivative works of it, any more than the model is a derivative of its training data, so it's not clear to me they can actually enforce what you do with them.

[filleokus]

> It's certainly not true that models are not copyrightable; databases have copyright protection if creativity was involved in creating them.

Are you talking about https://en.wikipedia.org/wiki/Database_right or plain old copyright?

I'm no IP lawyer, but I've always thought that copyright put "requirements" on the artefact (i.e the threshold of originality), not the process.

In my jurisdiction we have database rights, meaning that you get IP protections for the artefact based on the work put into the process. For example a database of distances between adress pairs or something is probably not copyrightable, but can be protected under database rights if enough work was done to compile the data.

EDIT: Saw in another place in thread speaking about the https://en.wikipedia.org/wiki/Sweat_of_the_brow doctrine, relates to Database rights. (Neither of which notably are not applicable in the U.S)

[kube-system]

These models are a product of much more creativity than simply a list of phone numbers in a phone book. I don't see how they wouldn't meet the modicum of creativity required for US copyright protection.

[Hizonner]

The software that creates the model is the product of creativity. The model itself is the product of mechanically applying that software to datasets that are (a) assembled with minimal, if any creativity, and (b) definitely not assembled with any eye to the specific form of the resulting model. The whole point is to get the software to form the model without you having to worry about what the result is going to look like. So you can't turn around and claim that the model is a creative work because of the choice of training data.

The only thing that's really specified about the model itself is its architecture, which is (1) dictated by function, and (2) usually deeply stereotyped.

[kube-system]

> mechanically applying that software to datasets that are (a) assembled with minimal, if any creativity, and (b) definitely not assembled with any eye to the specific form of the resulting model.

Fair enough, but those datasets are also primarily copyrighted material. If the software here merely transforms the input material (which I agree it does), then the output is a derivative work.

[Hizonner]

But the pitch-shifted song is still recognizably a creative work. It has identifiable, humanly comprehensible forms of all the original creative elements that Swift originally put into it (plus I guess a de minimis amount of extra creativity from the choice to pitch shift it).

If I take a string of data from a true hardware RNG, XOR it with a Taylor Swift song, and throw away the original random stream, is the resulting fundamentally random bit string still a derivative work of the song? As with the ML model, you can't recognize the song in it. And as with at least some training examples in the inputs of most ML models, you can't recover the song from it either.

It feels like the test for whether X is derivative for copyright purposes should include some kind of attention to whether X is a creative work at all. Maybe not, but then what test do you use?

I do recognize the possibility that the models might not themselves be eligible for copyright as independent works, yet still infringe copyright in the training inputs. It seems messy, but not impossible.

... and as I said elsewhere, it's also messy that while you generally can't recover every training input from the model, you can usually recover something very close to some of the training inputs.

[astrange]

There's a lot of effort involved in the training runs too, and you might be able to get away with saying the ops engineers made creative choices too (of which checkpoints were good etc.)

Since it costs millions to produce one of these models, it's not just taking the software and running it to compile them.

[andymcsherry]

I get the sentiment, but one of their models, albeit the worst one, is licensed under Apache without usage restrictions. The source to run the models is also open source.

[astromaniak]

> it is not open source

It would be nice here if you give some examples of what you call open source model. Please ;) Because the impression is that these things do not exist, it's just a dream which does not deserve such a nice term..

[Hizonner]

As far as I know, none have been released. And it doesn't even really make sense, because, as I said, the models aren't copyrightable to begin with and therefore aren't licensable either.

However, plenty of open source software exists. The fact that open source models don't exist doesn't excuse attempts to falsely claim the prestige of the phrase "open source".

[kube-system]

> the models aren't copyrightable to begin with

What criteria for copyright protection are they missing?

[astromaniak]

> As far as I know, none have been released.

I can tell you a secret. What you call 'open source' models are impossible. Because massive randomness is a part of training process. They are not reproducible. Having everything you cannot even tell if the given model was trained on the given dataset. Copyright is a different thing.

And a bad news, what's coming is even worst. Those will be the whole things with self awareness and personal experience. They can be copied, but not reproduced. More over, it's hard or almost impossible to detect if something undeclared was planted in their 'minds'.

All together means 'open source' model in strict interpretation is a myth, great idea which happen to be not. Like Turing test.

> However, plenty of open source software exists.

Attempt to switch topic detected.

PS: as for that massive downvote, I even wasn't rude, don't care. This account will be abandoned soon regardless, like all before and after.

[jillesvangurp]

> models aren't copyrightable to begin with

You are wrong about that. It's a file with numbers. Which makes it a database or dataset and very much protected by copyright. That's why licenses are needed. For the phone book, things like open street maps, and indeed AI models.

> The fact that open source models don't exist

The fact that many people (myself included) routinely download and use models distributed under OSI approved licenses (Apache V2, MIT, etc.) makes that statement verifiably wrong. And yes, I do check the license of stuff that I use as I work with companies that care about such matters.

> As far as I know ...

Now you know better.

[JimDabell]

> You are wrong about that. It's a file with numbers. Which makes it a database or dataset and very much protected by copyright. That's why licenses are needed. For the phone book, things like open street maps, and indeed AI models.

This is only true in jurisdictions that follow the sweat of the brow doctrine, where effort alone without creativity is considered enough for copyright. In other places, such as the USA, collections of facts are not copyrightable and a minimal amount of creativity is required for something to qualify as copyrightable. The phone book is an example that is often used, actually, to demonstrate the difference.

https://en.wikipedia.org/wiki/Sweat_of_the_brow

[Hizonner]

> Which makes it a database or dataset and very much protected by copyright.

Not every collection of numbers is a database, and a database is not the same thing as a dataset.

Databases have limited copyright-like protection in some places. Under TRIPS, that extends to only databases that are "creative by virtue of the selection or arrangement of their contents" or something along those lines. In the US they talk specifically about curation.

ML models do not meet either requirement by any reasonable interpretation.

> The fact that many people (myself included) routinely download and use models distributed under OSI approved licenses (Apache V2, MIT, etc.) makes that statement verifiably wrong.

The "source code" of an ML model is most reasonably interpreted as including all of the training data, which are never, ever available.

Now you know better.

[On edit: By the way, the people creating these works had better hope they're outside copyright, because if not, each one of them is a derivative work of (at least some large and almost impossible to identify subset of) its training data, so they need licenses from all the copyright holders of that training material, which few of them have or can get.]

[kube-system]

If we stop unnecessarily anthropomorphizing software, I think it is plainly obvious these are derivative works. You take the training material, run it through a piece of software, and it produces an output based on that input. Just because the black box in the middle is big and fancy doesn't mean that somehow the output isn't a result of the input.

However, transformativeness is a factor in whether or not there is a fair-use exception for the derivative work. And these models are highly transformative, so this is a strong argument for their fair-use.

[Hizonner]

Maybe, but...

"Fair use" is pretty much entirely a US concept, and similar concepts in other countries aren't isomorphic to it.

The model does have a radically different form from its inputs. So you could easily imagine that being "transformative enough" for US fair use. A lot of the other fair use elements look pretty easy to apply, too. Although there's still the question of whether all the intermediate copies you made to create the model were fair use...

In fact, I'll even concede that a court could find that a model wasn't a derivative work of its inputs to begin with, and not even have to get to the fair use question. The argument would be that the model doesn't actually reproduce any of the creative elements of any particular training input.

I do think a finding like that would be a much bigger stretch than a finding that the model was copyrightable. I could easily see a world where the model was found derivative but was not found copyrightable. And it's actually not clear to me at all that the model has to be copyrightable to infringe the copyright in something else, so that's another mess.

Somewhat related, even if the model itself isn't infringing, it's definitely possible to have most models create outputs that are very similar to (some specific examples in) their training data... in ways that obviously aren't transformative. Outputs that might compete with the original training data and otherwise fail to be fair use. So even if the model is in the clear, users might still have to watch out.

[simonw]

I'm personally comfortable calling a model "open source" if the license is compatible with the https://opensource.org/ definition.

The Llama models aren't. Some of the Mistral models are (the Apache 2 ones). Microsoft Phi-3 is - it's MIT.

[dagaci]

Open source must include source material so that another can reproduce that the model. I would expect that to be a minimum.

[simonw]

I agree, but that can't happen with the vast majority of these models because they're trained on unlicensed data so they can't slap an open source license on the training data and distribute it.

I've decided to draw my personal line at Open Source Initiative compliance for the license they release the model itself under.

I respect the opinion that it's not truly open source unless they release the training data as well, but I've decided not to make that part of my own personal litmus test here.

My reasoning is that knowing something is "open source" helps me decide what I legally can or cannot do with it when building my own software. Not having access to the training data downs affect my legal rights, it just affects my ability to recompile myself. And I don't have millions of dollars of GPUs so that isn't so important to me, personally.

[Hizonner]

> that can't happen with the vast majority of these models because they're trained on unlicensed data

Tough beans? There's lots of actual software that can't be open source because it embeds stuff with incompatible restrictions, but nobody tries to redefine "open source" because of that.

... and, on a vaguely similar-flavored note, you'd better hope that the models you're using end up found to be noninfringing or fair use or something with respect to those "unlicensed data", because otherwise you're in a world of hurt. It's actually a lot easier to argue that the models aren't copyrightable than it is to argue that they're not derivative of the input.

> I've decided to draw my personal line at Open Source Initiative compliance for the license they release the model itself under.

You're allowed to draw your personal line about what you'll use anywhere you want, but that doesn't mean that you should try to redefine "open source" or support anybody who does.