Big picture, I'll always believe we dodged a huge bullet in that "AI" got big in a nearly fully "open-source," maybe even "post open-source" world. The fact that Meta is, for now, one of the good guys in this space (purely strategically and unintentionally) is fortunate and almost funny.
Another funny possibly sad coincidence is that the licenses that made open source what it is will probably be absolutely useless going forward, because as recent precedent has shown, companies can train on what they have legally gained access to.
On the other hand, AGPL continues to be the future of F/OSS.
MIT is also still useful; it lets me release code where I don't really care what other people do with it as long as they don't sue me (an actual possibility in some countries)
The US, for one. You can sue nearly anyone for nearly anything, even something you obviously won't win in court, as long as you find a lawyer willing to do it; you don't need any actual legal standing to waste the target's time and money.
Even the most unscrupulous lawyer is going to look at the MIT license, realize the target can defend it for a trivial amount of money (a single form letter from their lawyer) and move on.
You can sue for damages if they have malware in the code, there is no license that protects you from distributing harmful products even if you do it for free.
And illegally too. Anthropic didn't pay for those books they used.
It's too late at this point. The damage is done. These companies trained on illegally obtained data and they will never be held accountable for that. The training is done and they got what they needed. So even if they can't train on it in the future, it doesn't matter. They already have those base models.
Then punitive measures are in order. Add it to the pile of illegal, immoral, and unethical behavior of the feudal tech oligarchs already long overdue for justice. The harm they have done and are doing to humanity should not remain unpunished.
And the legality of this may vary by jurisdiction. There’s a nonzero chance that they pay a few million in the US for stealing books but the EU or Canada decide the training itself was illegal.
Then the EU and canada just won't have any sovereign LLMs. They'll have to decide if they'd rather prop up some artificial monopoly or support (by not actively undermining) innovation.
It’s not going to happen. The EU is desperate to stop being in fourth place in technology and will do absolutely nothing to put a damper on this. It’s their only hope to get out of the rut.
If I can reproduce the entirety of most books off the top of my head and sell that to people as a service, it's a copyright violation. If AI does it, it's fair use.
>If I can reproduce the entirety of most books off the top of my head and sell that to people as a service, it's a copyright violation. If AI does it, it's fair use.
Assuming you're referring to Bartz v. Anthropic, that is explicitly not what the ruling said, in fact it's almost the inverse. The judge said that output from an AI model which is a straight up reproduction of copyrighted material would likely be an explicit violation of copyright. This is on page 12/32 of the judgement[1].
But the vast majority of output from an LLM like Claude is not a word for word reproduction; it's a transformative use of the original work. In fact, the authors bringing the suit didn't even claim that it had reproduced their work. From page 7, "Authors do not allege that any infringing copy of their works was or would ever be provided to users by the Claude service." That's because Anthropic is already explicitly filtering out results that might contain copyrighted material. (I've run into this myself while trying to translate foreign language song lyrics to English. Claude will simply refuse to do this)[2]
They should still have to pay damages for possessing the copyrighted material. That's possession, which courts have found is copyright violation. Remember all the 12 year olds who got their parents sued back in the 2000s? They had unauthorized copies.
I don't know what exactly you're referring to here. The model itself is not a copy, you can't find the copyrighted material in the weights. Even if you could, you're allowed under existing case law to make copies of a work for personal use if the copies have a different character and as long as you don't yourself share the new copies. Take the Sony Betamax case, which found that it was legal and a transformative use of copyrighted material to create a copy of a publicly aired broadcast onto a recording medium like VHS and Betamax for the purposes of time-shifting one's consumption.
Now, Anthropic was found to have pirated copyrighted work when they downloaded and trained Claude on the LibGen library. And they will likely pay substantial damages for this. So on those grounds, they're as screwed as the 12 year olds and their parents. The trial to determine damages hasn't happened yet though.
This was immediately my reaction as well, but I'm not a judge so what do I know. In my own mind I mark it as a "spice must flow" moment -- it will seem inevitable in retrospect but my simple (almost surely incorrect) take is that there just wasn't a way this was going to stop AI's progress. AI as a trend has incredible plot armor at this point in time.
Is the hinge that the tools can recall a huge portion (not perfectly of course) but usually don't? What seems even more straight forward is the substitute good idea, it seems reasonable to assume people will buy less copies of book X when they start generating books heavily inspired by book X.
But, this is probably just a case of a layman wandering into a complex topic, maybe it's the case that AI has just nestled into the absolute perfect spot in current copyright law, just like other things that seem like they should be illegal now but aren't.
Yea, that dipshit judge just opened the flood gates for more problems. The problem is they don't understand how this stuff works and they're in the position of having to make a judgement on it. They're completely unprepared to do so.
Now there's precedent for future cases where theft of code or any other work of art can be considered fair use.
So interestingly, free meant autonomy for Stallman and the original proponents of "copyleft" style licenses too. But autonomy for end-users, not developers. But Stallman et al believed the copyleft style licenses maximized autonomy for end-users, rightly or wrongly, that was the intent.
I read through and I think that the analysis suffers from the fact that in the case when the modifier is the user it's fine.
Free software refers to user freedoms, not developer freedoms.
I don't think the below is right:
> > Notwithstanding any other provision of this License, if you modify the Program, your modified version must prominently offer all users interacting with it remotely through a computer network (if your version supports such interaction) an opportunity to receive the Corresponding Source of your version by providing access to the Corresponding Source from a network server at no charge, through some standard or customary means of facilitating copying of software.
>
> Let's break it down:
>
> > If you modify the Program
>
> That is if you are a developer making changes to the source code (or binary, but let's ignore that option)
>
> > your modified version
>
> The modified source code you have created
>
> > must prominently offer all users interacting with it remotely through a computer network
>
> Must include the mandatory feature of offering all users interacting with it through a computer network (computer network is left undefined and subject to wide interpretation)
I read the AGPL to mean if you modify the program then the users of the program (remotely, through a computer network) must be able to access the source code.
It has yet to be tested, but that seems like the common sense reading for me (which matters, because judges do apply judgement). It just seems like they are trying too hard to do a legal gotcha. I'm not a lawyer so I can't speak to that, but I certainly don't read it the same way.
I don't agree with this interpretation of every-change-is-a-violation either:
> Step 1: Clone the GitHub repo
>
> Step 2: Make a change to the code - oops, license violation! Clause 13! I need to change the source code offer first!
>
> Step 1.5: Change the source code offer to point to your repo
This example seems incorrect -- modifying the code does not automatically make people interact with the program over a network...
"free software" was defined by the GNU/FSF... so I generally default to their definitions. I don't think the license falls afoul of their stated definitions.
That said, they're certainly anti-capitalist zealots, that's kind of their thing. I don't agree with that, but that's besides the point.
It's not really "virtually impossible to comply with". It's very restrictive, yes, but not hard to comply if you want to.
And yes, it is an EULA pretending to be a license. I'd put good odds on it being illegal in my country, and it may even be illegal on the US. But it's well aligned with the goals of GNU.
Hell is, by design, a consequence for poor people. (People could literally pay the church to not go to hell[0]). Rich people have no consequences whatsoever, let alone poor people consequences.
Not "by design", as historically the hell came first. It was only much later that they catholic church started talking about the purgatory and the possibility of reducing your punishment by paying money.
The people running AI companies have figured out that there is no such thing as hell. We have to come up with new reasons for people to behave in a friendly way.
We already have such reasons. Besides, all religious "kindness" was never kindness without strings attached, even though they'd like you to think that was the case.
Open source may be necessary but it is not sufficient. You also needed the compute power and architecture discoveries and the realisation that lots of data > clever feature mapping for this kind of work.
A world without open source may have given birth to 2020s AI but probably at a slower pace.
Dont make the mistake of anthropomorphizing Mark Zuckerberg. He didnt open source anything because he's a "good guy", he's just commoditizing the complement.
The "good guy" is a competitive environment that would render Meta's AI offerings to be irrelevant right now if it didnt open source.
The reason Machiavellianism is stupid is that the grand ends the means aim to obtain often never come to pass, but the awful things done in pursuit of them certainly do. So the motivation behind those means doesn't excuse them. And I see no reason the inverse of this doesn't hold true. I couldn't care less if Zuckerburg thinks open sourcing Llama is some grand scheme to let him to take over the world to become its god-king emperor. In reality, that almost certainly won't happen. But what certainly will happen is the world getting free and open source access to LLM systems.
When any scheme involves some grand long-term goal, I think a far more naive approach to behaviors is much more appropriate in basically all cases. There's a million twists on that old quote that 'no plan survives first contact with the enemy', and with these sort of grand schemes - we're all that enemy. Bring on the malevolent schemers with their benevolent means - the world would be a much nicer place than one filled with benevolent schemers with their malevolent means.
> The reason Machiavellianism is stupid is that the grand ends the means aim to obtain often never come to pass
That doesn't feel quite right as an explanation. If something fails 10 times, that just makes the means 10x worse. If the ends justify the means then doesn't that still fit into Machiavellian principles? Isn't the complaint closet to "sometimes the ends don't justify the means"?
You have to assume a grand ends is achievable through some knowable means. I don't see any real reason to think this is the case, certainly not on any sort of a meaningful timeframe. And I think this is even less true when we consider the typical connotation of Machiavellianism, which is through 'evil' actions.
It's extremely difficult to think of any real achievements sustained on the back of Machiavellianism, but one can list essentially endless entities whose downfall was brought on precisely by such.
Machiavellianism is not for everyone. It is specifically a framework for people in power. Kings, Heads of States, CEOs, Commanders. Competitive environments with allot at stake (peoples lives, money, future), in these environments it is often difficult to make decisions. Having a framework in place that allows you to make decisions is very useful.
Mitch Prinstein wrote a book about power and it shows that dark traits aren't the standard in most leaders, nor they are the best way to get into/stay in power
author is "board certified in clinical child and adolescent psychology, and serves as the John Van Seters Distinguished Professor of Psychology and Neuroscience, and the Director of Clinical Psychology at the University of North Carolina at Chapel Hill" and the book is based on evidence
edit: you can't take a book from 1600 and a few alive assholes with power and conclude that. there's a bunch of philanthropists and other people around
Im not saying that the end outcome wont be beneficial. I dont have a crystal ball. Im just saying that what he is doing is in no way selfless or laudable or worthy of praise.
Same goes for when Microsoft went gaga for open source and demanded brownie points for pretending to turn over a new leaf.
> Dont make the mistake of anthropomorphizing Mark Zuckerberg
Considering the rest of your comment it's not clear to me if "anthropomorphizing" really captures the meaning you intended, but regardless, I love this
Oh, absolutely -- I definitely meant that in the least complimentary way possible :). In a way, it's just the triumph of the ideals of "open source," -- sharing is better for everyone, even Zuck, selfishly.
> The price tag on this stuff, in human capital, data, and hardware, is high enough to preclude that sort of “perfect competition” environment.
I feel like we right now live in that perfect competition environment though. Inference is mostly commoditized, and it’s a race to the bottom for price and latency. I don’t think any of the big providers are making super-normal profit, and are probably discounting inference for access to data/users.
Only because everyone believes it’s a winner takes all game and this perfect competition will only last for as long as the winner hasn’t come out on top yet.
Everyone always thinks this at least in big tech I’ve never heard a PM or exec say a market is not winner take all. It’s some weird corpo grift lang that nothing is worth doing unless its winner take all.
A continuous stream of monetizable live user data?
The entire point of Meta owning everything is that it wants as much of your data stream as it can get, so it can then sell more ad products derived from that.
If much of that data begins going off-Meta, because someone else has better LLMs and builds them into products, that's a huge loss to Meta.
Meta's play is to make sure there isn't an obvious superiority to one company's closed LLM -- because that's what would drive customers to choosing that company's product(s).
If LLM effectiveness is all about the same, then other factors dominate customer choice.
Like which (legacy) platforms have the strongest network effects. (Which Meta would be thrilled about)
I think its about sapping as much user data from competitors. A company seeking to use an LLM has a choice between OpenAI, LLaMA, and others. If they choose LLaMA because it's free and host it themselves, OpenAI misses out on training data and other data like that
Well is the loss of training data from customers using self-hosted Llama that big a deal for OpenAI or any of the big labs at this point? Maybe in late-2022/early-2023 during the early stages of RLHF'd mass models but not today I don't think. Offerings from the big labs have pretty much settled into specific niches and people have started using them in certain ways across the board. The early land grab is over and consolidation has started.
Meta's primary business is capturing attention and selling some of that attention to advertisers. They do this by distributing content to users in a way that maximizes attention. Content is a complement to their content distribution system.
LLMs, along with image and video generation models, are generators of very dynamic, engaging and personalised content.
If Open AI or anyone else wins a monopoly there it could be terrible for Meta's business. Commoditizing it with Llama, and at the same time building internal capability and a community for their LLMs, was solid strategy from Meta.
So, imagine a world where everyone but Meta has access to generative AI.
There's two products:
A) (Meta) Hey, here are all your family members and friends, you can keep up with them in our apps, message them, see what they're up to, etc...
B) (OpenAI and others) Hey, we generated some artificial friends for you, they will write messages to you everyday, almost like a real human! They also look like this (queue AI generated profile picture). We will post updates on the imaginary adventures we come up with, written by LLMs. We will simulate a whole existence around you, "age" like real humans, we might even get married between us and have imaginary babies. You could attend our virtual generated wedding online, using the latest technology, and you can send us gifts and money to celebrate these significant events.
We would have to know their intent to really know if they fit a general understanding "the good guys."
Its very possible that China is open sourcing LLMs because its currently in their best interest to do so, not because of some moral or principled stance.
But that's precisely why Meta are the "good guys". They specifically called China the good guys in the same way that Meta is the good guys, though in this case many of the Chinese models are extremely good.
Meta has open sourced all of their offerings purely to try to commoditize the industry to the greatest extent possible, hoping to avoid their competitors getting a leg up. There is zero altruism or good intentions.
If Meta had actually competitive AI offering, there is zero chance they would be releasing any of it.
The country ruled by "people's party" has almost no open source culture while capitalism is leading the entire free software movement. I'm not sure what that says about our society and politics but the absurdist in me is having a good laugh every time I think about this :D
There’s actually a lot of open source software made by Chinese people. The government just doesn’t fund it. Not directly anyway, but there’s a ton of Chinese companies that do.
>There’s actually a lot of open source software made by Chinese people
Yea exactly, there is also a lot of chinese people out there, statistically a large chunk are cool with it.
Same dynamic as the US can be really - other countries see the US government and think to themselves, "I don't like these US people, look at what their government did" meanwhile US people are like "what do you mean, I don't like what the government did either". That's what a lot of Chinese people are thinking (but now allowed to say, in China criticizing the government is against their community guidelines)
I've recently been exploring PKM/knowledge management programs, and the best open source one is a Chinese project - SiYuan.
I have a feeling that their collaborative hacker culture is more hardware oriented, which would be a natural extension from the tech zones where 500 companies are within a few miles of each other and engineers are rapidly popping in and out and prototyping parts sometimes within a day.
Anecdotally, I've dealt with Chinese collaborative community projects in the ThinkPad space, where they have come together to design custom motherboards to modernize old ThinkPads. Of course there was a lot of software work as well when it comes to BIOS code, Thunderbolt, etc. I remember thinking how watching that project develop was like peering into another world with a parallel hacker culture that just developed... differently.
Oh there's also a Chinese project that's going to modernize old Blackberries with 5G internals. Cool stuff!
China does have their own Github as gitee.com¹ which runs a fork of Gitea but it's basically dead because it's impossible to have anything like Github with the current censorship aparatus. Here's the excerpt from wiki:
> On 18 May 2022, Gitee announced all code will be manually reviewed before public availability.[4][5] Gitee did not specify a reason for the change, though there was widespread speculation it was ordered by the Chinese government amid increasing online censorship in China.[4][6]
I won't pretend to be deeply familiar with China, but I think of two reasons: China doesn't take IP law seriously, so they can just copy, pirate whatever anyway. And the West has more wealthy idealistic techies with the free time for free software.
Capitalist countries (actually there are no other kinds of economies, in reality) are leading the open source software movement because it is a way for corporations to get software development services and products for free rather than paying for. It's a way of lowering labour costs.
Highly paid software engineers working in a ZIRP economy with skyrocketing compensation packages were absolutely willing to play this game, because "open source" in that context often is/was a resume or portfolio building tool and companies were willing to pay some % of open source developers in order to lubricate the wheels of commerce.
That, I think, is going to change.
Free software, which I interpret as copyleft, is absolutely antithetical to them, and reviled precisely because it gets in the way of getting work for free/cheap and often gets in the way of making money.
Copyleft isn't antithetical, see how many people are paid to work on the Linux kernel. I believe some other ecosystem software is also copylefted, like systemd.
And is building on top of the unpaid labour of SW engineers really a major part of the open source ecosystem? I feel open source is more a way for companies to cooperate in building shared software with less duplication of costs.
I disagree, the corporate open source is just half of the story. Much of free software space is pushed by idealists who can afford to pursue the ideals due to freedoms and finances provided by capitalist systems.
I don't think the intent really matters once the thing is out in the open.
I want open source AI i can run myself without any creepy surveillance capitalist or state agency using it to slurp up my data.
Chinese companies are giving me that - I don't really care about what their grand plan is. Grand plans have a habit of not working out, but open source software is open source software nonetheless.
It's really hard to tell. If instructions like the current extreme trend of "What a great question!" and all the crap that forces one to put
* Do not use emotional reinforcement (e.g., "Excellent," "Perfect," "Unfortunately").
* Do not use metaphors or hyperbole (e.g., "smoking gun," "major turning point").
* Do not express confidence or certainty in potential solutions.
into the instructions, so that it doesn't treat you like a child, teenager or narcissistic individual who is craving for flattery, can really affect the mood and way of thinking of an individual, those Chinese models might as well have baked in something similar but targeted at reducing the productivity of certain individuals or weakening their beliefs in western culture.
I am not saying they are doing that, but they could be doing it sometime down the road without us noticing.
I mean in some sense the Chinese domestic policy (“as in Xi”) made the conditions possible for companies like DeepSeek to rise up, via a multi-decade emphasis on STEM education and providing the right entrepreneurial conditions.
But yeah by analogy with the US, it’s not as if the W. Bush administration can be credited with the creation of Google.
Do we know if Meta will stick to its strategy of making weights available (which isn't open source to be clear) now that they have a new "superintelligence" subdivision?
Why not? Current open models are more capable than the best models from 6 months back. You have a choice to use a model that is 6 months old - if you still choose to use the closed version that’s on you.
There are plenty of companies that don't immediately qualify as "the bad guys".
For instance, of all companies I've interviewed with or have friends working at that developed tech, some companies build and sell furnitures. Some are your electricity provider or transporter. Some are building inventory management systems for hospitals and drug stores. Some develop a content management system for medical dictionnary. The list is long.
The overwhelming majority of companies are pretty harmless and ethically mundane. They may still get involved in bad practice, but that's not inherent to their business. The hot tech companies may be paying more (blood money if you ask me), but you have other options.
In my head at least, Bluesky are way closer to "the bad guys'. I don't trust them at all, pretty sure that in spite of what they say, they're going to do the same sort of rug pull that Google did with their "do no evil" assurances.
Funnily enough, I would actually flip it to say this about Kagi. With Bluesky, everything they have built is available to continue to be useful for people completely independent of what the folks over at Bluesky decide to do. There is no vendor lock in at all.
Kagi, on the other hand, has released none of their technology publicly, meaning they have full power to boil the frog, with no actual assurance that their technology will be useful regardless of their future actions.
Google was bad the moment it chose its business model. See The Age of Surveillance Capitalism for details. Admittedly there was a nice period after it chose its model when it seemed good because it was building useful tools and hadn't yet accrued sufficient power / market share for its badness to manifest overtly as harm in the world.
I would normally agree, but we're instantially talking about the company that made Pytorch and played an instrumental role in proliferating usable offline LLMs.
If you can make that algebra add up to "bad guy" then be my guest.
I wouldn't call mass piracy [0], for their own competitive gain, to be a "good" act. Especially when it seems they know they were doing the wrong thing - and that they know that the copyright complaints have grounds.
> The problem is that people don’t realize that if we license one single book, we won’t be able to lean into fair use strategy.
They're involved in genocide and enables near-global tyranny through their surveillance and manipulation. There are no excuses for working for or otherwise enabling them.
Most of Meta's models have not been released as open source. Llama was a fluke, and it helps to commoditize your compliment when you're not the market leader.
There is no good or open AI company of scale yet, and there may never be.
A few that contribute to the commons are Deep Seek and Black Forest Labs. But they don't have the same breadth and budget as the hyperscalers.
Llama is not open source. It is at best weights available. The license explicitly limits what kind of things you are allowed to use the outputs of the models for.
Yup, but that being said, Llama is GPLv3 weather Meta likes it or not. Same as ChatGPT and all the others. ALL of them can perfectly reproduce GPLv3 licensed works and data, making them derivative work, and the license is quite clear on that matter. In fact up until recently you could get chatGPT to info dump all sorts of things with that argument, but now when you try you will hit a network error, and afterwards it seems something breaks and it goes back to parroting a script on how it's under a proprietary license.
related stuff has, the core part being that if your model reproduces parts or all of a licensed work, it needs to comply with the license / copyright. Otherwise why aren't pirates just making 'models' that generate protected material, or music, and completely bypass all laws?
I know because I wanted to, as a form of protest/performance art, train a model to a few Disney movies and publicly distribute, but legal advice was this would put me directly into hot water not just because of who im pissing off (which i knew and was comfortable with) but also the fact there was precedent (i.e. news papers suing LLM providers).
It would be an open and shut case that would leave me in financial ruin.
The reason openAI hasn't been struck with this yet is, who has the time? and there isn't much to learn from all that either. Most open source tooling out competes openAI's offering as is, so the community wouldn't really win beyond punishing someone.
Yes. Having training obey copyright is a big coordination problem that requires copyright holders to group together to sue meta (and prove they broke copyright, which is not something proven before for LLM).
Whereas meta suing you into radioactive rubble is straightforward.
Just because someone is doing something you like doesn't automatically make them the good guys... In other, more clear words: Just because someone is offering you candy doesn't mean you should sit in their car with them...
