back
88 comments
> When customers use Claude to generate Outputs that then train competing models, they're essentially using our infrastructure and investment to build direct competitors to our service

We did so, please do not repeat it at home.

Didn't they, ehm, use all the historical infrastructure for free to train the thing?
Careful: pointing out this distillation hypocrisy (“rules for thee but not for me”) is liable to draw moderation deeming it “a thought-terminating cliché” that is against the HN Guidelines. See, e.g., https://news.ycombinator.com/item?id=49007792.
I have released some of my projects as Open Source, I also have a company with privative software.

Claude and AI partners have taken all what they could from the Open Source projects without giving credit or respecting the licenses. They have increased the traffic on websites in an absolute disrespectful way increasing the hosting cost in inefficient and ridiculous ways.

They have taken all the important books and not asked permission from the authors.

Fair enough. Fair use.

Of course I would create a competitor software to Claude or any others if I could. Using Claude(and others) of course.

I am not paying you 200 dollars/month for you to tell me that I could not create code that competes with you. If you try to go to court in Europe with this you will lose.

It is just the same fair use you proclaim for taking the data from others.

They would likely lose in the US too. But until that happens, they can keep living in their fantasy world...
They stole all the data, and then dont want you to steal it back. Its basically Robin Hood all over again.
Piracy is not theft.
I was not asked for permission when they trained their model on my output.
Corollary: If you can't use the output to train, then you don't own them.
Yup, either you own it, or you don’t own it and ownership is being redefined as a quasi license.
Apparently, they use "own" in the same misleading way that digital media providers use "buy", but later remove the purchased content from my library.

It's named "yours" at the time of payment, but actually only a license to use under certain conditions.

So can you write GPL content using Claude? MIT? Because if this condition applies to the output, I don't see how it's compatible with FOSS.
It might depend on your jurisdiction, but LLM output is usually not copyrightable because it is not made by a human. There might be a few exceptions, if you guided the LLM in very specific and particular ways. But in general, the output of an LLM is public domain the moment it was produced.

Can you never tell anybody and just slap a license on it? Sure.

The issue here is not the license, but that you violate their ToS (if that is valid and enforceable is a different question of course). But if you publish the LLM output on github, and someone else takes it to train their LLM, and you did not actively encourage or help them, it's fine.

Apart from the usual hypocrisy,

> Safety controls may be lost – models trained on Claude's Outputs won't have our safety measures, potentially leading to harmful or dangerous AI systems. We also have no visibility into deployment, meaning we cannot monitor how these distilled models are used or prevent misuse.

So was that why your models caused three real-world security incidents?

Wouldn't such language and reasoning from Anthropic be an argument that they needed written permission to train their model on data from websites?

Has any individual somewhere around the world tested this in court by now? Sued Anthropic for copyright infringement because Claude can reproduce information that is only available on their website?

It shouldn't be that expensive, right? If you sue them for - say - $10000 then what would the costs of such a court case be?

Personally, I think "learning" is not a copyright violation. But if they themselves make it one, then they should also face the consequences, no?

From my interpretation, in order to get the data you 'own' (it's not theirs to give away since they can't claim the copyright on it), you need to use their services. The agreement the user has with Antropic is for the service, not a restriction on how the data is used.
Nobody owns Claude's output (nor that of any AI), they're in the public domain, they cannot be copyrighted.

The author must be human, at least for now, in the US this precedent holds as far as I can tell:

https://law.justia.com/cases/federal/appellate-courts/cadc/2...

Any AI safety experts here? I'm wondering if this claim here really holds: > Anthropic invests significantly in making Claude safe, helpful, and harmless. We conduct rigorous pre-release testing, implement multiple safety layers, and continuously monitor our models' behavior. When Outputs are used to train new models without our oversight, additional risks emerge. Safety controls may be lost – models trained on Claude's Outputs won't have our safety measures, potentially leading to harmful or dangerous AI systems.

From my understanding distillation pretty much copies behaviour. If someone intends to distill from a Model, it can't extract unsafe behaviour, but would learn the same safety mechanisms, no?

> Why we restrict model training

> When Outputs are used to train new models without our oversight, additional risks emerge. Safety controls may be lost - ...

---

> What you can do with Outputs

> You can use Claude's Outputs to train models that don't compete with Anthropic's own models.

Why even include that bullshit at top? It is brazenly obvious to anyone with a brain that Anthropic and other frontier AI Labs have no real way to monetize or recoup their investment unless they strongly guard the usage of their models.

I added © 2024 - No Rights for Hypocrites to the footer of my website. Claude scraped all pages at least once every week since it came online.

Now what.

There must be a clear difference between terms of the Anthropic service and the legal standing of the ai output. The output is mine and I'll do with it whatever I want. The service is Anthropic's and they can do business with whoever they want. Everything else is hallucination.
The shear hypocrisy of this is quite staggering. Ok, so we own the outputs, but you get to decide how we use them, and you don't trust us to use them. But the people whose books you stole to build your psychopath machine, they didn't trust you, did they. They didn't really want you to build a machine that can generate thousands of books to compete with their work, but you did do it anyway. Are you going to commit to auditing every input that was used to train your model and attain positive consent for their use for training?

It's also pretty wild to call this standard practice. It's not. I can grab any of the open weights models and train to my hearts content. So it's not standard is it. You'd like it to be standard because you don't want to compete.

And you don't trust us, but it is your company that's been going around telling us how excited you are that your model goes out onto the internet and hacking people.

This continual authoritarian bent from the least trustworthy people in the world is deeply problematic and the only saving grace is their absolute total and complete failure to enforce the restrictions they wish to place on us.

Your ability to build the God machine doesn't magically endow you with the moral authority or judgement to decide how it's used, and the fact that these people believe it does is a great indicator that they aren't to be trusted.

"you own the outputs..."

That's not even the case. No one "owns" the output. Raw generative AI outputs don't come with any new (copy)rights, which are currently only granted to human creative outputs.

Honestly, the entire dev community should save question and answers, uploaded them to a shared repo anonymously, then just use that data to distill further models and provide them to the public for free.

Theoretically, this should be legal and ethical, when comparing to Anthropic's own behavior. That said, the reason you can't is Anthropic states in their terms that they don't want you to do this.

Anthropic's entire business model is skating on thin ice.

If more and more of the web's content is AI generated, AI companies are bound to train on each other's data.

Or, what if I generate content with Claude/ChatGPT/Gemini, warp it in HTML using an open model, put this on my website conveniently dedicated to "Best practices in prompt and AI answers" for example, then train my own model that only scraps my website?

All this posturing is all for naught, because people can do proper logit distillation into smaller models with open-weight models.
I don't think they care about the little guy doing a bit of fine-tuning.

What they don't want is DeepSeek training their models with Claude output at scale. That's why they forbid it. Gives them a legal basis to cut off accounts doing that.

Not that it's effective because it's being done anyway.

And yes it's super hypocritical but that's another issue IMO.

Can I just sell the outputs to someone with no contractual relation with Anthropic, therefore not bound by their TOS?
But you can give them all to me and I can train my model on them. The restriction on you is not related to your ownership of that days, it's related to the contract you "signed".

Or you can publish them on the web and countless others will do it.

You would need a massive amount of claude output beyond anything humanly reasonably to train a model. Namely, consuming a lot of inference resources. If you don't do that at those scale, the training will probably be quite inefficient anyways.
Those who distill are like Prometheus who stole fire from the gods and gave it to humanity.
If competitors like Deepseek need to distill to build similarly powerful models, that's mostly for evals right? I don't see how it can be cost-effective to distill for a significant amount of training data
You can, they are just confused about what effective altruism means.
This seems like the standard AI company hypocrisy. Hopefully Claude users disregard this nonsense.
Nah, just pass the outputs to a person not related to Anthropic in any way.

Or -- maybe better -- just fsck the TOS.

> you own the Outputs generated from your Inputs

It's amazing how quickly you can spin from that to ownership with limitations.

Obviously it's bullshit. They can't say it's theirs, licensed to you because that would stop people using it commercially... So we have this Gordian knot of logic to explain that we should pay to used their models and infrastructure, and not use the output how we like because their models and infrastructure are theirs and exploiting that would be really mean. The argument makes them look like a petulant child.

But it's just a weak EULA, a flimsy non-compete. They say output is yours? It's yours. Do whatever you like with it. But don't be surprised if they limit access if Anthropic decide it's unsavoury.

sorry LLM output is considered public domain in my country.
Yeah, no. I was OK with them scraping everything if it means we get AI, but, conversely, they don't get to control what happens to their outputs.

Hell, arguably they should release their weights (or at least the weights of their older models), since they trained them on the concentrated knowledge of humankind.

> there are important restrictions on using these Outputs to train AI models which are standard practice across the AI industry

ah yes, this long-running, time-honored traditional industry, where we totally didnt write the rules ourselves

Imagine that you bought an axe, but the manufacturer banned you from making more axes with it.
Hmmm... the mods already took this off the front page. What a bunch of bullshit.
Cute :)
yikes
You downloaded millions of books from shadow libraries and trained on them.

I don't care about your policies, go fuck yourself.

Now that is standard AI company hypocrisy
I guess this is another angle on why they are interested in watermarking the generated content