🚩 Report — Ethical Issue: Questions about Discriminative Licensing & Unfair Business Practice
The MPT-7b Chat model was licensed as non-commercial from the start due to relying on a CC-NC dataset. The MPT-7b Storywriter model was relicensed as non-commercial due to relying on a copyrighted books dataset.
However this model also contains a variety of works that are Copyrighted and under license, yet it's re-licensed for commercial use.
@MosaicML Could you clarify how the licensing decision is made to make a model inherit the dataset license, and when it overrides/inhibits the dataset licenses? If there's a technical standard you're applying for this (e.g. metric) then it'd be nice to explain that.
My concern, and why I raise an Ethical Issue, is that this looks like a form of discrimination, respecting the human rights (economic & moral rights) of certain authors and writers, but not others. (I believe this would be against the Terms of Service here.) Furthermore, it looks like an unfair business practice under applicable laws to selectively respect the IP of certain companies and organizations.
I've been very impressed with how MosaicML has handled this so far, where as other companies releasing lower quality LLMs simply wait to be sued. It's great that the company is setting such a high standard!
We recommend people consult a lawyer before deciding how they want to use a model. Licensing decisions do not constitute legal advice, and we are not responsible for the actions of third parties who use the model.
Licensing decisions are not legal advice, but MosaicML is responsible as a company for the consequences of its own licensing decisions. Especially if you don't have the rights to license the model, or you license things selectively. The topic here is not about downstream users but about MosaicML taking responsible action, not misleading users with licenses it doesn't have the right to give.
I'm glad you reverted the Storywriter licence back because it'll be more interesting to watch this play out over the next few days ;-). It still, however appears to be discriminatory licensing and unfair business practice.
[...] we are not responsible for the actions of third parties who use the model.
FYI, this likely won't fly in court if you end up there. The responsibility is on you to not misrepresent what you're licensing, otherwise it's a form of fraud relating to granting rights you don't have. Copyright fraud. Especially as you explicitly wrote "commercial use permitted" in brackets, you and your company has providing legal advice and interpretation of the license.
Additionally, MosaicML chose one license and then changed it back, licensed some models selectively, etc. It doesn't look like MosaicML really considered the implications of their decision at all before starting or when applying the changes. This looks like negligence in licensing and simply reinforces the claims against the company.
You should also consult lawyer before you release models, or if you did, consult a better one!
Better call Saul
Anyway, I don't understand the point of this discussion since OpenAI is doing the same thing, just on another scale. When it comes to copyright infringement in LLMs, there aren't (yet) any clear laws that would prohibit this. With that being said, we should be thankful that we've got hands on something which cost Mosaic 200k+ to develop.
@cekal The license wouldn't affect you getting your hands on the models, just the terms.
When it comes to copyright infringement in LLMs, there aren't (yet) any clear laws that would prohibit this.
The EU debated this for its AI Act and decided that Copyright laws were sufficient, and indeed the growing consensus is that they are. Besides, this post is about the discriminatory licensing, respecting the rights of some organizations and not others — so this would fall outside of Copyright anyway.
Closing as report seems to have been addressed by Mosaic team
@alexjc if you don't like the terms or license, move on to a different model.
you're coming off as an attention-seeking troll from the way you're behaving in this thread.
@nacs You missed the point, but I guess you didn't read the thread? The license claimed is not valid for the purpose they repeatedly stated. I'm not seeking attention as intentionally ignoring legal and ethical issues is the norm (in a bad way) for both the model owners and the platform.
@julien-c It appears you re-closed this thread intentionally. This action hides the concerns (which don't show up by default in the Community Tab) and accepts the fact that no response was provided by the model owner. You did not provide a position from HuggingFace, so I presume your conclusion is that everything is acceptable based on your investigation of the matter. Please correct if this is not the case, you will receive a notification by being tagged.