[License-review] Please review revised ModelGo licenses

Fri May 16 11:15:26 UTC 2025

Moming, 

I undestand your point, but this is still an obligation that is inherent to the output of the model, nor on the model itself, neither on the derivatives. I understand that there may be some "knowledge" that flows from the model to the output, but this is also true for the training data and it might be true for all kind of software. 

Besides, what is the right that gives you control over the output? It commonly understood that the copyleft conditions, if at all imposed, can only reach as far as the copyright control reaches to derivatives. Any obligations or conditions that go outside the distribution artifact and derivatives thereof, insofar as they necessarily require a copyright permission, is OK, anything that goes beyond that, including inconsequential and unenforceable ones like "do no harm", are overreaching and go beyond Open Source. 

If you say that models (and AI systems altogether) are not (only) software and necessary adaptations are required, you are actually preaching the choir. But if you say that this makes a "collection of Output as a dataset generated wholly or partly using the Licensed Material" a derivative and therefore you somehow control it, I think this is not correct on many grounds, except perhaps in limited cases. 

Therefore, I hold the personal opinion that this single provision would make the containing licenses non Open Source. The other provisions, conversely, do not look overtly problematic, though I defer to what my learned colleagues have already written. 

Cheers 

Carlo 

> Da: "Moming Duan" <duanmoming at gmail.com>
> A: "License submissions for OSI review" <license-review at lists.opensource.org>
> Inviato: Mercoledì, 14 maggio 2025 17:25:23
> Oggetto: Re: [License-review] Please review revised ModelGo licenses

> Hi Carlo,

> Thanks for your reply.

> Regarding your concerns about model output, I refer to the email titled "Re:
> [License-review] Notice requirement for model output: OSD-compliant or not?
> (ModelGo)” dated March 4, in which you wrote:

>> I tend to agree. The output of a program (and of an AI system/model alike)
>> should not be subject matter controlled by the maker of them. This would
>> control conditions of use, not of downstream licensing (Freedom #0 in FSF's
>> Four freedoms parlance).
>> I gather Simon has the same view.
>> If the output is "derivative" of anything, I believe it is of the training
>> material, but then again this is more an unwanted accident than the rule,
>> according to my to-date understanding of the interaction between those two
>> bits.
>> It may sound akin to the requirement of the AI Act to watermark or otherwise
>> make conspicuous that a content has been generated by AI, but I don't see a
>> rationale in allowing this. We don't conflate policy requirements (as the one
>> Dirk mentions, incidentally) and licensing conditions, the two must stay
>> separate and depend on a lot of things that don't belong in, and cannot be
>> figured out at, licensing.
>> This has nothing to do with attribution, it is more akin, if anything to the
>> advertisement clause, you are right, but misplaced, since it refers to the
>> output and the output should be controlled by the user. The fact that is
>> becoming commonplace is no excuse. Proprietary license was (is) commonplace
>> when OSD was drafted too.

> From my point of view:

>>> On 4 Mar 2025, at 5:40 AM, Richard Fontana < [ mailto:rfontana at redhat.com |
>>> rfontana at redhat.com ] > wrote:
>>> This would be akin to putting an editor under a license that required
>>> any file created with the editor to have an attribution notice. Or, if
>>> there is some distinction, I'm not immediately seeing it.
>> An AI model is not like an editor. As I discussed previously, the intention of
>> this clause is not to enforce legal compliance but to ensure that open source
>> remains substantial. It is a widely practiced approach in the ML community to
>> extract knowledge from one model to improve another. In some cases, like
>> DeepSeek, the new model can even outperform the original one.
>> Using your editor analogy, an AI model is more like an editor that is likely to
>> output part of its source code. My intention is to require downstream users to
>> provide attribution when they create a dataset using this content, which may
>> then be used by others for training their models (e.g., Llama, ChatGPT).

> In summary, I believe that when knowledge is transferred from an original model
> to a new one, the resulting model is likely to inherit similar characteristics
> (i.e., reasoning, math), or even outperform the original. The knowledge
> embedded in the outputs (e.g., chains of thought and reasoning capabilities)
> should therefore be regarded as part of the original model’s “source code.” For
> this reason, both MG-BY-2.0 and MG-BY-OS-2.0 state:

>> 2.2(b) If You Distribute a collection of Output as a dataset generated wholly or
>> partially using
>> the Licensed Materials and/or Derivative Materials, You shall provide a
>> prominent notice
>> in the Distribution stating that the dataset is generated using the Licensed
>> Materials
>> and/or Derivative Materials provided under this License.

> This clause does not restrict the use of model outputs; rather, it requires
> users to provide notice (not license) when using those outputs to create and
> distribute datasets. Notably, MG-0 does not include such a requirement. I
> understand that further discussion is needed on whether attribution should be
> required for model-generated datasets. My intention here is to emphasize that
> models are not exactly the same as software, and should not always be treated
> as such. Thanks.

> Best,
> Moming

>> On 14 May 2025, at 10:09 PM, Carlo Piana <carlo at piana.eu> wrote:

>> Josh,

>> sorry for long silence.

>> I think that the new version of the ModelGo license does not seem to addres=
>> s the concern I have expressed against it, following up on your own comment=
>> on output (now in 2.bb). I think that imposing anything on the output of t=
>> he model is against the OSD as it is a restriction on the use of the licens=
>> ed subject matter.

>> So no, I am confused at how this new text should be addressing the above co=
>> ncern.

>> In a separate thread I have expressed perplexity on certain clauses, these =
>> seem to have been removed, so no issue on that end.

>> This applies to the updated versions.

>> Cheers

>> Carlo

>> ----- Messaggio originale -----

>>> Da: "Josh Berkus" <josh.berkus at opensource.org>
>>> A: "License submissions for OSI review" <license-review at lists.opensource.=

>> org>

>>> Inviato: Marted=C3=AC, 15 aprile 2025 1:46:45
>>> Oggetto: [License-review] Please review revised ModelGo licenses

>>> Carlo, Pam, Eric, Shuji,
>>> =20
>>> Moming has re-submitted revised versions of his licenses based on your
>>> feedback. Please check them when you can and make sure that your
>>> concerns about the licenses have been addressed.
>>> =20
>>> --
>>> -- Josh Berkus
>>> OSI Board Member
>>> =20
>>> =20
>>> _______________________________________________
>>> The opinions expressed in this email are those of the sender and not nece=

>> ssarily

>>> those of the Open Source Initiative. Communication from the Open Source
>>> Initiative will be sent from an opensource.org email address.
>>> =20
>>> License-review mailing list
>>> License-review at lists.opensource.org
>>> http://lists.opensource.org/mailman/listinfo/license-review_lists.opensou=

>> rce.org

>> _______________________________________________
>> The opinions expressed in this email are those of the sender and not necessarily
>> those of the Open Source Initiative. Communication from the Open Source
>> Initiative will be sent from an opensource.org email address.

>> License-review mailing list
>> License-review at lists.opensource.org
>> http://lists.opensource.org/mailman/listinfo/license-review_lists.opensource.org

> _______________________________________________
> The opinions expressed in this email are those of the sender and not necessarily
> those of the Open Source Initiative. Communication from the Open Source
> Initiative will be sent from an opensource.org email address.

> License-review mailing list
> License-review at lists.opensource.org
> http://lists.opensource.org/mailman/listinfo/license-review_lists.opensource.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20250516/0171448e/attachment-0001.htm>