Re: DIscussion on FMLL v1.0: A License for Free Artificial Intelligence

--===============1864100919204436732==
Content-Type: multipart/alternative; boundary="000000000000c5f8fc0656cecf61"
Content-Transfer-Encoding: 7bit

--000000000000c5f8fc0656cecf61
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Dear David Woolley,

Thank you for your feedback.

I believe that a license should protect the rights holders of datasets, and
withholding datasets is unacceptable (for example, they might contain
private data). Also, users have the right to know what data a model was
"trained" on. Since datasets and weights are not code, existing licenses
cannot affect this in any way.

As for reproducibility, I did not intend to make "AI Body" fully
reproducible in practice, which is why I used the term "Theoretical
Reproducibility" =E2=80=93 this term is primarily meant to encourage datase=
t
openness. Perhaps I should refine the definition of "Theoretical
Reproducibility."

Sincerely,
Mikhail Potekha

=D0=BF=D1=82, 17 =D0=B8=D1=8E=D0=BB. 2026=E2=80=AF=D0=B3., 06:29 David Wool=
ley <[email protected]>:

> In general, I don't think it is close to being well drafted, but I'll
> pick some points specific to the application.
>
> On Thu, 2026-07-16 at 16:28 +1000, Mihail Poteha wrote:
> > I am an independent developer from Russia, and I am creating my own
> > AI license. To date, I have not found a license for Open AI. All
> > existing licenses, including OSAID, allow datasets to be withheld,
> > either partially or in full, thereby creating a =E2=80=9Cblack box.=E2=
=80=9D Such
>
> For most applications of AI training on datasets that you can freely
> redistribute will give them inadequate diversity to be useful in the
> real world.
>
> > licenses also fail to protect the rights holders of the datasets. I
> > disagree with this approach.
>
> Nothing in the main text stood out as achieving that, and it seemed to
> me that the licence only really works if none of authors of the
> training material intend to exercise any of their rights.
> >
> > Theoretical Reproducibility. A property of the model=E2=80=99s source f=
ile
> > that, in theory, makes it possible to recreate a complete copy of the
> > language model=E2=80=99s source file on the basis of the instructions a=
nd
> > tools specified in the AI Body. The Author must provide complete
>
> AI models are trained using random numbers, and training tends to be
> massively parallel.  Even with explicit seeds, programming languages do
> not specify the algorithm used to create the random sequences, so they
> could differ over time and between machines.  The parallelism also
> introduces the possibility of different scheduling orders.
>
> At inference time, simpler user interfaces won't include the ability to
> set seed values, and many users will not want exact repeatability.  For
> training that converges and inference that converges, you can only
> define statistical properties of the result, not exact values.
>
> I'm not sure if the OSD forces you to publish private modifications,
> but it is generally not part of the open source culture, and it is
> unenforceable, because it is happening in private.
>
> _______________________________________________
> The opinions expressed in this email are those of the sender and not
> necessarily those of the Open Source Initiative. Official statements by t=
he
> Open Source Initiative will be sent from an opensource.org email address.
>
> License-discuss mailing list
> [email protected]
>
> http://lists.opensource.org/mailman/listinfo/license-discuss_lists.openso=
urce.org
>

--000000000000c5f8fc0656cecf61
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"auto">Dear David Woolley,<div dir=3D"auto"><br></div><div dir=
=3D"auto">Thank you for your feedback.</div><div dir=3D"auto"><br></div><di=
v dir=3D"auto">I believe that a license should protect the rights holders o=
f datasets, and withholding datasets is unacceptable (for example, they mig=
ht contain private data). Also, users have the right to know what data a mo=
del was &quot;trained&quot; on. Since datasets and weights are not code, ex=
isting licenses cannot affect this in any way.</div><div dir=3D"auto"><br><=
/div><div dir=3D"auto">As for reproducibility, I did not intend to make &qu=
ot;AI Body&quot; fully reproducible in practice, which is why I used the te=
rm &quot;Theoretical Reproducibility&quot; =E2=80=93 this term is primarily=
 meant to encourage dataset openness. Perhaps I should refine the definitio=
n of &quot;Theoretical Reproducibility.&quot;</div><div dir=3D"auto"><br></=
div><div dir=3D"auto">Sincerely,</div><div dir=3D"auto">Mikhail Potekha</di=
v></div><br><div class=3D"gmail_quote gmail_quote_container"><div dir=3D"lt=
r" class=3D"gmail_attr">=D0=BF=D1=82, 17 =D0=B8=D1=8E=D0=BB. 2026=E2=80=AF=
=D0=B3., 06:29 David Woolley &lt;<a href=3D"mailto:[email protected].=
uk">[email protected]</a>&gt;:<br></div><blockquote class=3D"gmail=
_quote" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:=
1ex">In general, I don&#39;t think it is close to being well drafted, but I=
&#39;ll<br>
pick some points specific to the application.<br>
<br>
On Thu, 2026-07-16 at 16:28 +1000, Mihail Poteha wrote:<br>
&gt; I am an independent developer from Russia, and I am creating my own<br=
>
&gt; AI license. To date, I have not found a license for Open AI. All<br>
&gt; existing licenses, including OSAID, allow datasets to be withheld,<br>
&gt; either partially or in full, thereby creating a =E2=80=9Cblack box.=E2=
=80=9D Such<br>
<br>
For most applications of AI training on datasets that you can freely<br>
redistribute will give them inadequate diversity to be useful in the<br>
real world.<br>
<br>
&gt; licenses also fail to protect the rights holders of the datasets. I<br=
>
&gt; disagree with this approach.<br>
<br>
Nothing in the main text stood out as achieving that, and it seemed to<br>
me that the licence only really works if none of authors of the<br>
training material intend to exercise any of their rights.<br>
&gt; <br>
&gt; Theoretical Reproducibility. A property of the model=E2=80=99s source =
file<br>
&gt; that, in theory, makes it possible to recreate a complete copy of the<=
br>
&gt; language model=E2=80=99s source file on the basis of the instructions =
and<br>
&gt; tools specified in the AI Body. The Author must provide complete <br>
<br>
AI models are trained using random numbers, and training tends to be<br>
massively parallel.=C2=A0 Even with explicit seeds, programming languages d=
o<br>
not specify the algorithm used to create the random sequences, so they<br>
could differ over time and between machines.=C2=A0 The parallelism also<br>
introduces the possibility of different scheduling orders.<br>
<br>
At inference time, simpler user interfaces won&#39;t include the ability to=
<br>
set seed values, and many users will not want exact repeatability.=C2=A0 Fo=
r<br>
training that converges and inference that converges, you can only<br>
define statistical properties of the result, not exact values.<br>
<br>
I&#39;m not sure if the OSD forces you to publish private modifications,<br=
>
but it is generally not part of the open source culture, and it is<br>
unenforceable, because it is happening in private.<br>
<br>
_______________________________________________<br>
The opinions expressed in this email are those of the sender and not necess=
arily those of the Open Source Initiative. Official statements by the Open =
Source Initiative will be sent from an <a href=3D"http://opensource.org" re=
l=3D"noreferrer noreferrer" target=3D"_blank">opensource.org</a> email addr=
ess.<br>
<br>
License-discuss mailing list<br>
<a href=3D"mailto:[email protected]" target=3D"_blank" r=
el=3D"noreferrer">[email protected]</a><br>
<a href=3D"http://lists.opensource.org/mailman/listinfo/license-discuss_lis=
ts.opensource.org" rel=3D"noreferrer noreferrer" target=3D"_blank">http://l=
ists.opensource.org/mailman/listinfo/license-discuss_lists.opensource.org</=
a><br>
</blockquote></div>

--000000000000c5f8fc0656cecf61--


--===============1864100919204436732==
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: base64
Content-Disposition: inline

X19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX18KVGhlIG9waW5p
b25zIGV4cHJlc3NlZCBpbiB0aGlzIGVtYWlsIGFyZSB0aG9zZSBvZiB0aGUgc2VuZGVyIGFuZCBu
b3QgbmVjZXNzYXJpbHkgdGhvc2Ugb2YgdGhlIE9wZW4gU291cmNlIEluaXRpYXRpdmUuIE9mZmlj
aWFsIHN0YXRlbWVudHMgYnkgdGhlIE9wZW4gU291cmNlIEluaXRpYXRpdmUgd2lsbCBiZSBzZW50
IGZyb20gYW4gb3BlbnNvdXJjZS5vcmcgZW1haWwgYWRkcmVzcy4KCkxpY2Vuc2UtZGlzY3VzcyBt
YWlsaW5nIGxpc3QKTGljZW5zZS1kaXNjdXNzQGxpc3RzLm9wZW5zb3VyY2Uub3JnCmh0dHA6Ly9s
aXN0cy5vcGVuc291cmNlLm9yZy9tYWlsbWFuL2xpc3RpbmZvL2xpY2Vuc2UtZGlzY3Vzc19saXN0
cy5vcGVuc291cmNlLm9yZwo=

--===============1864100919204436732==--