Re: GCC LLM policy adopted

--000000000000c6da400657e58fc8
Content-Type: text/plain; charset="UTF-8"

On Fri, 31 Jul 2026 at 12:00, Gard Spreemann <[email protected]> wrote:

> Aigars Mahinovs <[email protected]> writes:
>
> > You've missed significant exceptions that GCC policy grants
>
> I wasn't specifically referring to the GCC policy at all. I was agreeing
> with and picking up on Sean and Russ' about the importance of
> distinguishing between AI-generated code and AI assistance while writing
> code. I should have changed the email subject, sorry.
>

IMHO the GCC policy is a well considered stop-gap and if we were to craft a
ballot option in a similar vein, then refering to it and implementing close
to it would be useful for building community consensus. And it would
clearly signal the intent to review that policy when GCC does the same.


> > - grant on changes below "legally significant" limit (~15 lines of
> > code)
>
> Instinctually to me, there's so little to be gained from automating the
> writing of ~15 lines of code that I'm not very worried about this.
>

You know the old joke about the TV repairman with a hammer?

TV does not work, a repairman comes by, inspects the TV and hits it with a
hammer, the TV works perfectly again.
- Here is your repair bill, 92$
- Why so much, you just hit it with a hammer!
- The 2$ is for hitting it with the hammer, 90$ is for knowing where
exactly to hit

Same with AI-generated contributions, especially bugfix patches and
security fixes. The most useful of them tend to be way smaller than 15
lines of code, but can take massive effort to actually find the right place
and the right change.


> > and grant to on adding tests
>
> True.
>
> > as well as writing aids, including spell check, translation,
> > speech-to-text and reverse and so on. They also make a point that AI
> > patch review is fine.
>
> Spell-check and AI patch review are explicitly fine under my draft
> ballot suggestion. Is there anything in the language that makes it seem
> not fine?
>

Documentation (or even code, especially in comments) that has AI-driven
translation will be "AI-generated content contributed to Debian" fully,
same (to lesser extent) for spell check and grammar clarifications.

One big benefit of AI patch review is that it often can suggest fixes for
its comments. Including those fixes (even if that is only a 2-line diff to
the original submission) will again be "AI-generated content contributed to
Debian" and thus forbidden under your current formulation, but not
forbidden under GCC policy.

These kinds of questions are why the GCC policy was formulated as it was -
these exceptions allow many trully useful AI-assisted workflows without
legally significant copyright risks.
-- 
Best regards,
    Aigars Mahinovs

--000000000000c6da400657e58fc8
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div class=3D"gmail_quote gmail_quote_container"><div dir=
=3D"ltr" class=3D"gmail_attr">On Fri, 31 Jul 2026 at 12:00, Gard Spreemann =
&lt;<a href=3D"mailto:[email protected]">[email protected]</a>&gt; wrote:<b=
r></div><blockquote class=3D"gmail_quote" style=3D"margin:0px 0px 0px 0.8ex=
;border-left:1px solid rgb(204,204,204);padding-left:1ex">Aigars Mahinovs &=
lt;<a href=3D"mailto:[email protected]" target=3D"_blank">[email protected]=
om</a>&gt; writes:<br>
<br>
&gt; You&#39;ve missed significant exceptions that GCC policy grants<br>
<br>
I wasn&#39;t specifically referring to the GCC policy at all. I was agreein=
g<br>
with and picking up on Sean and Russ&#39; about the importance of<br>
distinguishing between AI-generated code and AI assistance while writing<br=
>
code. I should have changed the email subject, sorry.<br></blockquote><div>=
<br></div><div>IMHO the GCC policy is a well considered stop-gap and if we =
were to craft a ballot option in a similar vein, then refering to it and im=
plementing close to it would be useful for building community consensus. An=
d it would clearly signal the intent to review that policy when GCC does th=
e same.</div><div>=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"ma=
rgin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:=
1ex">
&gt; - grant on changes below &quot;legally significant&quot; limit (~15 li=
nes of<br>
&gt; code)<br>
<br>
Instinctually to me, there&#39;s so little to be gained from automating the=
<br>
writing of ~15 lines of code that I&#39;m not very worried about this.<br><=
/blockquote><div><br></div><div>You know the old joke about the TV repairma=
n with a hammer?</div><div><br></div><div>TV does not work, a repairman com=
es by, inspects the TV and hits it with a hammer, the TV works perfectly ag=
ain.</div><div>- Here is your repair bill, 92$</div><div>- Why so much, you=
 just hit it with a hammer!</div><div>- The 2$ is for hitting it with the h=
ammer, 90$ is for knowing where exactly to hit</div><div><br></div><div>Sam=
e with AI-generated contributions, especially bugfix patches and security f=
ixes. The most useful of them tend to be way smaller than 15 lines of code,=
 but can take massive effort to actually find the right place and the right=
 change.</div><div>=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"m=
argin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left=
:1ex">
&gt; and grant to on adding tests<br>
<br>
True.<br>
<br>
&gt; as well as writing aids, including spell check, translation,<br>
&gt; speech-to-text and reverse and so on. They also make a point that AI<b=
r>
&gt; patch review is fine.<br>
<br>
Spell-check and AI patch review are explicitly fine under my draft<br>
ballot suggestion. Is there anything in the language that makes it seem<br>
not fine?<br></blockquote><div><br></div><div>Documentation (or even code, =
especially in comments) that has AI-driven translation will be &quot;AI-gen=
erated content contributed to Debian&quot; fully, same (to lesser extent) f=
or spell check and grammar clarifications.=C2=A0</div><div><br></div><div>O=
ne big benefit of AI patch review is that it often can suggest fixes for it=
s comments. Including those fixes (even if that is only a 2-line diff to th=
e original submission) will again be &quot;AI-generated content contributed=
 to Debian&quot; and thus forbidden under your current formulation, but not=
 forbidden under GCC policy.</div><div><br></div></div>These kinds of quest=
ions are why the GCC policy was formulated as it was - these exceptions all=
ow many trully useful AI-assisted workflows without legally significant cop=
yright risks.<div><span class=3D"gmail_signature_prefix">-- </span><br><div=
 dir=3D"ltr" class=3D"gmail_signature"><div dir=3D"ltr">Best regards,<br>=
=C2=A0 =C2=A0 Aigars Mahinovs</div></div></div></div>

--000000000000c6da400657e58fc8--