Re: LLM usage in Debian -- ballot options comparison

--0000000000006410030658396536
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Tue, 4 Aug 2026 at 15:07, Didier 'OdyX' Raboud <[email protected]> wrote:

> One way to look at this is that you used a 2026 proprietary Software-as-a=
-
> Service to produce a piece of software for yourself (and relatives).


We all use proprietary hardware, proprietary firmware. A lot of free
software users don't even run free software operating systems - they run
some free software on top of Windows or MacOS or Android or (in the past)
proprietary UNIX. That does not make the free software itself less free.


> One important point for me here is that I grew convinced that the Free
> Software movement is much more about the _movement_ that about the
> software.
>

Any community is good by itself. But the actual goal of the movement always
was to give users more freedom. A community is a side-effect. A pleasant
one, but still just a side-effect. If LLM coding gives more users more
freedom, then it would be fully consistent with the goals of the free
software community to embrace (and extend) the LLMs.


> To get back to your anecdote above: you built useful software, but you
> didn't
> build a community. Again, no offense meant: not all software needs a
> community, pushing to Github doesn't imply collaboration, etc. But I find
> that
> it's a too common trait of LLM-produced code: it's only good code.
>

Not all code needs to be shared or benefits from being shared or provides
benefits to others when it is shared. There is a *lot* of code that is only
ever useful for one user, one company, one computer. And that is great. The
more other software structures enable this kind of hyper-local free
software, the more freedom users have. Collaboration, maintainership,
policies, ... are simply not relevant to local code.


> A counterpoint to the above "it's just a massive proprietary SaaS offerin=
g
> currently cheap-enough" is that we have (or will have)
> open-{source,weight}
> LLM models that we can "just run ourselves". Right.
>
> An article from Micah Lee posted yesterday [0] about agentic coding had
> this
> excerpt:
> > I have a Framework Desktop server with 128GB of RAM.
> =E2=80=A6 and that's just the most recent article I stumbled upon (so I'm=
 not
> picking
> specifically at Micah). That setup is at least 3'600 USD as of today.


You should see what computer Linus uses to work on the Linux kernel. His
main workstation plus laptop cost at least as much. And yet, we all know,
that it can also be done in a much cheaper way. It just all depends on what
you want to do and and how fast you want to do that.

On the one extreme there is
https://www.cnx-software.com/2026/08/03/28-9m-parameter-llm-runs-locally-on=
-esp32-s3-at-9-tokens-s/
But in a more practical example, there are AMD XT 7900 cards with 20Gb RAM
for around 700=E2=82=AC that can run decently sized models with reasonable =
speed.
For similar price one can also find used Apple Mac mini with an M-series
processor and 32Gb of unified RAM. That can run even larger models but a
bit slower.
And you can rent the hardware, like from
https://www.hetzner.com/dedicated-rootserver/matrix-gpu/ for 0.375=E2=82=AC=
/hour

--=20
Best regards,
    Aigars Mahinovs

--0000000000006410030658396536
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 Tue, 4 Aug 2026 at 15:07, Didier &#39;OdyX=
&#39; Raboud &lt;<a href=3D"mailto:[email protected]">[email protected]</a>&gt;=
 wrote:<br></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">One way =
to look at this is that you used a 2026 proprietary Software-as-a-<br>
Service to produce a piece of software for yourself (and relatives).</block=
quote><div><br></div><div>We all use proprietary hardware, proprietary firm=
ware. A lot of free software users don&#39;t even run free software operati=
ng systems - they run some free software on top of Windows or MacOS or Andr=
oid or (in the past) proprietary UNIX. That does not make the free software=
 itself less free.</div><div>=C2=A0</div><blockquote class=3D"gmail_quote" =
style=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);pa=
dding-left:1ex">One important point for me here is that I grew convinced th=
at the Free <br>
Software movement is much more about the _movement_ that about the software=
.<br></blockquote><div><br></div><div>Any community is good by itself. But =
the actual goal of the movement always was to give users more freedom. A co=
mmunity is a side-effect. A pleasant one, but still just a side-effect. If =
LLM coding gives more users more freedom, then it would be fully consistent=
 with the goals of the free software community to embrace (and extend) the =
LLMs.</div><div>=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"marg=
in:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1e=
x">To get back to your anecdote above: you built useful software, but you d=
idn&#39;t <br>
build a community. Again, no offense meant: not all software needs a <br>
community, pushing to Github doesn&#39;t imply collaboration, etc. But I fi=
nd that <br>
it&#39;s a too common trait of LLM-produced code: it&#39;s only good code.<=
br></blockquote><div><br></div><div>Not all code needs to be shared or bene=
fits from being shared or provides benefits to others when it is shared. Th=
ere is a *lot* of code that is only ever useful for one user, one company, =
one computer. And that is great. The more other software structures enable =
this kind of hyper-local free software, the more freedom users have. Collab=
oration, maintainership, policies, ... are simply not relevant to local cod=
e.</div><div>=C2=A0</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">=
A counterpoint to the above &quot;it&#39;s just a massive proprietary SaaS =
offering <br>
currently cheap-enough&quot; is that we have (or will have) open-{source,we=
ight} <br>
LLM models that we can &quot;just run ourselves&quot;. Right.<br>
<br>
An article from Micah Lee posted yesterday [0] about agentic coding had thi=
s <br>
excerpt:<br>
&gt; I have a Framework Desktop server with 128GB of RAM.<br>
=E2=80=A6 and that&#39;s just the most recent article I stumbled upon (so I=
&#39;m not picking <br>
specifically at Micah). That setup is at least 3&#39;600 USD as of today.</=
blockquote><div><br></div><div>You should see what computer Linus uses to w=
ork on the Linux kernel. His main workstation plus laptop cost at least as =
much. And yet, we all know, that it can also be done in a much cheaper way.=
 It just all depends on what you want to do and and how fast you want to do=
 that.</div><div><br></div><div>On the one extreme there is=C2=A0<a href=3D=
"https://www.cnx-software.com/2026/08/03/28-9m-parameter-llm-runs-locally-o=
n-esp32-s3-at-9-tokens-s/">https://www.cnx-software.com/2026/08/03/28-9m-pa=
rameter-llm-runs-locally-on-esp32-s3-at-9-tokens-s/</a></div><div>But in a =
more practical example, there are AMD XT 7900 cards with 20Gb RAM for aroun=
d 700=E2=82=AC that can run decently sized models with reasonable speed. Fo=
r similar price one can also find used Apple Mac mini with an M-series proc=
essor and 32Gb of unified RAM. That can run even larger models but a bit sl=
ower.</div><div>And you can rent the hardware, like from=C2=A0<a href=3D"ht=
tps://www.hetzner.com/dedicated-rootserver/matrix-gpu/">https://www.hetzner=
.com/dedicated-rootserver/matrix-gpu/</a> for 0.375=E2=82=AC/hour=C2=A0</di=
v></div><br><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>

--0000000000006410030658396536--