Re: GR: Ban LLM contributions from Debian

De : Matthias Geiger <[email protected]>
À : [email protected]
Date : 22 juil. 2026 20:01:23
Objet : Re: GR: Ban LLM contributions from Debian

> On Wed, 22 Jul 2026 19:30, Pierre-Elliott Bécue <[email protected]> wrote:
> 
> Hi,
> 
> 
>> Hello,
>> 
> [...]
>> 
>> These apply exclusively to code and work done specifically for the
>> Debian project (Debian websites, applications, resources, packages,
>> etc.). They do not apply to any upstream work. In what follows, "work"
>> refers to work covered by these guidelines.
>> 
>> - All code and work assisted by a generative AI agent or tool must
>>   comply with the DFSG.
>> - The submitter is solely responsible for the submitted work and, by
>>   submitting it, certifies that:
>>   + any generated code has been reviewed and audited;
>>   + all licensing-compliance duties have been fulfilled;
>> 
> Sorry, but I fail to see how code generated by LLMs can ever comply with the DFSG. TTBOMK no model exists that a) was solely trained on DFSG-software, b) has DFSG-compatible licensing on itself and c) denotes from where in the training data the output comes from.
> 
> There is still a huge uncertainty to whether such output is copyrightable at all, and if so, under which license. If you let a function be generated where you can't be 100% if you don't have any license obligations due to the training data, this does not meet DFSG IMO.
> 
> best,
> 
> Matthias Geiger <werdahias>
> 
> Debian Developer -- KiCad Contributor
> GPG: ECBEDBB607B9B2BE

Following your logic, no upstream software using LLMs, eg the kernel, is DFSG-compliant.

-- 
Pierre-Elliott Bécue