Hi, On Tue, 2026-11-03 at 10:13 +0000, [email protected] wrote: > HAm 03.11.2026 09:18 schrieb Matthias Urlichs: > > The DFSG team can use an automated SPDX checker, if everything in a > > package is SPDX'd, or they can use an LLM > > An LLM would be not only waste of resources but also not a reliable > "checker". SPDX is well defined. A linter/parser/checker (don't have the > correct term for that) like "reuse" does the job for you. "spdx2debian" > does use "reuse" in the back.
How exactly does SPDX help for projects without the annotations? Isn't reuse/spdx2debian a waste of resources for such projects? Even more than any LLM? Also the LLM the NEW queue uses isn't limited to checking copyright notices. > > You want the > > DFSG people to not use an LLM for that? > > Despite all the critics about LLM, it is defiantly not the right tool > for such a job. An LLM is not even able to count words in a text or > provide PEP8 Python code. An LLM is not for parsing/linting. Humans are not able to count words in text accurately count words either. Nor are they good at writing d/copyright. The bar for a LLM to reach human-level for that is not that high and probably sufficient as we accepted that quality before. Ansgar
