On Wed, May 6, 2026 at 12:34 PM Ricardo Wurmus <[email protected]> wrote:
[...]
> In the same vein we import CRAN and Bioconductor packages
> automatically
> without requiring the use of LLMs in a loop:
>
>    https://codeberg.org/guix-science/guix-cran-scripts/
>
> These scripts rely on the CRAN importer/updater, which is what I
> use for
> bulk updates to the R package collection in Guix.

If someone was already packaging Guix then this project would be as
simple as importing CRAN.

> I think we'd benefit from improvements to the importers and
> updaters without encouraging the use of LLMs, which are orders of
> magnitude heavier in terms of complexity and compute resources.

LLMs use a tiny fraction of resources required for a build, and by
reducing builds are even eco-friendly.

> (This is, of course, beside the point of whether LLM output should
> or should not be included in Guix.  This tangent is a note on
> whether LLMs are really the tool of choice for package updates.
> It's disappointing to see a willingness to substitute simple,
> reproducible, and understandable code with LLMs.)

No one wants to replace simple tools with AI. The goal is to make
better tools. LLMs can of course use the existing tools.

Reply via email to