Recently someone commited a large set of LLM (Large Language Model often wrongly called AI or Artificial Intelligence as at best they are an AI model, or better phrased machine learning model, not even a machine learning reality¹) assistance files to the pixelfed repository. This caused quite some turmoil resulting in pixelfed banning AI generated contributions.
This is for good reason: too many LLM generated commits lack quality control. This results in shoving the burden of the checks from the contributor to the volunteers of the project teams. Those volunteers are already overwhelmed anyway (there is always an applicable XKCD, in this case [Wayback/Archive] 2347: Dependency – explain xkcd).
The offending commit was [Wayback/Archive] Add AI docs · pixelfed/pixelfed@65dff00 · GitHub. The contributor was lazy enough to wrongly name this commit, as these weren’t docs in the sense of documentation, but a set of Claude Skills and other LLM slop².
That behaviour is exemplary for almost all LLM contributions: most humans and systems generating them take them for correct and just commit their content.
I am still not convinced that the main contributor switching within a few days from “oh, let’s add LLM stuff” into “nope, we are not going to allow LLM” vows for a strong soul.
The Pixelfed statement is via [Wayback/Archive] pixelfed: “Pixelfed bans AI generated contributions…” – Mastodon
