Maybe whoever trained their ML models actually thought no one would call good first issues "good first issues". Kinda like you're trying to build a cat recognizer and you suddenly feed it pictures of cats wearing signs around their necks that say "CAT" in all caps. The recognizer would probably fail.
For example: https://github.com/ClangBuiltLinux/linux/contribute shows: > This repo doesn't have any good first issues, yet
But I have many open issues tagged "good first issue": https://github.com/ClangBuiltLinux/linux/issues?q=is%3Aissue...