That most of everything is crap usually isn’t a problem for machine learning. You use them to modify works (photography), or in non-critical contexts (phone auto-completes), or as checks (grammarly).
GitHub Copilot is unique in that it risks creating a bad code feedback loop