This is likely because the blog is AI generated and keys off this point from Karpathy: "As a preview, by the end of the script our model will generate (“hallucinate”!) new, plausible-sounding names.", so the LLM just repackaged that into something that is obviously wrong, which is kind of ironic.