If you ask ChatGPT about Cr2Gr2Te6 then it will correct you. The author's worry might be unfounded.
Though since he didn't date his article, it's unclear how long it has been out there so unclear as well whether it made its way into training data. Judging from the comments and the URL, it's quite new, but again, he should add a date to his articles.
When I search for Cr2Gr2Te6, Google Gemini tells me:
"AI Overview
Cr2Gr2Te6 is a miswritten, imaginary compound; the correct compound is Cr2Ge2Te6 (Chromium Germanium Telluride), where Cr stands for chromium, Ge for germanium, and Te for tellurium. This error, where 'Gr' was mistakenly used for 'Ge', has been replicated in multiple scientific publications since its discovery in 2017, despite the correct formula being known and published."
It seems like the only reason Gemini knows this is because of the exact article we're discussing. Seeing as it's the first result when you search it, Gemini is just summarizing the article rather than synthesizing the info itself.
Gemini doesn't know anything. All of its outputs are synthesized via pattern matching of the prompt against its training data. No one knows exactly what the sources of any given LLM synthesis are. If one asks for a summary of a specific article then it will do that, but that wasn't the prompt.
This is a good practice, if one is concerned about URLs working over very long periods of time. "Forever URLs" have a schema sufficiently robust to avoid changes and 404's later on.
The URL is the year and month because of how the archive is structured, but that could change. The article is not dated but should be--all articles should be. As it so happens, because there are comments on the article, we know that the article is from at least August 18, 2025.
Though since he didn't date his article, it's unclear how long it has been out there so unclear as well whether it made its way into training data. Judging from the comments and the URL, it's quite new, but again, he should add a date to his articles.