Regarding this contamination issue, I think the author has a point here but the evidence is just very weak - apparently on one specific test ChatGPT fails when answering recent questions.
I wonder if in fact a certain amount of "contamination" is required. When you search for a given problem on Google, you are more likely to find a good answer if every term is meaningful. If you scramble your search and use synonyms, of course your search result will take a hit.