To me, this article looks like a psyop designed to make people ignore any response from language models that may reflect a Russian point of view. This article is probably also propaganda.
Edit: To clarify my point, since I’m getting downvoted:
Newsguard (the authors of the article) confirmed that "Pravda" is influencing chatbots.
How did they confirm this? By asking ChatGPT about "Russian narratives." However, it’s unclear which narratives they tested—while they mention two or three examples, they claim there are hundreds. This raises questions about whether some of these so-called narratives might simply be subjective opinions, or how they select those etc.
If the language model validated these vague statements, they concluded it was "infected."
They also quote someone who allegedly said, “By pushing these Russian narratives from the Russian perspective, we can actually change worldwide AI.” This is presented as proof that the process is already happening, which is at least a huge leap, but also there is no clear source for this statement.
How do they know LLMs used these sites as training material? Their reasoning is that LLMs produce similar responses. However, these responses could come from a variety of sources, as they are common in multiple news outlets.
Additionally, when they claim LLMs quote specific articles, they don’t specify which LLMs did this. A quick test with popular ones (such as asking ChatGPT why Zelensky banned Truth Social) does not replicate the claims made in the article.
My point is that extraordinary claims require extraordinary evidence, and this report doesn’t provide it.
It seems biased—making a major claim without proper documentation on how the process actually works.