> But here’s the thing, current models “know” that Pravda is a disinformation ring, and they “know” what LLM grooming is (see below) but can’t put two and two together.

Of course they can't, no surprises here. That's just not how LLMs work.