We’re seeing a shift in #AIbullshit (aka #hallucinations). It used to be about invented facts, like fake academic references you could easily debunk with a Google search. Now, #DeepResearch tools generate impressive reports, but often with a deceptive synthesis: reality is rearranged, and true facts are combined in a false but plausible narrative that’s much harder to detect.

I've been testing Perplexity's free version (limited to five prompts), and I’ve found that it either takes too long to realize the result is unusable or it’s immediately obvious.

One interesting, easily verifiable test: ask it about your own CV. In my case, much of the information was accurate, but the tool fabricated details, like claiming I coordinated certification standards for 23 AVT specializations (false) based on the fact that I’m VP of ESIST (true).

Its sycophantic tendencies are also evident: it slightly inflates my h-index (from 13 to 16) and citation count (from 774 to 800+).

All of this applies when searching in English. A simple prompt in Greek returned someone else’s CV, so another attempt was required.

I have no idea whether the long reports these tools publicize have been verified by anyone. And asking DeepResearch to verify them only adds to the confusion, since it often marks true information as unverified.