I used #Perplexity’s #deepresearch feature to help me make a decision.
I provided it the context, the goal and four non-confidential documents, one was quite extensive (300 pages), the others were short.
First, I was impressed by its Greek. It had only one or two grammar errors. Its ability to synthesize information was also remarkable.
It generated a synthesis and even pulled in information from external sources. The result sounded logical, with clear inferences, but it wasn’t fully grounded in real sources. Verification was difficult, and in one crucial case, it hallucinated: It fabricated a sentence, presenting it as if it came from one of the documents I had provided, but it wasn’t there. Worse, that sentence would have been the most crucial for my decision.
I had to choose between four possible courses of action, and the proposal it made was contradictory. If you’re familiar with the Oracle of Delphi, it was like consulting Pythia: totally ambiguous.
Ultimately, it wasn’t useful for my specific purpose. If not for the side benefit of testing its capabilities, it would have been a complete waste of time.
I wouldn’t trust it for tasks where verification is difficult. But when I already know the correct answer and can easily cross-check the information, it is amazing.

The jury is still out for me, I will keep testing.