My experience is that Perplexity is slightly better at providing facts than ChatGPT (in default mode), probably because (almost) everything comes from a source, not just the model's training set. Although Perplexity does mess up numbers as well.
My most recent experiment:
How many Google CEOs there have been?
Followed by the question
So 3 CEOs for 27 years. How does that number compare to other companies of this size
ChatGPT just completely hallucinates the answer -- 5 Microsoft CEOs over 50 years, 3 Amazon CEOs over 30 years, 2 Meta CEOs over 20 years which are just obviously wrong. You don't need to do a search to know these numbers -- they are definitely in the training dataset (barring the small possibility that there has been a CEO change in the past year in any of these companies, which apparently did not happen)
But Perplexity completely nailed it on first attempt without any additional instructions.
My most recent experiment:
How many Google CEOs there have been?
Followed by the question
So 3 CEOs for 27 years. How does that number compare to other companies of this size
ChatGPT just completely hallucinates the answer -- 5 Microsoft CEOs over 50 years, 3 Amazon CEOs over 30 years, 2 Meta CEOs over 20 years which are just obviously wrong. You don't need to do a search to know these numbers -- they are definitely in the training dataset (barring the small possibility that there has been a CEO change in the past year in any of these companies, which apparently did not happen)
But Perplexity completely nailed it on first attempt without any additional instructions.