I know they are doing their funding (huge ad in their search bar for paid customers) and this is probably a hype piece sponsored by them. Anyways, I've been using them for a bit.
Perplexity is improved web-search but compared to ChatGPT and Claude it is quite lacking in depth.
It can't do as good an answer as ChatGPT and it can't do as good a retrieval as Claude. Yet, they claim to use ChatGPT and Claude in the backend.
I believe to save API costs and token costs, they are using tricks to summarize and make the queries small to the actual APIs.
Good for asking short questions but for digging into details it is pretty bad. It just spits out summary back at you or acts like a bad retrieval system and gets a tiny portion back and slightly rewords a sentence back to you.
Also it is very prone to giving you the wrong answers depending on the first web search results it gets.
All it seems to me is that this is a very shiny coat of javascript paint on top of web search results API being passed to AI API. So, they've gotten the web search results from someone, then do summarization with probably open source models and then pass it on to a cloud LLM API.
It is a decent web search tool but a bad AI LLM tool. Much better than bing search results from ChatGPT (esp after the nerf).
If advertising and manipulation hits this, I don't know what kinds of shadow prompts will be added to your search results. Just small changes seem to really sway results quite a bit.