Last reviewed
Correct answer: A. It actively searches the web for each query rather than relying only on fixed training data
Explanation
A language model's parameters are frozen the moment training ends. Everything it knows was baked in then, so a question about something that happened last week has no answer sitting in the weights, because the event was never in the data. That boundary is the knowledge cutoff. Perplexity closes the gap by searching at the moment the question is asked and handing the retrieved pages to the model, which is why its documentation points to web search for current information, recent news, and questions reaching past training data.
The caveat matters as much as the mechanism. Retrieval makes an answer only as good as the pages retrieved, and a stale post or an early, mistaken report gets cited just as confidently as a correct one. Open the source and check its date before repeating the claim.
Each wrong option breaks somewhere different. Factual questions are answered rather than refused. Models are still very much involved, since retrieval feeds a model instead of replacing one. And recency is the strength on offer here, not a restriction to history.
Sources
“Use it for current information, recent news, source-grounded research, and questions that need information beyond the model's training data.”
“The web search tool gives Claude direct access to real-time web content, allowing it to answer questions with up-to-date information beyond its knowledge cutoff.”
Practise 2 questions on this topic
Take Perplexity Basics — Timed Test (2 questions) — scored instantly, explanation for every question, no login.