Did the Perplexity CEO say how they are different than a basic RAG setup with citations?

Aravind Srinivas, CEO of Perplexity, highlighted the unique aspects of their platform compared to a basic Retrieval-Augmented Generation (RAG) setup. He emphasized that Perplexity is designed as an "answer engine" where answers are directly backed by citations from human-created sources on the web. This ensures factual grounding and significantly reduces the chances of hallucinations by language models. Unlike a basic RAG model, which retrieves documents and uses them to generate an answer, Perplexity strictly enforces that only the retrieved information is used to formulate responses. If sufficient information is not available, the model is instructed to indicate the lack of sufficient data, thus maintaining accuracy and reliability 1 2.

To ensure that all its sources are "human-created," Aravind Srinivas explains that Perplexity meticulously follows a process similar to the citation principles used in academic papers. This involves verifying that the sources referenced are notable and credible. Like Wikipedia, which expects edits to be backed by reliable citations, Perplexity implements a strict sourcing policy to ensure the information provided comes from validated and reputable human-created content available on the web. This is part of their strategy to eliminate inaccuracies and reduce hallucinations by the language model 1.