How Did Perplexity Become an Answer Engine?

Henry Modisett says Perplexity became an answer engine through truly iterative experiments. A working setup already hit a database with SQL, and the team kept adding more text. More text meant people could ask questions beyond what anyone expected the tool to do.

They were not thinking about Google or about search. They were thinking about asking questions of a large data set. Someone changed the backend and plugged it into a Bing search API. Suddenly a person could ask anything and get an answer. After many years at Quora spent on how people obtain knowledge and how information is organized on the Internet, an on-demand question with an instant answer was what they had always wanted. It lived in a Slack bot, not a product. Because it worked immediately, an iPhone app and a company felt obvious. If everyone could ask a question on a phone and get the answer instantly, that had to be valuable.

The path was a six month project with no plan beyond a hunch, stuck in a hyper iterative cycle. They were onto something about a month before ChatGPT came out, right behind it. Even then, the models were dated by several months from the last training run.