The first employee we hired came and asked us about health insurance. Normal need, I didn’t care. I was like, “Why do I need a health insurance? If this company dies, who cares?” My other two co-founders were married, so they had health insurance to their spouses, but this guy was looking for health insurance, and I didn’t even know anything.

Who are the providers? What is co-insurance, a deductible? None of these made any sense to me. You go to Google. Insurance is a category where, a major ad spend category. Even if you ask for something, Google has no incentive to give you clear answers. They want you to click on all these links and read for yourself, because all these insurance providers are bidding to get your attention.

We integrated a Slack bot that just pings GPT 3.5 and answered a question. Now, sounds like problem solved, except we didn’t even know whether what it said was correct or not. In fact, it was saying incorrect things. We were like, “Okay, how do we address this problem?” We remembered our academic roots. Dennis and myself were both academics. Dennis is my co-founder. We said, “Okay, what is one way we stop ourselves from saying nonsense in a peer reviewed paper?”

We’re always making sure we can cite what it says, what we write, every sentence. Now, what if we ask the chatbot to do that? Then we realized, that’s literally how Wikipedia works. In Wikipedia, if you do a random edit, people expect you to actually have a source for that, and not just any random source. They expect you to make sure that the source is notable. There are so many standards for what counts as notable and not. He decided this is worth working on.

It’s not just a problem that will be solved by a smarter model. There’s so many other things to do on the search layer, and the sources layer, and making sure how well the answer is formatted and presented to the user. That’s why the product exists.

• Aravind Srinivas

这似乎就是 pplx 的起源了。