Hacker News Reader

Comment by streets11703 | Hacker News Reader

(co founder) We have worked on neeva search for four years and have really sweated building out a real system with a massive crawl, index and serving systems. We see AI as a way to get people to answers faster and provide an overall better experience. We tried hard to tune neevaAI to make sure that it doesn't hallucinate. And to be very clear, we are using LLMs to help distill information from sites and aren't pretending to be the source of truth. Hence the citations for each and every sentence.

View on HN

How are you tuning it to make sure it doesn’t hallucinate? I think that’s a fundamental property of neural networks (see, for instance, the adversarial examples literature, or the recent work demonstrating adversarial exploits in AlphaZero style agents in Go).

Interesting.

I've always viewed there as being a sort of inherent contradiction between chat search and document search. You can do either one well, but not both.

Although as far as I understand this approach of yours seems to still suffer from having to identify the most reliable source(s) to interrogate first... which is essentially The Problem™ of search and if you can do that well enough to make a conversational search engine give good answers, then you already do search way better than Google's ever done.

To be clear, my comment wasn't a critique - if you're an early-stage startup, it makes sense to try new things and attempt to catch new trends early, since they can carry you past your established competitors.

I think the search engine with citations is a great idea - you're grabbing onto a big trend while solving a clear, hugely-discussed problem (you can't trust LLMs to give truthful answers, and that's exacerbated by the fact that you can't see where their info is coming from). Hope it works out for you.

> To be clear, my comment wasn't a critique...

GP is responding to other naysayers without directly replying to them (:

And to be clear, I think going down the LLM rabbit-hole is absolutely the thing for them to do (see the quip about AI being their "core biz"), but you wonder why it wasn't them that came up with ChatGPT-like system in the first place, if disrupting is what's Neeva is about? Though, I don't think Neeva is out to disrupt Google Search when the differentiator is merely a different business model; I mean, per definition of the term as Christensen meant it [0], it is pretty hard to disrupt free.

[0] What is disruptive innovation (2016), https://archive.is/IzaKy (hbs.org)

Is it fair for publishers to scrape their info and then provide it so users don’t even give them traffic?

Reply on news.ycombinator.com