What if you could control the sources that go into your AI search engine?

At @Neeva, we're making that a reality.

We're excited to share we added #NeevaAI:
✅ Answer support for verified health sites, official programming sites, blogs, etc.
✅ Availability in the News tab

🧵 First, at @Neeva we're passionate about generative search engines combining the best of search & AI.

But it's clear generative AI systems have no notion of sources or authority.

Their content is based on their reading of source material, which is often a copy of the entire Web.

Considering using a LLM for production, but it's too slow?

Well fear not! Today you can with Double Pseudo Labeling!

Here's why the @Neeva team did, and we've never looked back!

🧵 Before we get started, let's briefly go over #NeevaAI and what it is...

✅ Our AI generates answers
✅ Our Search makes sure answers are timely & factual

All in all, @Neeva combines search + AI to generate a single, cited for each of your searches.

Yesterday, we talked about why AI chatbots generate frankenanswers.

As promised, today we're going over how NeevaAI implements our solution to frankenanswers.

BIG thanks to our talented AI/ML team: @avinashparchuri @rahilbathwal

Keep reading for how we tackle this problem ⤵️ And if you didn't read our thread on this yesterday, check out part 1 below.

(we promise it makes more sense if you do! 😜)

https://twitter.com/Neeva/status/1623778167193206784

Have you seen ChatGPT combine info on multiple entities into an answer that’s completely WRONG? 😬

Generative AI and LLM models can mix up names or concepts & confidently regurgitate frankenanswers.

Neeva is solving this problem on our AI-powered search engine.

Here’s how 🧵 FYI This is a two-part thread series.

Today, with the help of @rahilbathwal, we’ll explain why the problems happen technically.

Tomorrow, we’ll talk through how we’re implementing our solution with our AI/ML team.

Make sure you're following... 👀

1/ NeevaAI serves abstractive summaries of web pages that are generated in real-time.

We achieved this by a ~10x reduction in latency of a fine-tuned t5-large encoder-decoder model.

TY @asimshankar, @rajhans_samdani, @AshwinDevaraj3 + @spacemanidol

See our lessons learned.. 🧵 2/ First off, we found that there are far fewer resources available for optimizing encoder-decoder models (when compared to encoder models like BERT and decoder models like GPT).

We hope this thread will fill in the void and serve as a good resource. 📂

1/ At #Neeva, design is the act of giving form to an idea: we gather data and inspiration, think, make, and iterate through feedback. 💡

Here's how our team, working alongside the ✨Neeva Community✨, shaped our latest news tool, #BiasBuster...

(read on 📖) 2/ To improve the news experience on Neeva, we solicited insights from users of various news outlets.

One early finding 👉 the journey to get daily news typically started from news providers' sites and apps, but NOT from a search engine.

🤔

1/ Have you heard? Bias Buster is now available in #Neeva's main search tab!

🔎 Try a search here: neeva.com/search?q=calif…

And if you're wondering how we crawled and evaluated topics to create our 5 point scale slider, stay tuned! 🤓

We dive into it on this thread 🧵… 2/ Our goal? 👉 Show a variety of POVs on particular news topics.

To reach this goal, we categorized results based on 5 buckets to ensure a smooth experience while interacting with the slider. This includes:

🪣 Far Left
🪣 Left Leaning
🪣 Center
🪣 Right Leaning
🪣 Far Right

🥁 Introducing #NeevaAI 🪄

🌟 Powerful AI
🌟 Cutting-edge LLMs
🌟 Authoritative answers in real-time (solving major drawbacks of ChatGPT)

It’s like nothing we, or anyone, have built before.

US users w/ a Neeva account can try it now or create one here: neeva.com/search?q=shoul… While ChatGPT is a groundbreaking product, it comes with two significant drawbacks.

1️⃣ ChatGPT’s output does not include sources or references, making it impossible to determine the credibility of an answer.

2️⃣ ChatGPT does not retrieve real-time data or information.

1/ We redesigned our marketing home page and got 50% more clicks, and 20% more installs.

Want to learn how we did it? 🧵⬇️ 2/ A picture is worth 1k words 🖼

Instead of lines of text telling people why to use #Neeva, we show them!

Our new website visualizes installing Neeva so they:
🅰️ Know what to expect
🅱️ Show them the benefits of tracker blocking 🛑 and address bar search 🔎

1/ It's not the size, it's the skill - now releasing #Neeva's Query Embedding Model!

Our query embedding model beats @openai’s Curie which is orders of magnitude bigger and 100000x more expensive. 🤯

Keep reading to find out how... 📖 2/ Query understanding is the life blood of #searchengines. Large search engines spent millions of SWE hours building various signals like synonymy, spelling, term weighting, compounds, etc.

We don’t have that luxury. 🙄

Fortunately for us, #LLMs are here to build upon.

1/ Google will do just about anything to maintain its monopoly power

Fear-inducing pop-ups with misleading designs to trick users into going back to Google search ✅

What’s a competing search engine to do? The only thing we can…Design and innovate our way out of it! 🧵 2/ By default, Google Chrome comes with Google search – no surprise there.

However, if a user prefers a more privacy focused search engine, they have to jump through a few hoops to install an extension and make it as the default.

All in all not terribly difficult so far, but…

(1/) You know the drill. Head to a website, and get hit with a pop up asking to enable cookies.

Too often we just click “accept all” to get rid of the prompt, privacy be damned

Rather than 🤷‍♀️, we set out to get rid of these annoying banners & keeping your privacy intact.

🧵 (2/) Why should you care about cookies?

Sites save these small text files on your device and use them to track you.

Little crumbs that track sites you visit, what you shop for, and even what you search for.

That private & valuable info is sold and ads are targeted back at you.

1/ When someone types “neeva” into search, how do we know they mean “neeva.com” instead of “neevaneevaneeva.com”? After all, the second has 3 times as much neeva!

See how you can do much better than vanilla TF-IDF / cosine similarity for textual relevance!🧵 2/ Textual relevance is only one part of document ranking (alongside signals like centrality, page quality, and click rate)

But it’s one the most important parts and the one we’ll be covering in today’s thread.

(1/) Competing in search starts from crawling the web. You can't serve up search results if you can't crawl the web.

Crawling the web is very hard today because of the discriminatory and #anticompetitive nature of how crawlers are treated on the web. 🧵 (2/) We need "crawl neutrality" on the Internet, similar to net neutrality
@matthewstoller @econliberties @EFF @superwuster @amyklobuchar @ChuckGrassley @RepAnnaEshoo, @Deese44 @OmidyarNetwork @accountabletech, @ChuckGrassley, @lutherlowe @Public_Citizen @publicknowledge @amprog

The Next Google? @dmitribrereton author of “Google Search is Dying” calls @Neeva “The Everything Engine.”🔎

✅ Transparency, agency and deep integrations.

No matter what you’re looking for, or where it is, Neeva will help you find it, often with only one click. 🕵️‍♂️

More ⬇️ Search across the accounts you choose to connect to Neeva and the web, all from one search box. 🔎

See results from Gmail, G Drive, Dropbox, Office 365, Slack, Github, and more. All with one click.