looks to innovation to chip away at Google’s search dominance

Founder Richard Socher's research helped set stage for today's AI leaps CEO Richard Socher may not be a household name, but if you travel in AI circles, chances are you know who he is. Beginning in 2014 when he wrote his influential doctoral thesis about natural language processing, he has helped pave the way for today’s headline-grabbing AI technology, a fact that doesn’t escape him as he tries to build his next-generation search engine.

He launched MetaMind, shortly after that paper was published, a startup that Salesforce acquired two years later. In fact, as chief scientist at Salesforce, he helped build the AI layer that the company calls Einstein.

After he left Salesforce, he started, a consumer search engine, in 2020. Obviously he’s up for a challenge, but he also recognizes that with time on his side, and an ability to innovate, he has two advantages that could allow his company to begin to chip away at Google’s search hegemony.

He’s not intimidated by the fact that fellow search engine startup Neeva, launched by two Google veterans, was acquired recently by Snowflake after failing to find product-market fit. He says his search engine is already much larger in terms of users than Neeva ever was.

But he also understands that what he is doing is more than theoretical; it’s a business, and he has to look to a future where he is not just burning money but also bringing it in. He has reached the point, three years into his startup journey, where he has to switch focus from growth and start concentrating on revenue, having raised a modest $45 million.

We helped get this party started

While AI research has been ongoing for decades, Socher and his peers at Stanford helped open the door to today’s breakthroughs with their groundbreaking research in 2014. He has continued his research and has published papers as recently as 2020 when he began putting all his efforts into his current company.

He says as a scientist, it’s been amazing to see his research put to work in the ways that it has. “I had this realization the other week, which is as a researcher, if you’re ahead of your time, you’re a visionary. And you get a lot of credit for people using your ideas later, and you’re like ‘Oh, wow, we invented word vectors and contextual vectors, and then prompt engineering and the single model for all of NLP and LLMs for protein design,” Socher told TechCrunch+.

As he pointed out, just a few years later, all of these things have become a huge part of what’s happening in AI right now causing such a sensation. “ChatGPT 2 and 3 came out trying to do everything in NLP with a single model, too, and now the prompt engineering wave is really big now,” he said.

He said by contrast, as a CEO, being early can be harmful because the market isn’t ready for you. But at the same time, it’s important to keep innovating and finding new ways to reinvent, like his company is attempting to do with search.

Let’s get innovative

One of his company’s chief strengths, as he sees it, is its ability to innovate. As fast as he comes up with ideas, the more people copy them. The company has indeed innovated, coming out with chat search before OpenAI, Microsoft, Google or others did. It has also released an app store, multimodal chat search, and it introduced citations, showing the source of a particular answer.

One of the biggest issues with LLMs is the hallucination problem where the model simply makes up an answer when it doesn’t know the actual one. One way to combat that is forcing the model to show its work by looking at the sources it used to deliver the answer, an approach that’s gaining in popularity, but one that Socher says was among the first to do.

“I think our users are learning that if the facts are really important, and it has a citation, then it’s much more likely to be correct. And then if it says a bunch of other things, and none of those have citations, then you probably want to double check on the links if it’s a really important fact for your life or decision-making,” he said.

The company’s app store, which is analogous to OpenAI’s plugins, launched several months before OpenAI publicly announced them, and a recent blog post points out that OpenAI appears to have put the idea on hold as it searches for product-market fit.

The same post indicated that multimodality, already available with, won’t be widely available until next year. With multimodality, instead of just a text link, you might get a code snippet, map or stock ticker, depending on the nature of the search. The company has also added native tools for writing and image creation inside the search engine.

“It’s clear that we have to continue to innovate. This space is so dynamic, there’s so much going on,” he said.

Of course, there’s also a 10,000-pound gorilla the company must reckon with, too.

I ain’t afraid of no search giant

As Neeva showed, even with a strong pedigree, you aren’t guaranteed success when it comes to competing in consumer search. Socher certainly understands that, but he doesn’t feel like it’s a zero-sum game, either. If he can begin to capture a piece of the multibillion-dollar search market, that would not be a bad place to start.

He says he’s glad the Neeva founders landed in a good spot but says his company had outgrown them by 4x or 5x in terms of just traffic. “We raised less money and try to be as frugal as we can. It’s not not cheap to run a search engine,” he said.

The app has tons of users now, and he understands that as he scales the company, it gets more costly to run. “We have grown the app to millions of users, and so the cost to serve has become very, very important for us, and we have actually been able to reduce the cost of serving by 70% on the LLM side and over 50% overall. So we are close to being able to run it at a similar cost to just a web index now,” he said.

On the revenue front, Socher said the company’s working on some ideas, including what it calls “private ads,” as well as some things he wasn’t ready to discuss publicly just yet.

The difference between this variety of ad and the kind you have on most other search engines is that the advertisement is confined to this application. He says that it won’t follow you around the internet the way that Google ads do.

Another issue for him is defensibility, both against other startups and the giants, something that all companies in the generative AI space will have to deal with. While he didn’t go into too many details, he is thinking about how to add more features.

“It’s very clear that distribution continues to play a major, major role in all of this, and without revenue, you just can’t play the big game. And so that’s why we’re focused mostly on revenue right now, rather than continuing to grow massively. I think once you have revenue in place, it sets you up for being able to grow,” he said.

You don’t necessarily start a company like this hoping to take down Google, but he feels like he has laid the foundation for becoming a viable search engine competitor without having to grab huge chunks in the short term.

“It’s such a vast market. You don’t need to eat this massive trillion-dollar company to do well, even though that is obviously our super long-term ambition, our hope. But as long as we’re growing — and on many metrics we are growing in double digits, week over week — we will look to continue to do that.”