Search engine

a new “neural” search engine hopes to transform the market

In the field of search engines, there is a holy grail. Think of human intelligence as asking your question to a friend – but the friend exists through the computer screen and in the digital ether, a pansophical internet librarian who will rummage through virtual shelves to retrieve the answers you seek.

There is no such librarian in the world today. To use modern search engines, we often pretend the machine is dumber than it is smart: we digest queries into a basic mix of keywords and clunky half-phrases, then deploy more intelligence to foolproof the language, so that search algorithms cannot go astray. But George Sivulka, a specialist in artificial intelligence, wants to end the era of what he calls “Googlese”. He is the founder of Hebbia, a two-year-old startup building the world’s first “neural search engine”, due to launch today.

Since its inception in August 2020, Hebbia has amassed a roster of investors that includes Peter Thiel (as one of six pre-seed investments, after Facebook, OpenAI and Deepmind), Stanley Druckenmiller, Yahoo founder Jerry Yang, Google board members parent Alphabet, former chairman of the Federal Communications Commission and venture capital firm Index Ventures. It recently closed a $30 million Series A funding round.

Thinking tools, not keywords

The company’s rationale stems from the fact that in today’s market, the dominant commercial search engines including Google, Yahoo, Bing, Baidu and DuckDuckGo are all powered by keyword matching. They must be able to identify words from queries in all web pages they launch; then they rank the pages through an abstruse synthesis of consumer statistics, word counts, and semantic mapping (for example, past data may have linked thematically similar words like “solar” and “sun”) . This model has its limitations – in a way, you must already know something about the results you are looking for.

But Hebbia hopes to transcend limitations by offering answers keyword-agnostically, using a set of machine-learning trained neural networks. If you were looking for “What is the meaning of life?” Google might create a blog post with that very title, but Hebbia might take it a step further, giving you a treasure trove of scholarly literature from famous existential philosophers. According to the company, it performs 57% better than other leading search algorithms in finding the most optimal results.

Make no mistake: the technology for understanding like a human, called natural language processing (NLP), has been around for years. The the breathtaking brilliance of AI minds like GPT-3 have already wowed the crowds. But when it comes to research, NLP is vastly underutilized by companies who thought it was interesting academically, but perhaps not commercially. In recent years, Google has tinkered with NLP “transformers” called BERT and MOM, but failed to integrate them into the core of its search engine; its current transformer architecture is minimal. For Sivulka, the question was glaring: “How do we apply and produce this new technology to build thinking tools?

Look into the mysterious “Deep Web”

Hebbia’s first product, spear two years ago, was a Ctrl-F function that allowed users to search text on screen more intelligently, now followed by a standalone search engine. But the company’s business model also draws on its academic roots: it carves out a specific niche in the financial, legal and medical spheres, where analysts are bent on searching for details buried in mountains of documents and transcripts (they could now hopefully type in: “What are the latest sales figures” or “When was the patient diagnosed” and have the answers in seconds). To achieve this, Hebbia is partnering with institutions and governments to create a database of private internal documents and texts — a solution that Sivulka says will also tackle one of Google’s weak spots.

The data on the web is vast, almost unfathomable, with experts claiming that the “Deep Web”, which is invisible to most web crawlers, is about 500 times larger than the web as we know it, yet it’s estimated that Google has only indexed 4% of the world’s data. Much of reaching the final 96% involves the cooperation of data providers who are reluctant to make their valuable archives fully public, such as Hebbia’s partner institutions. For them, Hebbia could offer peace of mind. Because its algorithms do not require quick access to text keywords, it is able to encrypt all content stored in its index, which means the data would be secure even in the event of a hack.

Hebbia isn’t a Google killer yet, but Sivulka says he hopes to eventually branch out into the realm of Google’s public search engines that crawl the World Wide Web, perhaps as a hybrid product – probably for a skilled “knowledge worker” – who will connect to a personal database of highly specialized information, operating within a larger search engine. He notes that Hebbia is already indexing public documents, including Congress’s Cut Inflation Act and bipartisan infrastructure bill, as well as documents filed in Johnny Depp’s libel trial, and he imagines that all of this will intensify in the next six months to a year.

A powerful status quo

The idea of ​​a better search engine may seem like a no-brainer. So why hasn’t it been done yet? Google, with its trillion-dollar war chest, may have all the ammunition it needs to build a version of Hebbia. But according to Sivulka, he is held back by an “innovator’s dilemma”: his current business model is far too lucrative, as his dominance in the search market allows him to sell valuable ads based on his traditional method. keyword research and page ranking.

“If you’re talking to a founder who believes in neural research and computers that understand you,” Sivulka says, “I think it’s going to be inevitable that Hebbia, or some other neural research company, will take over Google. “

Its investor, Mike Volpi of Index Ventures, has a more tempered view of Hebbia. “I’m really glad our founders have big ambitions, but I don’t think [Hebbia] going to challenge Google in public search,” he says. “I don’t see a consumer business in the near future. It’s a tool that Index Ventures, or fast businesscould buy to do intensive research.

George Sivulkathe founder of Hebbia.

For Volpi, part of the draw is Sivulka himself, whom he first met through his daughter, a classmate of Sivulka’s at Stanford (“I told him, if the one of them seems smart or has a cool business plan, make sure to send him my way,” he laughs. To start a business, you need “a special kind of person,” he says, no only sharp, but also charismatic, “a pleasant, commercial and regular person. [Index Ventures], when we invest, it is an important factor. Where you end up isn’t always where you were pointing when you started.

Sivulka started his first job at 16 at NASA, working on satellite land mine detection software, then did physics research on a dark matter detector for the US Department of Energy at Stanford, then became the fastest undergraduate to complete a math degree at Stanford. He then dropped out of his Stanford PhD program in computational neuroscience to start Hebbia.

The name, he says, was inspired by his doctoral research, on a neural network that mimicked the patterns of a coral reef. “You could argue that if the coral reef had a soul, the AI ​​doing the exact same thing also had a soul,” he says. “It was this weird connection between the artificial and the organic, or the natural and the human.” Training for this algorithm followed a process called Hebbian learning. Now the next evolution, he asks: how do we train neural networks that look like humans?