When it comes to search engines, there is a holy grail. Think of human intelligence, as if you were asking your question to a friend, but the friend exists through the computer screen and in the digital airwaves, a pansophic Internet librarian who rummages through virtual bookshelves to find 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 smart: we process queries into a basic hodgepodge of keywords and clunky half-phrases, then put in more brainpower to make the language foolproof so search algorithms can’t go astray. But George Sivulka, an artificial intelligence scientist, 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’, launching today.
Since its inception in August 2020, Hebbia has built an investor list that includes Peter Thiel (as one of just six pre-seed investments, following in the footsteps of Facebook, OpenAI and Deepmind), Stanley Druckenmiller, Yahoo founder Jerry Yang, board members of Google parent company Alphabet, former chairman of the Federal Communications Commission, and venture capital firm Index Ventures. It recently closed a $30 million Series A financing round.
Tools to think about, not keywords
The company’s raison d’être stems from the fact that in today’s market, dominant commercial search engines, including Google, Yahoo, Bing, Baidu, and DuckDuckGo, are all keyword-driven. They must be able to identify words from searches within web pages they access; they then rank the pages through an in-depth synthesis of consumer statistics, word counts, and semantic mapping (for example, past data may have made connections between thematically similar words like “sun” and “sun”). That model has its limitations – in a sense you already need to know something about the results you’re looking for.
But Hebbia hopes to push the limits by offering keyword-agnostic answers, using a series of machine learning-trained neural networks. If you were to search for “What is the meaning of life?” Google could launch a blog post with the same title, but Hebbia could go deeper and bring you a wealth of scientific literature from famous existential philosophers. According to the company, it outperforms other state-of-the-art search algorithms by 57% 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 mind-boggling genius of AI minds like GPT-3 have astonished the public. But when it comes to search, NLP is severely underused by companies that found it interesting from an academic perspective, but perhaps not commercially. In recent years, Google has tinkered with “transformer” NLPs called BERT and MOM, but stopped integrating it into the core of its search engine; the current transformer architecture is minimal. For Sivulka, the question was poignant: “How do we apply and manufacture this new technology to build tools for thought?”
Peer in the mysterious ‘Deep Web’
Hebbia’s first product, launched two years ago, was a Ctrl-F function that allowed users to search more intelligently in on-screen text, now followed by a standalone search engine. But the company’s business model also leans on its academic roots: it’s carving out a specific niche in the financial, legal and medical fields, where analysts toil for details buried in mounds of documents and transcripts (they could now hopefully be typing: “What are the latest sales figures” or “When was the patient diagnosed” and have answers at your fingertips in seconds). To achieve this, Hebbia is working with institutions and governments to develop a database of setting up internal private papers and texts – a solution that, according to Sivulka, will also attack one of Google’s pain points.
The data on the web is vast – almost unfathomable, with experts claiming that the ‘Deep Web’, that invisible to most web crawlers, is about 500 times larger than the web we know, but it is estimated that Google indexed only 4% of the world’s data. A large part of reaching the final 96% is due to the cooperation of the data providers who are reluctant to make their valuable archives fully public, such as Hebbia’s partner institutions. For them, Hebbia could provide peace of mind. Since the algorithms do not require quick access to text keywords, it is able to encrypt all content stored in the index, meaning data is safe even in the event of a hack.
Hebbia isn’t a Google killer just yet, but Sivulka says he hopes eventually to expand into Google’s realm of public search engines that crawl the World Wide Web, perhaps as a hybrid product — probably for an experienced “knowledge worker” — that will log into a personal database of highly specialized information, operating within a broader search engine. He notes that Hebbia is already indexing public documents, including Congressional Inflation Reduction Act and the Bipartisan Infrastructure Act, as well as Johnny Depp’s libel lawsuit files, and he envisions all of this increasing over the next six months to a year.
A powerful status quo
The idea of a better search engine may seem like a good idea. So why isn’t it done yet? Google, with its trillion dollar war chest, may have all the ammunition needed to build a version of Hebbia. But according to Sivulka, it is hampered by an “innovator’s dilemma”: the current business model is far too lucrative, because its dominance in the search market allows it to sell valuable ads based on its traditional method of keyword search and page rankings.
“If you talk to a founder who believes in neural search and computers who understand you,” Sivulka says, “I think it’s inevitable that Hebbia, or some other neural search company, will take over Google.”
The investor, Mike Volpi of Index Ventures, has a more tempered view for Hebbia. “I’m very happy that our founders have great ambitions, but I don’t think so [Hebbia] is going to challenge Google in public searches,” he says. “I don’t see a consumer business in the near future. It is a tool that Index Ventures, or londonbusinessblog.comcould buy to do intensive research.”
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 her, if any of them seem smart or have a cool business plan, send them my way.” ”, he laughs). Building a business requires “a special type of person,” he says — not only sharp, but charismatic, “a personal, commercial, ordinary person. Bee [Index Ventures], when we invest, that’s a big factor. Where you end up isn’t always what you pointed to when you started.”
Sivulka began his first job at NASA at age 16, working on satellite landmine detection software, then doing physics research on a dark matter detector for Stanford’s U.S. energy department, becoming the fastest student to earn a math degree. to Stanford. He then dropped out of his Stanford PhD program in computational neuroscience to start Hebbia.
The name, he says, was inspired by his PhD 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 link between the artificial and the biological, or the natural and the human.” The training for that algorithm went through a process called Hebrew learning. Now, the next evolution, he asks: How do we train neural networks that resemble humans?