If you build an AI search product, you’re competing with Google.
But it’s much easier for Google to answer a query with a single, simple answer like “How many is 12” than it is to answer a complex question like “How did Thomas Paine’s ‘Common Sense’ influence the ideals of the Enlightenment?”
That's why You.com is committed to answering the second type of question.
Buoyed by a new $50 million funding round, the AI company is looking to gain an edge in a perennially innovative yet often overlooked field where other multi-billion dollar AI companies have faltered.
As founder and CEO Richard Socher put it, “From first principles, where can we be 10 times better than Google?”
There would be no point competing on simple questions like basic facts, conversions and referrals, which make up the majority of Google searches.
“But the people who pay for You.com are productive knowledge workers,” Socher says, “and that's really the sweet spot of this technology, the killer app point: building this productivity engine that tells agents how and when to search the Internet.”
The term “productivity engine” may not be immediately intuitive, but the idea is that you can tell the system what you want to know, regardless of complexity, using natural language, just as you would tell a human assistant (it's close to an agent, but not the same).
For example, say you want to know about the side effects of a new drug, you can ask the system to “Summarize the literature on acute side effects of flimflamazone.”
A language model probably can't answer this kind of question right away. It's entirely possible that you've never heard of flimflamazone. In that case, you might admit to a lack of knowledge, or you might hallucinate. Even if you do have some knowledge about the drug, it's likely not up to date.
You.com focuses on these types of more demanding tasks, where agents must first research the query itself so they can arm themselves with the right information and techniques. In this case, agents must grade several papers online. Crucially, in this type of research, citations are deep linked and displayed in context. So when you look at an assertion or figure, you'll see a clickable citation that not only takes you to the source document, but also highlights the relevant text.
Image credit: You.com
Socher also gave an example of how the model estimated how much to invest in index funds to ensure the fund would cover Stanford University tuition when a child turns one year old.
Walking through the process step by step, the model said he first had to run a search to find out the average yield of compound interest funds, the average cost of education at Stanford, the average age to enter college, the inflation rate, etc. Given these, the model drew up a Python script to calculate how much different seed amounts would grow, eventually arriving at a reasonable answer (about $51,000).
You can have Claude or ChatGPT do something similar; in fact, You.com relies on these and other models for its LLM feature. But Claude, for example, can't find new documents to reference, and ChatGPT isn't as meticulous about sources and processes. Socher says You.com's goal is to get it right the first time, every time, by carefully controlling which models to prompt with.
Image credit: You.com
He also showed off a demo of what he calls a “multiplayer” AI, which is essentially a shared AI workspace where multiple users can add documents, summarize, ask questions, and perform other “productivity engine” type tasks, but remain fully visible to other users.
Socher says You.com's service is good for business, too: While others are racing to the bottom, he's worked his way up the food chain, winning more and more paying customers. You.com's subscriber base is five times larger than it was at the start of the year.
“Companies are raising money to offer their products for free, but they don't give enough thought to chat ads,” he said. “We're becoming more cautious about it and we think now is the time to scale.”
He didn't name any specific names, but said that some large companies are essentially using You.com to handle certain queries that their systems receive. I can imagine that large companies offer specific automation services, but are limited in how much they can handle with their internal models and APIs. You.com is expensive, but if it gets the job done, it could have a place in their stack.
Socher said the outlook is bright, and investors seem to agree. The $50 million B round was led by Georgian, with participation from DayOne Ventures, DuckDuckGo, NVIDIA, Salesforce Ventures, and SBVA (formerly SoftBank Ventures Asia). The amount was slightly lower when Socher spoke, but the fact that millions more had been added as of press time makes this a notable round.
While the round is certainly large, it may seem trivial compared to what You.com's multi-billion dollar competitors are raising. But those companies have astronomical run rates, given the growing headcounts, eye-watering hardware investments, and server fees they have to pay. The strategy seems to be to frontload costs to go to market, and while there's a good chance of success, the list price is in the ten-figure range.
But for now, You.com is making money from at least some of the companies.
“We have positive unit economics when we work with large enterprises, some of whom use us millions of times a day,” he said.
The idea that AI shouldn't cost hundreds of millions of dollars just to exist seems novel today, but if You.com can execute this strategy, it could soon catch on.