If you build an AI search product, you compete with Google.
But Google has a lot easier time answering queries with a single, simple answer, such as “how many is a dozen?” than it does answering complex questions like “what influence did Thomas Paine’s ‘Common Sense’ have on Enlightenment ideals?”
That’s why You.com is betting the company on answering the second type of question.
Emboldened by a new $50 million funding round, the consistently innovative yet often overlooked AI company aims to excel where other AI companies raising billions falter.
As founder and CEO Richard Socher puts it: “Just from first principles, where can you be 10x better than Google?”
It would be futile to compete on the simple questions that make up the vast bulk of Google searches – basic facts, conversions, and references.
“But people willing to pay for You.com are people that do productive knowledge work,” Socher says. “And this is actually where the sweet spot, where the killer app for this technology is: making this productivity engine, telling these agents when and how to search the internet.”
While the term “productivity engine” may not be immediately intuitive, the idea is that you will be able to employ natural language to tell the system what you want to know, whatever the complexity, in the same way you might tell a human assistant. (It’s agent-adjacent but not the same thing.)
For instance, say you wanted to catch up on the side effects of a new drug. You could tell the system, “Summarize the literature around acute side effects of flimflamazone.”
A language model probably can’t answer this kind of question off the cuff. It’s entirely possible it’s never heard of flimflamazone — in which case, it might admit its lack of knowledge, or it might hallucinate an answer. Even if it does have some knowledge of the drug, it’s likely not up to date.
You.com is focusing on this kind of more demanding task, where the query itself needs to be examined first so the agent can arm itself with the proper information and techniques. In this case, it would need to go online and score a few papers. Importantly, for this kind of research, citations will be deep-linked and in context. So when you see a claim or figure, it will have a clickable citation that not only takes you to the source document, but also highlights the relevant text for you.
Socher also showed me an example of asking the model to estimate how much someone should invest in an index fund when their kid turns one, in order to ensure the fund grows to cover their Stanford tuition.
Explaining its process step by step, the model said first that it needed to perform searches to find out the average yield of a compound-interest fund, the average cost of a Stanford education, and the average age someone goes to college, plus inflation and some other stuff. Using those as its assumptions, it sketched out a Python script to calculate how much different seed amounts would grow, and ultimately arrived at a reasonable answer (about $51,000, if you’re wondering).
You could get Claude or ChatGPT to do something similar. In fact, You.com relies on these and other models for its LLM capabilities. But Claude, for instance, would not be able to go and find new documents to reference. And ChatGPT is less painstaking about its sources and process. Socher said that You.com’s goal is to get it right the first time, every time, by carefully controlling which models are prompted and how.
He also showed a demonstration of what he called “multiplayer” AI — essentially a shared AI workspace where multiple users can add documents, summarize or ask questions about them, and do other “productivity engine” type tasks, but with full visibility to others.
Socher said that You.com’s services compared favorably business-wise too. While others are racing to the bottom, he’s moving up the food chain and adding paying customers left and right — You.com has five times more subscribers now than it did at the start of the year.
“Companies are raising money so they can give away their product for free, and ads haven’t really been figured out for chat,” he said. “We’ve been more careful about this, and we think it’s time for us to scale.”
He wouldn’t name any names, but said some large companies essentially use You.com to handle certain queries their own systems get. One can imagine a large company offering certain automation services but having internal models or APIs that can only handle so much — if You.com is more expensive but gets the job done, it has a place in their stack.
Socher did make the outlook seem rosy, and apparently investors agree. The $50 million B round was led by Georgian, with participation from Day One Ventures, DuckDuckGo, NVIDIA, Salesforce Ventures, and SBVA (formerly Softbank Ventures Asia). The amount was slightly less when I talked with Socher; you know it’s a hot round when a few million get added to it while you’re writing the article.
Though the round is inarguably large, it may appear insignificant compared to those being raised by You.com’s billion-dollar competitors. But with swelling employee numbers, eye-popping hardware investments, and server bills to pay, the run rate of those companies is astronomical. The strategy appears to be that they are frontloading the cost of inventing the market — and they may well succeed, but the sticker price is in the 10-figure range.
But You.com is making money, at least from some companies, right now.
“The unit economies for large enterprise deals are positive — and some companies are using us millions of times per day,” he said.
The idea that AI shouldn’t cost hundreds of millions just to exist seems novel today, but if You.com can make this play, it may just catch on.
