If you create an AI search product, you compete with Google.
But Google has much more time to respond to requests with a single simple answer, as “how much is a dozen?” That he makes people answer complex questions like “what influence did he have the” common sense “of Thomas Paine on the ideals of enlightenment?”
That’s why You.com I bet that the company answers the second type of question.
Enhanced by a new financing cycle of $ 50 million, the company of innovative IA still innovative, but often neglected, aims to excel where other IA companies increasing billions of dollars.
As the founder and CEO Richard Socher says: “Just from the first principles, where can you be 10x better than Google?”
It would be futile to compete on simple issues that constitute the big major part of Google – Basic facts, conversions and references.
“But people wishing to pay for you.com are people who make productive acquaintances,” explains Socher. “And this is actually where the Sweet Spot, where the Killer app for this technology is: to do this productivity engine, to say to these agents when and how to search on the Internet.”
Although the term “productivity engine” is not immediately intuitive, the idea is that you can use natural language to tell the system what you want to know, whatever the complexity, in the same way you could say to a human assistant. (It’s the adjacent agent but not the same thing.)
For example, let’s say you want to catch up with the side effects of a new medication. You can say to the system: “Summarize the literature around the acute side effects of flimflamazone.”
A language model probably cannot answer this type of question on the cuff. It is quite possible that he has never heard of flimflamazone – in this case, he could admit his lack of knowledge, or he could hallucinate an answer. Even if he has a certain knowledge of the drug, he is probably not up to date.
You.com focuses on this type of more demanding task, where the request itself must be examined first so that the agent can arm with appropriate information and techniques. In this case, you should go online and mark some articles. Above all, for this type of research, the quotes will be deeply linked and in context. So, when you see a complaint or a figure, it will have a clickable quote which brings you not only to the source document, but also highlights the relevant text for you.
Fin also showed me an example of a request to the model to estimate how someone should invest in an independent fund when their child makes one, in order to ensure that the fund is developing to cover its Stanford tuition fees .
Explaining its step -by -step process, the model first declared that it had to carry out research to discover the average yield of a compound fund of interest, the average cost of a Stanford education and the average age that Someone goes to college, the more inflation and some other things. Using these as hypotheses, he sketched a Python script to calculate the quantities of different seeds increase, and finally arrived at a reasonable response (around $ 51,000, if you are wondering).
You could ask Claude or Chatgpt to do something similar. In fact, you.com relies on these models and in others for its LLM capabilities. But Claude, for example, would not be able to find new documents with reference. And Chatgpt is less meticulous of its sources and its processes. Socher said that you.com’s goal was to do things the first time, each time, carefully controlling which models are invited and how.
He also showed a demonstration of what he called “multiplayer” AI – essentially an AI work space where several users can add documents, summarize or ask questions about them and perform other tasks of Type “productivity engine”, but with complete visibility to others.
Socher said that you.com services are also compared to business. While others run down, it rises the food chain and adds paid customers on the left and right – you.com has five times more subscribers now than at the start of the year.
“Companies collect funds so that they can give their product for free, and advertisements have not really been determined to discuss,” he said. “We have paid more attention to that, and we think it’s time for us to scale.”
He would not name any names, but said that some large companies mainly use You.com to manage certain requests that their own systems obtain. One can imagine a large company offering certain automation services, but having internal models or APIs which cannot manage as much – if you are more expensive but does the work, it has a place in their battery.
Socher has published the pink perspectives, and apparently investors agree. The $ 50 million tour B was led by Georgian, with the participation of Day One Ventures, DuckduckGo, Nvidia, Salesforce Ventures and SBVA (formerly SoftBank Ventures Asia). The amount was a little less when I spoke with wizards; You know it’s a hot round when a few million dollars have been added while you write the article.
Although the tour is undoubtedly large, it may seem insignificant compared to those raised by competitors of a billion dollars from you.com. But with the numbers of inflated employees, material investments and server bills to pay, the execution rate of these companies is astronomical. The strategy seems to be that they are on fronts the cost of the invent of the market – and they may well succeed, but the price of the stickers is in the 10 -digit range.
But you.com earn money, at least of certain companies, at the moment.
“Unit economies for large companies are positive-and some companies use millions of times a day,” he said.
The idea that has it should not Cost of hundreds of millions just to exist seems new today, but if you can make this game, it can simply reproduce.