Sep 10, 2024

Too Long; Didn't Read You.com announced a $50 million Series B funding round, as well as a new team plan called Multiplayer AI. You.com was founded in 2020 by Richard Socher, one of the leading NLP (Natural Language Processing) researchers in the world. The company’s ARR has grown by 500% since January 2024. You.com showcases the state of AI today The story of you.com is multi-faceted and telling in many ways. You.com was founded in 2020 by Richard Socher, one of the leading NLP (Natural Language Processing) researchers in the world, to offer a better search experience to users and compete with Google. With a startup exit and a Chief Data Scientist stint at Salesforce, Socher got the experience, network, and backing he needed to pursue his long-term ambition of taking on Google. That’s something few people have tried, with moderate success. Socher had diagnosed early enough that the way to success is by carving a niche for you.com. You.com focuses on serving knowledge workers in “complex informational / action searches”: elaborate queries and queries that are really about accomplishing a task, respectively. In 2022, in the pre-ChatGPT era, Socher set out a course for you.com based on AI, apps, privacy, and personalization . In 2024, you.com is staying the course, but a few things have changed. In the GenAI era the competition is growing, and borrowing pages from you.com’s book. Language model providers such as OpenAI and Anthropic now offer services similar to you.com. Upstarts such as perplexity.ai have sprung up, and Google itself is embracing the AI approach to search. You.com is making progress too. Since launching in November 2021, you.com has served 1 billion queries and has millions of active users, including from Fortune 500. The company’s ARR has grown by 500% since January 2024. Today, you.com announced a $50 million Series B funding round, as well as a new team plan called Multiplayer AI. We caught up with Socher, talked about the news, and took you.com for a spin. You.com’s pivot and the funding to back it up You.com’s $50M Series B is led by Georgian, with participation from SBVA (formerly Softbank Ventures Asia), Salesforce Ventures, NVIDIA, DuckDuckGo, Day One Ventures, and others, bringing its total funding to $99 million. Socher praised you.com’s investors, calling out Georgian’s deep technical expertise. Socher said that NVIDIA appreciated the fact that you.com has both innovation and revenue to show for, and hinted at future collaboration. What may seem peculiar at first, however, is the fact that DuckDuckGo is also investing in you.com. Aren’t you.com and DuckDuckGo competitors, after all? As Socher reiterated, you.com is not in the search engine market anymore. “A marketing campaign or deep research for biotech or a hedge fund is where we can provide a 10X better experience. Why? Because we have deep search expertise and deep Language Model expertise. So we’re very excited to bring that technology to other companies in the business context”, Socher said. You.com has two lines of business for enterprises, Socher added. One is enterprise site licenses, where an entire company can make all employees more productive. You.com offers a consumption-based pricing model, as well as APIs. This is what got DuckDuckGo interested, Socher related, as they share values around privacy. You.com goes enterprise: collaboration and security In a way, you.com is moving from a “bring your own AI” approach, selling directly to knowledge workers, to targeting enterprises. That requires a new array of features as well as an upgrade in the sales process, and Socher says both are there. The competition is there too, with Anthropic being the last one to check that box . Feature-wise, security, and collaboration are at the heart of the new you.com. You.com now offers single sign-on and is getting its SOC2 compliance. You.com also offers collaborative workspaces and massive context windows, enabling custom agents built by users to know everything they need to know to be effective. Users in those collaborative workspaces will be able to share chats, upload and share files, and work together on building workflows. Sales-wise, Socher is leveraging his Salesforce network and experience to build an enterprise sales team which he said is hard at work already. What they have found, Socher said, is that many companies had tried to build their own prototypes, big ones too. Some don’t have the resources, and you.com can help them build the first version. Some larger customers had actually built prototypes similar to you.com, or they tested out the competition. Reportedly, however, they were dismayed by how inaccurate that is. “In GenAI, you can quickly hack up a prototype, but it’s very hard to build something that really works at scale every day very accurately, has strong search, strong Language Model orchestration, and puts it all together in a package that you can rely on in your day to day work” Socher said. Smart, Genius, Research, and Custom agents You.com lets users create custom AI agents on top of any AI model for any task. It offers three different models or agents, as well as the option to create custom ones. Out of the box, you.com offers the Smart, Genius and Research agents. Smart is meant to provide quick, smart, and accurate answers. It won’t go into too much detail or give you a huge research report, but it will be fast and multimodal, Socher promised. Genius is meant to solve complex problems with a conversational prompt. The idea is that users don’t have to input in the perfect equation or formula. They can just explain what they want to accomplish in natural language, and Genius will use Python code and chain-of-thought reasoning to solve the problem precisely. Research is meant to provide in-depth research in minutes. It executes multiple searches from one prompt, analyzes dozens of sources, and delivers a report that you.com claims has more citations than any other LLM while being the only experience with citations that link directly to the exact sentence it cites. Users can also create Custom agents for any task using models from OpenAI, Anthropic, Meta, Google, and more. For example, a marketing team can train the AI to follow a brand’s voice and style guide and work collaboratively with the AI Agent without manual prompting each time. Customer choice, mixture of experts and prompt engineering Socher sees helping people and businesses to be future-proof as a part of you.com’s mission. As he noted, many customers feel like they’re locked in when they sign a big annual contract with one of the foundation model providers, and then a few weeks later, another better model comes out. Custom agents mix and match 3rd party models as well as models developed in-house by you.com. While many power users take a Mixture of Expert approaches to accomplish their various tasks, automating and engineering this to the scale needed for a service such as you.com presents many challenges. Besides offering customers choice, you.com also benefits from orchestrating different models and choosing the one most appropriate to provide accurate answers for different types of questions. For instance, Socher said, Anthropic’s Claude currently outperforms OpenAI when it comes to programming. So, programming questions are now routed to Claude. Another challenge in managing AI model workflows has to do with managing prompt engineering efforts. Oftentimes, users find that the prompts they have carefully crafted and fine-tuned stop working well (or even at all) as models evolve. Socher acknowledged the challenge and related that part of you.com’s orchestration layer is actually to do what is called dynamic prompting. For every query that comes in, you.com needs to understand whether users want to ask a factual question or whether they want to have the model actually hallucinate something. Doing research for a meeting with a client does not have the same requirements as writing a poem. The former requires accurate citations, while the latter does not. Dynamic prompting allows the model to decide to look for facts on the Web and then feed those into the prompt. Accuracy and the state of AI today One recurring theme in the conversation with Socher has been accuracy. Indeed, lack of accuracy is something that all language models are not known to suffer from. Different approaches are being applied to tackle this, such as RAG and context windows. Since accuracy is central for you.com, we wondered how it goes about improving and measuring it. Socher knows that if each step of an AI agent is 95% accurate, none of the 30-step workflows will work reliably. He believes that going from 95% to 99.9% accuracy is similar to the last-mile problem in self-driving cars. That is, it’s easy to hack up a prototype, but making it work reliably at scale is hard. There are several dozen modules used towards that goal on you.com. These include modules, such as intent classification and query rewriting, that are research directions in their own right. Socher acknowledged that making search accurate is a nontrivial engineering task, but he cited investors who trust you.com as well as third-party benchmarks that show you.com to have unparalleled accuracy. That’s all fine and well, but how about a little test? During our conversation, Socher showcased some examples of the Genius and Research agents at work, and they both seemed to handle complex tasks successfully. However, when trying out a simple use case of our own, things did not go as well. What we wanted to achieve was to identify whether a quote attributed to Umberto Eco (“There is no News in August”) is real and to retrieve a citation for it. A basic prompt was formulated, instructing the agent to act as an experienced journalist. We tested the exact same prompt on you.com and perplexity.ai. The results showcased the state of AI today. You.com and perplexity.ai each gave an acceptable answer, but one which should not be trusted as-is. Sources had to be checked manually one by one, and doing so showed that both you.com and perplexity.ai tripped up. Both were right as to the conclusion (Umberto Eco seems to have said that indeed), but all but one of the sources they used as references were either irrelevant or wrongly interpreted, or both. Did this save time? Did it produce a better result than an average knowledge worker would do without resorting to AI-powered agents? Will said agents continue to improve to the point where the answers to these questions will no longer be hard? We’ll leave that to your judgment. From our side, we acknowledge the effort and innovation that Socher and you.com bring to the table and wish them luck.