Richard Socher: "We'll never be as bad as Google. We'll never sell your data."

salesforce.com video

Are you happy with Google search? Regardless of how you answer this question, chances are you still use it. With the notable exceptions of China and Russia, where Baidu and Yandex lead, respectively, Google's market share in search is over 90% worldwide.

It's not that Google is the only game in town. Besides Baidu and Yandex, the likes of Microsoft and Yahoo have tried their luck too, with Bing and the eponymous search engine, respectively. The privacy-focused DuckDuckGo is another option. Yet, none of those has a market share of over 3% worldwide. Can a new entry do better than so many others before it?

Richard Socher thinks so. Socher, the founder and CEO of upstart search engine you.com, has had this mission impossible on his mind ever since his Stanford days. Today, almost a decade later, with lots of distinctions and plenty of startup and enterprise experience under his belt, Socher is heads-down on mission impossible.

The birth of you.com

When Socher came to the US from Europe in his twenties, his dream was to get a university faculty job and he worked very hard to make it come true. He got into deep learning early on, when it was just a niche topic, and worked with deep learning pioneers Andrew Ng and Chris Manning at Stanford.

Recommends

The best browsers for privacy

Most people use Google Chrome as their default browser. But privacy is another matter for the online ad giant.

Read now

After having won the best computer science thesis award for his Ph.D. on Recursive Deep Learning for Natural Language Processing (NLP) and Computer Vision, Socher thought that founding a startup would be just a detour on the way to academia. Life proved him wrong.

Socher described his first startup, MetaMind, as "an enterprise AI platform that worked in medical imaging and eCommerce images and NLP and a bunch of other things, a horizontal platform play as a machine learning tool for developers." If that sounds interesting today, it was probably ahead of its time in 2014.

Salesforce acquired MetaMind in 2016, and Socher became the Chief Data Scientist at Salesforce. He led more than 100 researchers and many hundreds of engineers, working on applications that were deployed at Salesforce scale and impact. Socher was instrumental in creating Salesforce Einstein, a wide-ranging initiative to inject AI capabilities into Salesforce's platform.

In 2020, Socher left Salesforce to pursue his longtime ambition of building a search engine, which he named you.com. You.com has raised about$20 million from a number of investors, including Salesforce co-founder, chairman, and co-CEO Mark Benioff. The first version was implemented by Socher at the end of his Ph.D. but he was initially hesitant to pursue this.

"At the time, I thought, man, it's just too ambitious. People were probably like, Google's going to sue me. All my smart friends are going to work at Google. It's going to be so hard to compete with them. No one's really complaining about Google very much in my circles and online. And so I kind of discarded the idea," Socher said.

Socher claimed that he's not into this for a quick acquisition, and added that he and the small team at you.com are very motivated, and have the runway to work on this for many years. Socher acknowledged that this will in fact take many years, and gave three different groups of reasons for taking on Google: User-specific, macro, and timing.

What's wrong with Google?

Many of the user-specific reasons Socher cited have to do with privacy. Most online journeys start with a simple search, and the fact that our privacy gets so massively invaded at almost every step we take online as our lives go more and more online is unfortunate, he said. However, he added, users are becoming aware of it, and that's a good thing.

Ads are also part of Socher's user-specific reasons. As a user, it's just annoying to see five, seven different ads before you see some content, Socher said. Plus, once you learn a bit about how content ranking works, you realize all these search engine optimized (SEO) microsites are also just ads trying to funnel Google into affiliate links and cookies, he added.

Then, there is the issue of control. "A lot of people think about their food diet, but I think our information diet is incredibly important, too. It's important to be able to [..] say, I want to see more Reddit or less Reddit, or I want to see New York Times or ZDNet and others, versus just being sold with your information desires to the highest-bidding advertiser and having no control over it," Socher said.

Socher's macro reasons mostly come down to the fact that "the entire economy is moving online, and having a single gatekeeper that wants to sell you to the highest advertiser is not an ideal setup for the web, period," as he put it. 

Google has always maintained that Google Ads and organic ranking are entirely independent. Socher questioned the validity of this claim, although we were not able to verify this independently. Socher commented that "it's like a bad movie, and it's kind of nuts that it's happening." On the bright side, he added, now there is some tailwind in terms of antitrust and realizing the issues at stake for the entire economy".  

You.com is Richard Socher's bet to take on Google search

Somewhere in between macro and timing would be what we might call the information deluge. Twenty years ago, it was amazing to be able to have access to information. Today, accessing information is table stakes, and the problem is how to deal with it all, Socher noted. His answer: "You need to have AI that summarizes it for you".

Socher strongly believes that now is the time to innovate in search, as there hasn't really been that much innovation in recent years. Initially, Google provided an insane amount of value, but now it's logarithmically flattened off, Socher said. The data that people provide to Google was not very valuable initially, but now we're reaching an inflection point where people's data becomes more valuable than the services they get from Google, he added.

It could be argued that over time Google has added AI to power its search as well, notably by using BERT, one of the Large Language Models (LLMs) pioneered by Google. However, Socher did not hold back on his critique, noting that the only way to get "something real" out of Google search is to instruct it to get results from sites like Reddit explicitly every time and that Google's idea of innovation seems to come down to adding an ever-growing list of ads to its results to increase sales.

Taking on Google with AI, apps, privacy, and personalization

There is a certain grounding in Socher's critique of Google. However, it's a well-known fact to anyone even remotely familiar with search engines that Google has built a very effective moat around its business by creating what is arguably the most comprehensive and efficient index of the web.

Plus, by now Google is so entrenched in the routine of billions of people around the world, and the default for most browser search options, that to make users switch, as one Yandex executive once told ZDNet, you have to be 10X better. Is that even possible for anyone, let alone an upstart like you.com? How do you go about that?

Socher's reply to this obvious question was based on the fact that not all queries are the same. Sometimes, he said, people just want to get factual information, such as the weather today, or the leader of an organization. Sometimes, they want to get to a specific site, and instead of typing it, they enter it in a search.

For those types of queries (quick information queries and navigation queries, respectively) all you can do is serve them as quickly as possible. There is no room for differentiation. Where things get interesting is in what Socher called "complex informational / action searches" or elaborate queries, and queries that are really about accomplishing a task, respectively.

Socher claimed that you.com already does better than Google in complex informational searches because it provides much more rich information. As for action searches, such as ordering takeaway or booking a flight, Socher made it clear that this is the goal for you.com. He referred to you.com apps, which are domain-specific modules that are fine-tuned to the needs of specific tasks/audiences.

One domain that you.com is targeting is coding and developer searches. Socher offered the example of a developer looking for how to train a model using PyTorch. You.com can help in a number of ways. There is a Stack Overflow app, there are code snippets, there is access to documentation, Reddit discussions, and even a code-generating app, Socher said.

These are all things that Google does not offer, they come with a copy-paste button, and they provide great value by helping developers save anywhere between 30 seconds and 30 minutes for each search, Socher claimed. There's "a ton of AI and NLP in there," he added.

Domain-specific search applications are how you.com aims to deliver 10X better results than Google. Developers are one of the key audiences

The same goes for things such as product reviews, which aggregate and summarize information from different sources, rather than having to open a multitude of tabs. That is 10X better, according to Socher. He also referred to how you.com works with content providers such as Stack Overflow for its apps, alluding to announcements with more details on "building an ecosystem" coming soon.

Socher also talked about you.com's business model and its stance on privacy. He is confident that you.com apps will provide value that enough people will be willing to pay for. Another feature that Socher believes adds value is personalization -- the ability for users to customize results according to their preferences.

Clearly, in order for this to happen, user profiles must be supported. That opens the door to discussions around data collection, privacy, advertisement revenue, and related policies. At this point, Socher sees advertisements as secondary revenue streams and takes a middle-ground approach to privacy. You.com offers a private mode, and Socher promises better privacy: "We'll never be as bad as Google. We'll never sell your data".

However, he also believes that if you make privacy your focal point, then "the hardcore privacy people at that point want you to be a fully encrypted, fully open source, no revenue, no data, nothing kind of project. Essentially, you can't really be a company, [..] you will never be able to compete with Google." You.com will use data from logged-in users to serve localized results, which Socher believes is something most users want.

Ultimately, however, the choice between privacy and convenience will be up to the users. As for where the information is coming from: some of it, for generic queries, comes from Bing's index. For domain-specific queries, you.com has its own indices. This is a dependency all search engines except Google and Bing have, Socher said, although some like DuckDuckGo are "just a thin wrapper around Bing".

The way forward

It's still early days for you.com, so the verdict on whether this can work is still out. Besides "lots of love on Twitter and other channels," which Socher referred to as an encouraging sign, there are more solid reasons for optimism too.

Socher does have a well-rounded analysis of Google's weaknesses, and the background, motivation, and backing to at least give this a shot. The approach you.com is taking, although not fully operational or unveiled yet, seems promising. You.com was recently included in CB Insights' AI 100 list of the most promising artificial intelligence startups of 2022.

You.com's founder does not seem to have any illusions about the fact that this is going to be an uphill battle. Getting users to adopt a pay-to-use model for search, beating Google at its own game of powering search with AI, and walking the fine line between keeping users happy and running a viable business are all big bets for you.com. If nothing else, however, some competition in the stagnant search market would probably be good for everyone.

What Socher identified as a key premise for you.com is the idea of making AI controllable by the people affected by it. For you.com, that translates to users being able to nudge the search engine as to what they'd like to see more or less of. As for the bigger picture in AI, Socher seems to have been spot on in his 2017 TED Talk in which he identified NLP and multi-modal AI as key directions for the future.

Socher believes that LLMs are already "doing amazing things", and is hopeful that more progress will be made in terms of multitask learning, enabling them to be better at more tasks. However, he also believes that eventually LLMs will need to be injected with certain rules, or enabled to learn them, as scaling up does not seem able to achieve this.

In terms of moving AI forward, Socher also noted that current hardware favors a specific type of AI model architecture, which relies on matrix multiplications. That may or may not be the way forward, but this "hardware bias" has sidelined alternative model architectures. It's a bit like looking for your keys under the lamppost, Socher noted.

Socher is naturally aware of all major AI talking points these days, including bias (it's not just the datasets), sustainability (perhaps overblown, but we can and should do better), ethics (no easy answers, it depends on each person's stances and beliefs), and more. It's a conversation worth exploring -- perhaps even more so if you.com ends up working out.

Artificial Intelligence

The impact of artificial intelligence on software development? Still unclearAndroid 14's AI-generated wallpapers are super fun. Here's how to create themAI aims to predict and fix developer coding errors before disaster strikesGenerative AI is everything, everywhere, all at once

• The impact of artificial intelligence on software development? Still unclear
• Android 14's AI-generated wallpapers are super fun. Here's how to create them
• AI aims to predict and fix developer coding errors before disaster strikes
• Generative AI is everything, everywhere, all at once