Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More
It’s not every day you get to sit down with one of the world’s leading researchers in artificial intelligence (specifically, the fourth most cited in the field of natural language processing) for an hour at his private ranch hidden in the Portola Valley near San Francisco. I was lucky enough to do so last week, and you can read all about it in our story about the launch of YouChat 2.0.
But there was a lot we talked about that didn’t make it into the article. So below are some outtakes from our interview, including You.com cofounder and CEO Richard Socher’s deep analysis of the looming AI search wars.
Socher also talked about how Google is facing the classic innovator’s dilemma; how he imagines conversational AI being monetized; and what he sees as the future of search. (Editor’s note: This interview has been edited for length and clarity.)
VentureBeat: Let’s start with your take on the Google versus Microsoft news last week. The Google presentation was a little underwhelming in my view, but given its history with AI, shouldn’t Google be on equal footing with OpenAI/Microsoft?
Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.
Richard Socher: Google has done incredible research and propelled the field forward in many dimensions in terms of its research, but they are making $150 billion a year by invading your privacy and showing you ads on a search results page. So when you develop technology that would work great if, instead of six ads you had just one but it means you’d lose $500 million a day, you’re probably not incentivized to launch that new technology into the world.
It’s the classic innovator’s dilemma. Google is very good at running things not profitably for a long time, like Google Maps and YouTube, and then pulling ads into it. The problem is that this particular technology [conversational AI] changes their core.
VB: So does Microsoft Bing have a real chance of unseating Google’s search?
Socher: Microsoft also makes a lot of money with ads, but much less than Google. So there is a little bit of that revenue that will pull you into the old world. But they clearly have much less to lose. They’re making money in a lot of other areas in order to be able to lose money on this for quite some time. And that puts them in a better position, and certainly they have announced [they will] soon launch something that we at You.com have launched last year already.
We launched a text-only chat interface that’s similar to ChatGPT last year, but YouChat had recent sources and citations. That’s what we launched in December. That’s what Microsoft announced that they want to launch in the future.
This week, we are launching something new that we haven’t told the world about yet, so no one could copy it, and that is incorporating the apps into that chat experience. Because when you ask for something like, “What’s the stock of the biggest CRM company?,” you don’t want to have a bunch of text answers potentially hallucinated by the AI. You want the ground truth stock ticker, and that’s what we’re showing you for that query and lots of other kinds of similar ideas.
VB: In your opinion, what developments from Bing or OpenAI are worth keeping an eye on?
Socher: OpenAI has done a bunch of really interesting things, including image generation. They worked on playing games too. DoTA was interesting. They had some robotic hand stuff. I mean, they’ve done a lot of interesting research. With ChatGPT, they really put forward and allowed a lot of people to understand the power of large language models — and that was the most interesting thing they’ve built so far.
VB: So the impact of OpenAI’s work has been more about getting the conversation started and sparking interest in AI?
Socher: Exactly. It’s allowing people who weren’t following AI and generative models and large language models to get very excited. We have to thank them for that.
We’ve launched various similar features. We had large language models within the search results page that will write an essay for you, like [if you search] how to write well or how to write this essay [the answer] will just be right there and you can write the essay. But thanks to ChatGPT, a lot of people realized that you can make search better. And somehow, the cleverness and combination of the amazing technology and also amazing marketing unlocked the search engine space — and we predicted this would happen. That’s why we started You.com in 2020. But, you know, it really pushed a lot of people over a threshold to be like, oh yeah, I could actually imagine not using Google anymore.
VB: How do you imagine monetization working with conversational AI search? You can take share away from Google, but can chat answers monetize as highly as traditional AdWords?
Socher: Great question. I think you can have advertisements within chats. In fact, there’s likely going to be some ad-free chatbots, and there’s going to be some other chatbots where you might not know if the answer they gave you was actually paid for. Or maybe there’s some little ad symbol in the text. That’s one potential future.
It’s not something we’re working on — just for the record — but you know, it’s obvious how you could do it right and you could push harder on that front. You can also just have simple private ads next to the chat, similar to what DuckDuckGo might do, like an ad that only depends on the query. You see the ad and that’s it. It’s not that hard to think of just having an ad next to the chat based on your query.
Now, I do think that the massive level at which Google monetized where, you know, it’s sometimes now six ads on that first page. And then you scroll down, it’s just SEO’d microsites, which are mostly there for you to click on an ad from them that also often Google gets money for. I think that future — I hope — is over, because it sucked, and people didn’t like it and it gave users bad experiences.
I think a lot of these larger search engine companies learned a really painful lesson, which is that the more useful your actual organic content is, the less likely people are to click on an ad. And the more relevant and interesting and useful the ad is, you know, and then the content is not that good, you get even more ad clicks. So there was like, for the last decade, Google being the de facto monopoly and you getting a default that’s always Google on your iPhone and Android and Samsung and basically everywhere.
The fact that Google pays $15 billion a year to Apple to be that default search engine tells you how powerful that is. Now, in the European Union already there are various ways to fix that and hopefully You.com will take part and you can actually choose You.com, so at least there’s some randomization in the default choice. It just makes it easier to actually be added to that list of defaults, which is currently also very hard.
And with every browser company just to be added to the list — not even to be the default, but just make it a choice that you can go in your configurations, which very few people do anyway, just to be in that list of choices — every browser company charges you money for. So it’s really tough. I do hope and think that model will be slightly less monetizable. I don’t think users want to see six ads and then a bunch of SEO’d microsites with ads. I don’t think that’s the optimal experience for the internet. So in that sense, I think it’ll monetize less well in the future. I think you can only monetize that well when you’re a monopoly and users don’t know that there is any other choice than to keep scrolling down until you see relevant content.
VB: Okay, but there have been threats to Google in the past. Social graph was supposed to replace search, Amazon was going to replace product search, Reddit provided better content than Google. TikTok search is growing. Is this time any different, or will Google be able to integrate AI and retain market share?
Socher: I think actually in each of these cases, they actually did carve off a little bit of the search engine space. If you want to search in order to buy something between $5 to $50, you would probably just go search on Amazon, right? So those queries have actually disappeared for Google fully.
But most of these — Reddit, TikTok, ChatGPT — they all show us that users are looking for something different, but none of them are a viable, full daily driver search engine alternative. That’s why at You.com, we keep in mind that exact issue every day. As we innovate, we say let’s not innovate in a way that it’s just a tiny subset of what you could do in a search engine. It needs to capture, and this is very hard, both in design and everything, but it needs to be able to be your default search engine and your actual default in your URL nav bar.
So you can get the Chrome extension for instance, or install our iPhone app or Android app, so that you can use it every day and you can still find the weather and you can find the stocks and you can find driving directions and all of these little things. They still need to be viable to be used. That’s why we have a TikTok app. We have a Reddit app. We have generative AI apps. We’re a chat. But we also have all these other capabilities that you would want from a search engine at You.com.
VB: It seems to me like these AI assistants are becoming kind of a commodity, with each major tech company controlling a large language model and building an AI assistant to work in tow with you. Do you see it the same way, that users will have to choose between Siri, Alexa, Google, etc.?
Socher: A hundred percent. I think often when you ask something like, “Can you generate an image with AI?” you don’t want to get a bunch of links in a text message saying, “Yes, that’s totally possible. You can create an account here.” You just wanna do it, and that’s what we’re starting to do with You.com. That’s a big part of why we have this open platform. Because also, if you think about it, every publisher in the world, every company in the world, what are they going to do when the chatbots just give you answers, and don’t necessarily forward traffic — to publishers and magazines and everyone. The answer is, hopefully You.com wins and publishers can have an app right there. And if users like it, it will keep coming up in the chat response, and that’s what we’re launching with YouChat 2.0.
VB: The cost to serve these in conversational AI results is like 10 to 100 times higher than search.
Socher: It’s actually more like 50% to 70% higher at most, for us.
VB: How does that impact your approach to growth?
Socher: It’s interesting because, if we grew to a hundred million users, we would have to raise a lot more money like OpenAI did in order to keep [YouChat] alive at all. And not every venture firm out there is on board with that massive growth and supporting it before you have financials figured out. So that’s one part of the answer. But overall, the cost of service is actually not 100 times higher. It’s about 50 to 70% higher, and it’s doable.
It’s a repeating pattern that I feel like has happened so often in technology that I’m surprised people bring it up and, maybe it makes sense to bring it up, but, you know, most every medication was very expensive in the beginning, sequencing your DNA used to cost lots of money in the beginning. Fancy phones in the beginning were very expensive. And you know, if there’s enough demand for scale, scale will push prices down, and there is certainly enough demand for scale in large language models. So demand will continue to increase and hence, prices will continue to go down and hence, cost of service will go down.
VB: Over the last few years, search query growth has slowed to single-digit growth. Do you think that the pie will grow with AI-enabled search?
Socher: I think if you expand what search can do, then yes. Search is such a crazy and interesting and tough market for many reasons. One is that a famous mantra of startups is to focus on a tiny niche, like only a college, then only the other Ivy League colleges, then only this city and then after that city, another city, and then the next city, and then you go to five more cities. So that is kind of a very well known strategy for startups — owning a tiny niche and then expanding in search.
In search, if you want to be the actual default search engine for someone, no matter if you say, oh, I only focus on developers, which we actually did last year. Developers are people too, I sometimes joke. They go outside. They want to know the weather, they trade stocks, they want to know stock prices, they want to know driving directions and so on.
So we had to build all of these capabilities to really be a viable default search engine for people. That was very hard, and you have to do all of it with incredible speed, because otherwise you’re dead. You know, if you have the coolest search engine, but it takes 10 seconds to load it, it’s not gonna work.
So, if you think about what people are hiring a search engine to do, then I think the market can be much broader. At You.com, we already have generative AI apps. That’s not really a search anymore. You’re actually creating content rather than searching for content, right? But when you’re searching for the question, can I create this image of a baby skydiving, the answer is it probably wouldn’t exist in the world, but we can make it for you, right?
VB: What inspired you to create an AI search engine that takes on Google in the first place?
Socher: Lots of different strains of reasoning and thoughts went into this. One is, personally, I’ve worked in natural language processing for over a decade. I’m actually the fourth most cited person in the entire field of AI natural language process. I thought about what’s the biggest impact application that I could build now that we’ve figured out a lot of fundamentals. I kept coming back to search for almost a decade now, nine years. Nine years ago, I built the first prototype of a search engine, and I was just like, I couldn’t quite let the thought off.
VB: I understand you built the fundamentals of natural language processing through some of your research and professional history. But then how do you land on search? Is there a lightbulb moment?
Socher: It’s the biggest application of natural language processing ever. It’s sort of, from an AI perspective, it’s a big one. But then you also look at the entirety of the economy: The entire economy is moving online, and at the beginning of most people’s online journey, you start with a search engine, and then in natural language, you tell that search engine what you want to do, and then it’s kind of absurd that every other company out there has to pay a tax in order to exist on that first page through an ad.
It’s kind of crazy that you have that single gatekeeper in the beginning of so many people’s online journey that just wants to sell you a priority to the highest bidding advertiser. And that’s what they’ve been focused on for the last decade. That just can’t be the right setup, right? The technology gets so much better and the user experience keeps getting worse. It’s just absurd.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.