Interview with Richard Socher Google is stuck in the innovator's dilemma

Mr. Dot Socher, you have big plans for your search engine You Dot Com. You want to outperform the market leader Google and, in some ways, even revolutionize how people search the Internet. You . com is a new search engine that aims to make users' Internet searches as simple and as effective as possible before we get into the specifics.

Our better privacy is our first priority. Second, we are returning user control over the information consumers consume to them rather than to advertisers.

You can decide whether you want to see more Reddit, less Twitter, or more FAZ news in the search results overall, for instance. Third, we provide a variety of artificial intelligence-based functionalities.

For instance, we offer users the option to have a blog post written about Elon Musk or aliens using an AI function. This feature is available in the YouWrite app, which also writes essays, marketing copy, and other types of content. Alternatively, we have a programming AI.

We have an artificial intelligence called "YouImagine" that can produce images. Finally, we have a chat feature based on a very large, extremely complex AI that can carry on a conversation but, in contrast to other chat skills or chat algorithms, it has access to the internet and can also reference sources.

Backtracking to the search one more time. Three sub-settings, titled simply "Your time," "Your sources," and "Your privacy," are also available to anyone who visits You.com in addition to the search field.

The three functions can be used to set true privacy, among other things. Google does not currently allow users to conduct a simple private search.

It has already been proven that the "incognito mode" isn't actually incognito either, even legally. Since nothing is examined in our truly private mode, we as a search engine are actually completely ignorant of how users interact with it.

We want to make it simpler to post sources because a lot of people also want search to be able to identify their interests and provide them with results that are customized for them. Additionally, a lot of users occasionally only want a certain amount of information. For example, if you're looking for a recipe, you might only need to know the ingredients and not whether or by whom and when it was created; this will save you time.

Searching on You.com feels different from searching on Google because it makes us feel like she is much more American. Unfortunately, our attention is still primarily on the American market.

The internationalization of our website for additional languages and regions has not yet been completed. We have a few German news websites that we could potentially post as apps.

However, there are still a lot of functions that are lacking for the German market. That's a very good way to describe it, but if we combine the world of concepts you just described with a search engine, then the You . com project goes well beyond that - is it ultimately supposed to be something like an operating system for the Internet?.

To use a term that many people are familiar with, we continue to refer to this as a search engine, but in reality it is more akin to a portal to the Internet. Artificial intelligence was where she started her career.

You received your doctorate from Stanford after studying computer science in Leipzig and Saarbrücken. After going independent, you sold your business to the Salesforce group, where you spent many years as chief AI scientist, and you left the group to start your own business. You also concentrated on the issue of how a computer can comprehend and process language within the context of AI.

Although some people were more dubious, they thought deep learning techniques would also be effective. When I started in 2010, the field of deep learning and artificial neural networks was making a comeback, particularly in applications for image processing and image understanding, or "computer vision" as it is known in English.

Additionally, I had studied a great deal of linguistic computer science, and I thought that combining it with language processing would be the most exciting. The fundamental premise is that everything is ultimately just a list of numbers in language processing as well.

The numbers, however, now indicate a point in a semantic space rather than the values of individual pixels as in an image. For instance, the words "dog" and "cat" are closer together than "walk" and "run," which in turn are closer together than "walk" and "sleep.". There is only a list of numbers throughout the language.

Singular and plural nouns both have numerical lists as constituents. You can also transform words and sentences into neural networks once they have been reduced to a list of numbers or a vector.

After that, the neural network can pick up new information and adjust how this space is laid out. The teaching of computer language has been attempted in the past.

What is unique about deep learning? The approach of neural networks and deep learning is no longer to specify countless details, but rather to say using the example of language: That is a sentence that is positive. I've also included 10,000 encouraging words.

And the program needs to figure out for itself, whether it's because of the use of adjectives, negations, or something else, why this sentence is positive. The ability of the AI and these neural networks to learn how various words go together is continually improving.

It is no longer indicating whether this sentence is positive or negative or how it might be translated into another language for chat because we simply chose a different objective function. Instead, each text should have a prediction for the following word from the program.

This concept is very straightforward, but it works incredibly well. One needs to be completely knowledgeable about the world in order to predict the next word in any sentence they may find online.

Please use the following sentence as an example: I'm traveling from Berlin to dot. - It needs to understand geography in order to be able to foretell the following word.

Additionally, if you provide the neural network with sufficient parameters, it will eventually incorporate such general knowledge because it can actually remember everything. The fascinating aspect of language processing is how closely it is connected to knowledge, civilization, communication, and culture.

There are currently many powerful language AIs. Even bigger neural networks, more processing power, and more data are what's behind it, despite the fact that many people once believed scaling alone might not get you very far.

Given that it can now provide answers to more complex questions in a more complex manner, it is even possible to wonder whether such an AI will ultimately prove to be superior to current search engines. Yes, that is definitely very exciting.

These systems simply provide the user with the desired information by combining information from numerous websites. By the way, since a user could request that an AI write an essay about Kafka and the Beetle or something similar, this will also affect high schools.

Your chatbot not only knows information up to and including the year 2021, but also information from as recently as yesterday, which is what sets it apart from the ChatGPT that everyone is currently raving about. Additionally, he is capable of conducting his own Internet searches, finding fresh information from them, and citing the sources.

Apart from that, they are very similar. No, it's still challenging for users to completely rely on the information provided by the AIs, which is the central source question.

For commercialization, that ought to be necessary. Yes, that is what we also believe.

Now let's talk specifically about your business: Google has a huge physical infrastructure at its disposal thanks to its tens of thousands of employees, numerous data centers it runs all over the world, and undersea cables. All good questions, which regrettably we don't want to answer at the moment because we have a lot of competition that is closely following us, including how you intend to keep up, how many employees you currently have, and generally: what are you doing to counter that.

And I'm pretty sure there will be a ton of copycats looking to offer features similar to what we do in the next three to six months. Of course, in three to six months, we'll be further along. However, since that involves a bit of cat and mouse play, we are unable to and do not wish to disclose too many business specifics at this time.

Google is caught in a classic innovator's conundrum. Will they respond with their own models, which they also have at their disposal, or will they ask if they can buy your company in a phone call?. They make so much money from privacy and advertising that it would be very challenging for them to suggest that, rather than five advertising lists, they just show a cool app that provides the answer directly and is beneficial.

Because if that happened, no one would have to click on advertisements, the source of their primary income would disappear, and they would probably have to lay off a lot of people. The value of the stock would drop.

I believe Google is firmly entrenched in the advertising industry. Of course, Google possesses all of these technologies. They also employ top talent, and many of my friends work there and conduct extremely fascinating research.

The stakes, however, are a billion dollar company. On the other hand, as a startup, we still have the chance to simply be as innovative as we can for the users.

Currently, we don't even have a commercial business model, so what should your business model be in contrast to Google with its massively advertising-driven strategy?. We initially place a heavy emphasis on growth, much like Facebook, YouTube, Google, and Instagram, and like all successful online businesses.

We declared that we wouldn't engage in so-called targeted advertising, i. e.

undermines privacy with its advertising. But we are already considering the possibility of having private advertising, much like other private search engines.

Ads that are solely based on search terms are therefore displayed; we haven't been completely following the user for a few months. Because of this, the advertising should be more like a widely visible poster than something that is directed specifically at the individual.

However, we'll also make an effort to market our helpful search apps, like "YouWrite," and avoid relying solely on advertising.

To use a term that many people are familiar with, we continue to refer to this as a search engine, but in reality it is more akin to a portal to the Internet.