Microsoft: “With ChatGPT you are not alone in the face of a problem”

Microsoft could turn the hype around ChatGPT use it to find out with the in-house search engine Bing suddenly emerge as a competitor to Google. Even if the AI ​​search still has problems, the demand for the new “Copilot” large. We spoke to Harald LeitenmullerCTO of Microsoft Austria, on the strengths and weaknesses of search and how the system becomes profitable.

futurezone: After years with Bing, Microsoft has managed to position itself in the public eye as an alternative to Google. What does it take to compete with the market leader?
Harald Leitenmuller: There hasn’t been any real innovation in searching for a long time. Users need to know what terms to enter, and that’s not really satisfying. On average, people enter 3 search terms and hope that something will come up. This limits the context, which has a massive impact on relevance. We have the combination of artificial intelligence and a relatively good search engine. Together with the supercomputer that we developed with OpenAI, the results become up to 20 percent more relevant and that is of course an innovation boost. Of course, the hype about AI also contributes to marketing.

What do you use Bing Chat for?

I don’t usually use this for details and facts, but for brainstorming or creative processes. It’s just fun because you get the feeling that you’re not alone in the face of a problem. An example is interviews. You usually get a list of topics and I can ask the chat to generate possible questions and answers about the topics. There are things that are obvious, but also things that I wouldn’t have thought of. Then I let the chat find weaknesses in its own answers. This creates an iterative process and I have a kind of portfolio of topics to work with.

I also teach responsible use of artificial intelligence. Last year I asked which exam questions I could ask the students, based on my agenda. Then I asked: What would the students ask about? I discussed this with the students. They in turn tried to guess which questions could come up. Now a competition has emerged as to who can predict which questions.

How exactly does artificial intelligence know what a person is really looking for?

On the one hand, the ChatGPT language model has learned which intentions are conveyed via language. The Bing chat derives relevant search queries from this intention. You can see which terms the AI ​​is looking for. Behind it is a classic search engine. These results are then summarized in order to work out the important aspects that are relevant for users.

Now our market companions have tried to enrich the context with advertising. This is how you create a profile of people and over time a context for the person develops. Today, however, that is no longer ideal. We believe there will be big changes, also in terms of advertising models. It has to be honestly said that users are not always happy with the fact that a lot of things are now advertising-driven. Better solutions have to be found.

Will there be advertising in Bing chat?

There is currently no advertising directly integrated into the chat. The ads that appear now come from the search results. Bing will then mark it as advertising.

The AI ​​search does not yet work completely without errors. Where is the problem?

With experience and the way you interact with the AI, this could be even better. The competence of how to conduct the dialogue with the AI ​​massively influences the result. If a manager simply says to his employees: “Do something smart”, he shouldn’t be surprised if nothing comes of it. It’s a good idea to give clear instructions, and that’s the same with Bing Chat.

The search has problems, especially with very current data and facts. Why is that?

The trained ChatGPT model is not up to date. It is there to understand language. The results that come up are the ones that are searchable over the internet given the information. These may not always be the most up to date queries, this is something we are still working on. That’s where we come into the commercial area.

How is this related?

The focus is not on advertising, but on the connection of data sources. It’s about signing contracts with newspapers and other companies in order to always be able to deliver the most up-to-date information. The business model is not final yet, but we want to create a fair market so that everyone who contributes their information today can also benefit from it.

So Microsoft will also work with news platforms there?

This is already happening on MSN. Integrating the data into ChatGPT to make it work well is one of the most important next steps. The data that we either buy or that come from media partnerships flows in there. We don’t have an endless number of current sources, that’s all subject to a fee. There is now a marketplace.

It was said that the AI ​​had to be “restrained” after it made questionable statements. What’s behind it?

It’s interesting how much energy and effort people put into testing the limits of a system. On the one hand, this is good because we are in a test phase and can make improvements. But you also have to deal with it responsibly. As with finding security vulnerabilities in software, it is common practice to first notify manufacturers of the vulnerabilities before publishing them on the Internet. I would also wish that we were given the opportunity to improve what we did. We’re in the first phase and when the context in the chat gets too big and people push too much, then things happen that you might not want. We’ve limited the amount of dialogue so the dialogue doesn’t get too overwhelming.

What is ChatGPT for and what not?

It should be seen as a tool, as a “co-pilot”. Like a colleague, a partner you can talk to for support. This is a creative process in which you can definitely learn something new. We write labels, i.e. intended uses, for our products. We try to make that even clearer. It is not intended for creating creative works or writing code. Bing is good for searching. There is special software for programming or creative writing, for example to become a writer without committing copyright infringement.

This is a very expensive product with a lot of effort. How does this become profitable?

If we look at the OpenAI growth rates, it had a million users in the first 2 months. The profitable aspect is that people are willing to pay monthly for this added value. The amount of requests for the Azure OpenAI Services is so high that people have to be really patient as we unlock the resources. Making that available globally is now a focus. But of course there are many other requirements, such as data protection and control.

What can companies do with the OpenAI services?

There are partner companies building on our Azure cloud services that develop solutions with OpenAI. An example is to develop legal databases. This allows me to have contracts checked, for example. In order to develop this, you need specialist knowledge and partners who have specialized in it take care of that.