Munich IT professor challenges the US tech giants

Along with Midjourney and Dall-E 2, Stable Diffusion is now one of the most widely used image generators in the world. Dall-E 2 was launched by Californian company OpenAI, which also developed and uses the ChatGTP chatbot, among others. Microsoft sponsored. Midjourney comes from the American company of the same name, about which little is known. It describes itself as an independent research firm with no investors. According to its founder David Holz, it is now profitable.

Read also about this

Ohmer and his team didn’t have supercomputers and didn’t have the billions to train their AI algorithms. In order to save, they rely on smaller but particularly smart systems. Still, money was crucial to success. Finally, Stability AI has supported the startup financially and enabled it to transform its AI model into one of the most powerful in the world.

Usually, generative programs need a great deal of computing power. Therefore, it only works online: the actual computing operations run in server farms and high-performance computers – mostly in United States of America. This differs with Stable Diffusion, because the program manages with just a few gigabytes. The reason for this is that the creator of the images ignores details that are barely perceptible; Like the shape of individual blades of grass in a meadow. Such little things require a lot of computing power.

Bjorn Ohmer put the code online for free

What’s revolutionary is that Omer made the programming code for the program freely available on the Internet in August 2022 – as open source is called. This means that users do not have to pay for it and they can also develop the code further and try it out.

Bjorn Omer He asserts that he invented the image generator with the research group he heads – not alone. And that he and his team initially just passed the code on to the scientists to give them a head start. The Munich research group later made the code available to the general public. Free. So Omer is not worth any of it.

then he I am convinced that research should be accessible to all. When it comes to generative AI, he’s talking about a “significant technology” — both economically and socially. “The way we did it was probably the best solution to start a timely discourse about what society does want in generative AI — and what it doesn’t want,” he says.

Software that costs money can be manipulated just as easily as open source

According to Ommer, the decision to make the programming code for the image generator freely available on the Internet was also criticized. If the software is freely available online, it can be misused more quickly, for example to spread fake news, as has always been said. Ohmer says he and his team knew that despite some security locks, the code was still immature when they released it.

But he thinks it’s a fallacy to think that software that costs money is less likely to be manipulated. “Not making the software freely available and selling it as a security feature doesn’t work. After all, it can also be leaked or reprogrammed,” he says. This is exactly what has happened in the past few months. For example, the code for the AI ​​chatbot LLaMA was leaked only a few days later and was made freely available for download on the Internet.

Björn Ommer considers it so difficult that so far only a few companies are able to train, develop, and systematize generative algorithms. In his view, Germany and a large number of other countries make themselves completely dependent on powerful American corporations. Also in the public discussion about AI and its regulation, the demands of European or even German AI models are becoming increasingly urgent.

“I am amazed that the EU has simply allowed tech companies to do what they want,” German Ethics Council President Alina Boeks said in an interview with our editors. Something needs to be done about the fact that the EU relies on a few digital companies, all of which come from China and the USA. The German Ethics Council considers that European solutions and alternatives are urgently needed. “And if we don’t start now, the train has completely left the station.”

Ohmer opposes corporate America’s monopoly of power

Companies like Microsoft, Apple or Google can afford to feed huge amounts of data to algorithms on thousands of computers. But what data do they use to train their systems? They are careful not to reveal it. Perhaps the fear of getting into trouble and sharing the profits with others is paramount. For example, when they’re violating copyright because they’re training their systems with images from artists. Ohmer believes that artists should be able to actively speak out against the machines that fuel the imagery of their work.

Eager to democratize artificial intelligence and not leave the technology to a few large corporations with massive capital and massive server resources, Bjorn Ohmer opposes the monopoly of power of US technology companies worth billions. Although he made fun of them at first. At just 28, he preferred a professorship in computer science in Heidelberg to Silicon Valley. His independence was more important to him than working with the tech giant. This also applies to his research.

Artificial intelligence has rapidly blossomed into a powerful tool in recent years. But what remains constant: Omer’s desire is to teach machines to see better, to explore their limitations, and to make artificial intelligence an ally. For the general public, not for tech companies. In Munich, in Europe – not only in Silicon Valley.