Start-up Aleph Alpha wants to compete with OpenAI

Dusseldorf It’s once again an American company starting a technological revolution. The start-up OpenAI, with the software company Microsoft as a partner, has developed a new generation of artificial intelligence that impresses both the public and experts with its language skills. The competitors are called Google and Baidu and come from the USA and China.

Jonas Andrulis wants to ensure that Europe also plays a role in this future field. With his start-up Aleph Alpha, the 41-year-old business IT specialist has built a large language model called Luminous. The technology is able to answer complex questions and compose texts, similar to ChatGPT.

Aleph Alpha primarily addresses business customers. “We don’t want to build a machine that you can have a nice chat with – but one that can support complex work steps with high safety requirements,” Andrulis formulates the vision. For example in healthcare, in the financial sector or in the judiciary.

One benefit of this strategy is that Aleph Alpha requires less capital than OpenAI, which is expected to spend millions of dollars each month on computing capacity alone. The German start-up has to invest a lot in view of the great competition. “Now the point has come where it will be decided whether Europe will create or buy artificial intelligence in the future,” says Andrulis.

Anyone who would like to try Aleph Alpha’s technology can find it on the portal of the city of Heidelberg, home of the start-up. The digital citizen assistant Lumi has been in use there since October 2022 – a robot with big eyes hovers up and down in the chat window. He briefly answers questions about day care centers, housing benefit or garbage disposal and links to further information.

Assistant to lawyers and doctors

The chatbot is still in a test phase, which citizens occasionally notice. But he makes Aleph Alpha’s vision clear: The company wants to fundamentally change the interaction between man and machine.

Funding is tough for many startups right now, but that’s not necessarily the case for generative AI. Aleph Alpha Founder Jonas Andrulis

The language model in the background is able to understand input in natural language, research suitable information from a database and output answers in simple sentences. He can also interpret images.

“It’s an enabling technology, like electricity,” says Andrulis, who spent three years at Apple in the Special Projects Group leading AI research, working on projects like self-driving cars and the voice assistant Siri. According to the IT specialist, information-based value-added processes could be rethought – with the potential for “a new industrial revolution”.

Aleph Alpha intends to focus on complex activities in areas such as medicine and law. On the one hand, accuracy is important there, on the other hand, the legal requirements are particularly strict, for example with data protection or the ability to explain the results – “that is the point where we are particularly strong in an international comparison”.

>> Read here: How Sam Altman will change your life with ChatGPT

A demo shows how it works: One day, a program for lawyers could not only answer questions about legal documents such as employment contracts, but also mark the original passages so that the information can be understood at a glance – like a hard-working assistant. OpenAI does not currently offer such a source, but is rumored to be working on it.

Digital citizen assistant Lumi

The system answers questions about childcare, housing benefit or garbage disposal.

The OpenAI language model is also suitable as a basic technology for other programs, especially since Microsoft wants to market it via its own cloud platform Azure. According to estimates in industry circles, Google and Amazon Web Services (AWS) will develop their own offers.

Independence from the corporations could be an advantage for Aleph Alpha, says Andrulis: “Many companies don’t want to be dependent on Microsoft or Google.”

Aleph Alpha in comparison test on par with OpenAI

The technology works similarly for Aleph Alpha and OpenAI. The language models use statistical methods to analyze how words are related to each other – for example, “sun”, “summer” and “sunscreen” are often mentioned in a context. Millions of texts from the Internet and from books serve as material.

>> Read here: Will AI help employees – or replace them?

Analogous to a human brain, an artificial neural network has nodes that process information. The more of them, the more powerful. Internationally, Aleph Alpha is one of the top ten: Luminous Supreme has 70 billion parameters. The company is currently testing the new Luminous World version with more than 300 billion parameters.

A previously unpublished comparison available to the Handelsblatt is intended to show that the technology is competitive. Aleph Alpha pitted Luminous against several major language models, including OpenAI’s DaVinci and Meta’s Opt. The systems had to complete texts, answer multiple choice questions and write articles themselves. And that in five languages, in addition to German, for example, English and Spanish.

Arms race worth billions

In many tasks, Aleph Alpha was on par with the competition. The company can “play in the top class,” says Andrulis. “Luminous is a strong alternative in many environments and thus an important step towards Europe’s technological sovereignty.” The start-up wants to publish the data of the comparison in a reproducible manner so that they can be checked.

>> Read here: Microsoft attacks Google’s core business

However, the comparison is a snapshot. The hype surrounding ChatGPT is increasing investments in generative artificial intelligence, i.e. systems that produce text, images or program code. According to media reports from Microsoft, OpenAI alone received an additional ten billion dollars in the most recent financing round – to pay for the computing capacity, but also to further develop the product.

Aleph Alpha’s financing, on the other hand, is modest. The start-up has only received 28 million euros so far. The investors include the investor Klaus Hommels with his Lakestar fund, the Berlin early-stage investor Earlybird and the Munich tech financier UVC. It is impressive that the German start-up has developed a competitive technology with comparatively little funding. According to Andrulis, the need to continue to be among the best in the world is in the hundreds of millions.

The entrepreneur is planning a new round of financing for the current year. He has no concerns about finding financiers: “The financing situation is difficult for many start-ups at the moment, but that doesn’t necessarily apply to the field of generative AI.” The big hype has also reached the venture capital scene.

For the founder, it’s not just about his own company, but also about digital sovereignty: the goal must be to bundle competence and value creation in Europe. “Now it will be decided whether we will also lose this basic technology and become a customer of the USA.” In view of the revolutionary technology, this would have consequences for all forms of knowledge work.

More: With these four stocks you are betting on artificial intelligence.

source site-12