The startup Mistral from Paris wants to build an AI for Europe. The founders are industry professionals and come from Meta and Google. It’s worth millions to investors. But why?

AI connoisseurs: The founders of Mistral previously worked on some of the most well-known language models for artificial intelligence

Three-digit million amounts in financing rounds are currently in short supply. In the funding crisis, startups have to focus on profitability and often adjust their valuations downwards. The report that the AI ​​startup Mistral from Paris was able to collect 105 million euros is all the more surprising. Also because Mistral has not been on the market for years, but is only a few weeks old. It’s still in the very early stages, in the seed phase, with a product currently in development and not expected to hit the market until 2024. Even the company’s website is still under construction, there is only a contact address.

Two well-known German VCs, Headline and La Famiglia, were involved in the three-digit million round of this fresh start-up. Gründerszene asked the sponsors: Why is this AI startup so special that such high sums are justified at an early stage? Is that because of the current hype about artificial intelligence? Or the product, the vision or the team?

read too

Lilium Investor on Plunging Valuations: “The New Reality of the Market Will Last”

The founding team was at Deepmind and Meta

For Viet Le, Principal of La Famiglia, the founding team had a major influence on the investment decision. Mistral was made by the AI ​​experts Guillaume Lample, Arthur Mensch and Timothée Lacroix brought to life. Human was previously at Alphabet subsidiary Deepmind, working on the Chinchilla AI language model. It is one of the successful language models that the Google parent company is researching, as is PaLM, which is used in the Google chatbot Bard.

Lample and Lacroix worked for Meta (ex-Facebook) on the AI ​​language model LLaMa. The popular model is open source: Other programmers and scientists can look at the code, make improvements – and build their own apps on it.

“The founders of Mistral have proven that they can train and deploy state-of-the-art models,” says Viet Le. “We were convinced by the history and Mistral’s vision to build a strong and internationally serious European machine learning company”.

Christian Miele, General Partner of Headline, also relies on the founding team: They have “a unique starting position due to their experience”. It also shows a more open approach to science and technology compared to its competitors.

Mistral relies on open source

Mistral is working on a Large Language Model (LLM) for industry. Like Meta’s LLaMa model, Mistral is open source. Can this also have disadvantages from an investor’s point of view because an important asset – namely the source code – is freely accessible? Viet Le describes himself as a supporter of open source. According to the investor, it is “the best way to attract users, build a community and build transparency and trust for a technology”.

Christian Miele adds that “this allows users to check the fairness, robustness and ethical aspects of the AI ​​systems.” It also opens up “flexibility and adaptability, since engineers can adapt the models and algorithms to specific use cases and adapt them to different requirements”. It also means creating ecosystems and platforms instead of just functions.

read too

These 5 startups are the absolute insider tips of AI investor Adrian Locher

Many AI experts advocate the open source approach to artificial intelligence because it can speed up and improve development. While Mistral, like Meta, takes a more open approach to programming, companies like OpenAI and Google keep a low profile when it comes to their AI code. This has already led to internal debates there, as it could harm the companies in the long term.

AI training costs millions

What does Mistral actually need the 100 million for? The majority will be used to recruit the “world’s best” researchers and developers in the field of AI, according to Viet Le.

In addition, the hardware infrastructure to train the language models is very expensive. The German AI Association, for example, is committed to building a dedicated AI infrastructure to train large language models. Cost point: 400 million euros. “The high cost of computing power to train and run AI models is a very real hurdle for newcomers,” says Christian Miele. Headline’s general partner cites the training of the company’s own language models as the main source of costs.

Mistral wants to focus on European industry with its AI, which should be ready in 2024. Similar to scene star Aleph Alpha or the newly founded AI startup Nyonic by the former AI boss of SAP, Feiyu Xu.

One of the reasons: Under European law, EU companies could find it difficult to use AI software from abroad, such as OpenAI’s ChatGPT. The problem is data protection, as it is not always clear how the data will be used. In Italy, the use of ChatGPT had therefore been restricted in the meantime, and Germany was also debating a ban. In addition, the EU is currently working on an AI regulation to regulate artificial intelligence.

read too

AI pioneer: “Our inspiration was the Knight Rider series”

European AI as a counterpoint to the USA and China

How do Mistral investors assess the competitive situation? According to Headline’s Christian Miele, “Mistral competes with the best – eyes are on OpenAI”.

And what about the German competitors like Aleph Alpha and Nyonic? “This isn’t about Mistral vs Aleph Alpha vs Nyonic,” said La Famiglia’s Viet Le. He advocates “sending all horses into the race so that we in Europe can continue to be taken seriously economically and technologically,” says Viet Le. So it’s not about which startup wins in Europe, but “not being completely left behind and made dependent on China and the USA”.

He is currently relaxed about the regulation of AI in Europe. He does not expect the actual introduction of the EU AI Act for two years. By then, he hopes to have developed and deployed competitive models in Europe “that are aligned with our values.”

That means, for example: democracy instead of autocracy. Because: China wants to become an AI superpower by 2030 and is investing billions. And the Chinese government has clear rules for developing AI language models domestically: they must reflect the core values ​​of socialism and not spread information that could disrupt the economic or social order. So there are good reasons for an AI with European values.


Leave a Reply

Your email address will not be published. Required fields are marked *