After OpenAI became popular, Microsoft began to raise its "shotgun" again, seeking to expand its footprint in the rapidly developing artificial intelligence industry. Its next goal is to target a French AI startup that has only been established for ten months.
On February 26th, Microsoft announced a new partnership with French startup Mistral AI. Although detailed financial details were not disclosed, Microsoft stated that this partnership will help it unlock "new business opportunities" and expand its business to global markets.
According to the agreement, Mistral's large-scale language model will be provided on Microsoft's Azure cloud computing platform, and Mistral's multilingual dialogue assistant "Le Chat", which directly targets ChatGPT, will also be launched.
On February 26th, Mistral AI released the latest Mistral Large large-scale language model. Although the parameters of Mistral Large are smaller than those of GPT-4 in terms of large model parameters, it can almost reach the same level as GPT-4 in important performance aspects such as inference.
Microsoft President Brad Smith stated that this transaction is an important signal of the company's support for European technology. "I do believe that this day is one of the most important days for Microsoft to support European technology," he said at the MWC World Mobile Conference held that day.
Smith stated that investment in Mistral AI also includes funds specifically for research and development, such as establishing artificial intelligence models for European public sector services.
Microsoft is not the only one collaborating with Mistral AI. Just a few days ago, Amazon Cloud Technology (AWS) announced on its official website that two high-performance Mistral AI models, Mistral 7B and Mistral 8x7B, are about to be launched on Amazon Bedlock. This is also the seventh basic model provider introduced by AWS on Amazon Bedlock. Amazon Bedlock is a platform promoted by AWS to support the construction and expansion of generative AI applications.
Only ten months after its establishment, Mistral AI has been endorsed by Microsoft and Amazon, indicating that this startup team has a considerable background. The valuation of this company had already reached $2 billion before Microsoft's investment, and it is even considered the next AI enterprise that is most likely to compete with OpenAI.
The birth of Mistral AI can be said to be "holding a golden spoon". Just four weeks after its establishment, the company received over $100 million in seed funding and declared its intention to break the monopoly of American technology companies in the AI industry.
The founding members of Mistral AI come from DeepMind, a subsidiary of Google, as well as Meta. The company's co-founder and CEO, Arthur Mensch, worked in DeepMind's Paris office. The other two co founders, Timotee Lacroix and Guillaume Lample, worked in Meta's AI laboratory in Paris. Meta's chief AI scientist, Yann Lecun, is also French and led the development of Meta's large-scale language model.
The three founding team members are all young and in their thirties, but Mistral AI has a very senior founding advisory team, including Cedric O, former Minister of Digital Affairs in France, and the other two are the CEO and CTO of French health startup Alan.
Although the company is not large, Mistral AI aims to build a "world-class team" with the goal of building open source models. The company's mission is to elevate publicly available models to the most advanced performance. Co founder Arthur Mensch proposed that Mistral AI should become the most cost-effective among large model companies.
According to the introduction of Mistral AI, a major highlight of the company's large model is that it achieves a balance between cost and performance, and has excellent inference speed with low memory requirements but relatively high throughput.
Although the company's most advanced Mistral Large is not yet open source, Arthur Mensch stated that finding a balance between building a business model and adhering to open source values is very delicate. The company hopes to create new things and architectures, but also wants to provide customers with some additional products and services.
One important reason why Arthur Mensch left Google was his unwillingness to "develop opaque technology in large enterprises.". For this reason, he emphasized that Mistral AI's model is transparent and customizable, available to a wide range of users, and able to meet strict regulatory requirements. He also believes that everyone should have more control over the AI they use.
The EU's regulation of AI big models is becoming increasingly stringent. Microsoft's $13 billion investment in OpenAI deal also faces scrutiny from EU antitrust regulatory agencies. Some market insiders believe that this investment in Mistral AI may be a measure taken by Microsoft to appease competitive concerns.
Smith stated that the company hopes to provide diversified products. He said, "For us, it is important to demonstrate that this (AI big model) is not just Microsoft's technology, nor is it just a product from the United States; it will become the engine of technology, innovation, and growth in Europe."
At present, besides serving commercial companies, the government is also a more important major customer for technology giants in the future. NVIDIA founder and CEO Huang Renxun told attendees at the 2024 World Governments Summit held in Dubai on February 12 that each country needs to have its own AI data sovereignty. He also emphasized the importance of building AI infrastructure and suggested that governments input language and cultural data from various countries into their own AI big models.
With the increasing popularity of emerging technologies, rival technology giants such as Google and Amazon have also been increasing their investment in artificial intelligence. Driven by the fruitful performance of artificial intelligence chip manufacturer Nvidia, the global market reached a historic high last week.