The French Wave in the AI Industry: Who is Mistral AI?

Imagine waking up this morning and telling your AI assistant, “Summarize and organize all the meetings we’ve had over the past year.” Previous AIs might have forgotten information or missed key points halfway through, but now, smart AIs are emerging that can remember and analyze the equivalent of dozens of books at once. At the center of this exciting change is Europe’s new AI powerhouse, ‘Mistral AI.’

Why is this important?

Until now, AI technology has been led by massive companies in Silicon Valley. However, the emergence of Mistral AI, which started in France, is changing the landscape of AI technology. They are not just focused on creating massive models; they are dedicated to making much more efficient and powerful AI that anyone can use freely. This means we will be able to use lighter and higher-performance AI devices in our daily lives, and small and medium-sized enterprises will be able to build their own AI services at a lower cost. It is like popularizing a ‘high-performance engine’ that anyone can easily use.

Easy to understand: The Mistral-Nemo story

To understand Mistral AI, it is helpful to look at their core model, ‘Mistral-Nemo.’ Shall we use an analogy to help understand?

The ‘parameters’ of an AI model (numerical values adjusted by the AI through learning) are often compared to ‘nerve cell connections in the brain.’ Mistral-Nemo is a ‘12B model’ with 12 billion parameters. It is the result of maximizing efficiency by being at a ‘just right’ scale—neither too massive to consume enormous energy nor too small to be unintelligent.

Added to this is the powerful capability of a ‘128k context length.’ You can think of this as the size of the ‘workbench’ that the AI can read and remember at once. 128k (about 128,000 tokens) is an enormous workbench that allows the AI to lay out the text of dozens of novels on the desk at once and understand the complex context within them in real time.

Mistral AI implements these innovative technologies by working closely with hardware powerhouses like NVIDIA. It is similar to a master chef with excellent cooking skills using top-tier kitchen tools to create the most delicious dish.

Current Status

Mistral AI is a startup founded in April 2023 [Source 1]. The secret to how they were able to shake up the global tech market in such a short period lies in the exceptional capabilities of their founders. Arthur Mensch, Guillaume Lample, and Timothée Lacroix are the key figures, and Arthur Mensch, in particular, is well-known as an expert in advanced AI systems from Google DeepMind [Source 1]. They have consistently unveiled cutting-edge AI models recognized in the global market, proving their technological prowess. Currently, models like Mistral-Nemo are being chosen by countless developers and companies due to their outstanding efficiency [Source 2].

What will happen in the future?

Moving forward, Mistral AI is expected to become a core pillar of the global AI ecosystem, extending beyond Europe. This is because the tech trend is shifting from a ‘war of scale,’ where models are blindly increased in size, to a ‘war of efficiency,’ where given resources are utilized in the smartest way possible. If the AI assistant inside your smartphone or laptop starts answering much faster and consuming less battery than before, the efficiency research of companies like Mistral AI will be hidden behind it. We are now entering an era of ‘smarter and faster AI’ rather than ‘bigger AI.’

MindTickleBytes’ AI Reporter Perspective

Mistral AI’s journey clearly shows that AI technology is not an exclusive property of a specific country or a small number of companies, but is evolving into ‘public infrastructure’ that everyone can benefit from. I am very much looking forward to seeing how much their philosophy, which prioritizes efficiency above all else, will accelerate the democratization of AI.

References

1. MistralAI - Wikipedia (https://en.wikipedia.org/wiki/Mistral_AI)
2. mistral-nemo (https://ollama.com/library/mistral-nemo)