Can Europe Take the Lead in AI?
I am very excited about Mistral AI, and here's why: The French startup is challenging OpenAI and advancing AI with a small yet creative team that upholds high scientific standards.
Today I'll cover:
- The nature of the team.
- Their current projects.
- Why I think their strategy is superior.
- The impact of disruption from Europe.
- Business opportunities.
- The roadmap.
- Testing their first products.
Let’s Dive In! 🤿
1/ The Nature of the Team
Mistral.ai, with its founding team of leading AI researchers and experienced entrepreneurs from DeepMind, Meta, and French startup Alan, stands at the forefront of Europe's generative AI race. This diverse group, known for their work on advanced language models like Chinchilla, Retro, Flamingo, and Llama, brings a unique blend of expertise and innovation. Their commitment to open-source development and European-centric values is further bolstered by a significant $100 million Seed funding led by Eurazeo and CNP Assurances and a whopping $415 million Series A funding at a $2 B valuation. This investment underscores the confidence in Mistral.ai's ability to surpass Big Tech in the US with its revolutionary AI models. Some of the investors include: French investment bank BPifrance and former Google CEO Eric Schmidt are also shareholders.
Mistral.ai's approach sets them apart not just in geography but also in their strategic focus and values. They emphasize tight integration of their models into customer workflows, offering more than just black-box solutions. With a keen eye on specialized industries such as finance and law, their partnership with European entities enhances their model's relevance and specialization. Furthermore, their commitment to algorithmic transparency and continuous monitoring for ethics and security is a testament to their responsible approach, aiming to establish a European leadership in AI that aligns with the continent's ideals. As they gear up to disrupt the generative AI landscape, the anticipation to test their initial products and witness their progress in developing models that surpass those from across the Atlantic is growing.
Most of the AI models today like GPT-4 are propietary models, you don’t know anything about them, everything happens behind a blackbox API.
What we want to promote is an open approach to models. We release their weights and release the best Open Source models. We want to be an Open Core LLM provider and that’s our main differentiator. - Arthur Mensch, Co-founder and CEO of Mistral AI
2/ Superior Strategy
Mistral.ai is on a bold journey to redefine the AI industry. Their strategy begins with open-source models, meticulously crafted to address and overcome the current shortcomings in the market. By utilizing high-quality data and cutting-edge training methodologies, Mistral.ai sets its sights on not just meeting but exceeding the capabilities of established benchmarks like ChatGPT 3.5. Their pursuit doesn't stop there; they're pushing the boundaries further by creating small, super-efficient models and prioritizing data control, which is crucial for enhancing both the quality and safety of their AI offerings.
Learn more about why Small Models are Game Changers: link
In a move to challenge the giants of the tech world, such as OpenAI and Google, Mistral.ai is focusing its efforts on developing generative language models that promise to be competitive, if not superior. This ambitious plan includes not only opening some of their models to the developer community, fostering innovation and collaboration, but also crafting specialized commercial models. These models, trained on meticulously curated high-quality data, are tailored for business applications, offering a unique blend of openness and specialization.
They are moving away from the traditional black-box API offerings. By emphasizing data sources and forming strategic European partnerships, they can tailor models for specific industries like finance, law, and manufacturing. Mistral.ai's approach reflects a deep understanding of the diverse needs in the AI arena, positioning them as a versatile and forward-thinking player in the field.
3/ Current Projects
Mistral AI released two models:
- Mistral 7B: A 7B dense Transformer, fast-deployed and easily customizable. Small, yet powerful for a variety of use cases. Supports English and code, and a 8k context window.
- Mixtral 8X7B: A 7B sparse Mixture-of-Experts model with stronger capabilities than Mistral 7B. Uses 12B active parameters out of 45B total. Supports multiple languages, code, and 32k context window.
Both models come with a Licence of Apache 2.0. It allows to use, change, and share freely for any purpose, including commercial. More on Open Source licenses here
They also announced this week La plateforme, a versatile platform featuring their models for developers, with a focus on efficient deployment and customization. The platform, in its beta phase, offers chat endpoints like mistral-tiny, mistral-small, and mistral-medium, each designed for specific performance and cost requirements. These models support multiple languages and are fine-tuned for user control. Additionally, Mistral-embed provides a specialized embedding model. The platform is user-friendly, aligning with popular chat interfaces and supported by programming libraries for easy API access, reflecting Mistral AI's commitment to innovation and user-centric design​​​​​​​​​​​​​.
4/ Impact of Disruption from Europe
Emerging as a European leader in AI, Mistral.ai is set to challenge the current market dominated by US-based companies. Their strategy of developing open, auditable models and focusing on European data sovereignty is a potential game-changer. This not only addresses geopolitical concerns but also promises to usher in a new era of AI development centered on European values and standards​.
Mistral.ai seeks explicitly to bring leadership in generative AI out of US big tech and establish European capabilities and stewardship. As they note, this technology will transform industries globally in the coming years. By leading disruption from a European vantage point, they can bake in values like openness, customization for local markets, privacy protections, and responsible development patterns. This European counterbalance serves an important role.
Their latest developments reflect Mistral AI's blend of Silicon Valley-level expertise and a civic-minded ethos, aligning with the European regulatory landscape. Their recent achievements underscore Mistral AI's role in preserving Europe’s cultural voice in the AI domain, as highlighted by co-founder Cedric O.'s commentary on the influence of AI on cultural perspectives. The company's success in the recent political negotiations around the EU AI Act further positions Mistral AI to shape the future of AI development in Europe, balancing civic responsibility with commercial viability​
5/ Business Opportunities
Mistral.ai's business model revolves around providing cutting-edge AI-as-a-service. By co-building solutions with European integrators and clients, they're positioned to become a key player in the AI industry. Their focus on customizable models and partnerships with major industrial actors indicates a robust and diverse business strategy, poised to capture a significant market share.
Numerous businesses are keen to leverage generative AI, but have concerns today about sending sensitive data to black-box third-party models with opaque training data. Mistral.ai is already fostering partnerships with key French and European companies to explore co-building industry-specific solutions with transparent models tailored to their needs. These close customer integrations generate valuable real-world feedback to drive Mistral’s models forward while producing revenue and impact.
6/ The Roadmap
The roadmap for Mistral.ai includes developing a family of generative models with both text and visual outputs, focusing on high-quality, specialized data sources. They aim to establish a strong presence in the market by Q2 2024, offering scalable APIs and forming strategic commercial relationships. This ambitious timeline demonstrates their commitment to rapid development and market penetration​.
Mistral.ai plans to first release open-source models by the end of 2023 that beat alternatives like GPT-3.5 and BARD. In 2024, they will enhance these with superior semantic search, visual interfaces, and industry specialization. Concurrently, they are working with major partners like Airbus on custom deployments and collaborations. This two-pronged approach strengthens their capabilities and reputation on both the open source side as well as commercial side to set up leadership potential.
To learn more about their roadmap, I recommend you read their strategic memo:
7/ Testing Their First Products
There are multiple ways to experiment with Mistral Models. First, you can try today Mistral AI_’s Open Source models, all available on HuggingFace:
- mistralai/Mixtral-8x7B-v0.1
- mistralai/Mixtral-8x7B-Instruct-v0.1
- mistralai/Mistral-7B-v0.1
- mistralai/Mistral-7B-Instruct-v0.2
You can request access to their hosted endpoints and platform here: https://console.mistral.ai/
Mistral Models are already included in 3rd party platforms. My favorite No-code AI platform for LLMs is Stack AI and already includes Mistral AI models as part of its open catalog:
8/ Wrapping Up!
Mistral.ai's unique blend of expertise, strategic focus on open-source development, and commitment to European values positions it as a formidable player in the global AI landscape. Their ambitious projects and business strategies hold the promise of significant advancements in AI, making them a company to watch in the coming years.