20VC: Mistral's Arthur Mensch: Are Foundation Models Commoditising | How Do We Solve the Problem of Compute | Is There Value in the Application Layer | Open vs Closed: Who Wins and Mistral's Position

Arthur Mensch is the Co-Founder and CEO of Mistral AI. Since its inception in May 2023, Mistral has raised over $520M in funding from investors like Andreeseen Horowitz, General Catalyst, Lightspeed Venture Partners, and Microsoft with a current valuation of $2 billion. Before founding Mistral, Arthur was a research scientist at DeepMind, one of the leading AI institutions in the world.

In Today’s Episode with Arthur Mensch We Discuss:

1. From Models to Team Building: Arthur’s Greatest Lessons at DeepMind

• What were Arthur’s biggest lessons from his time at DeepMind?

• How did DeepMind shape how Arthur built Mistral?

• Why does Arthur believe smaller teams are better for AI?

• Why did Arthur decide to leave DeepMind and start Mistral?

2. Scaling Mistral to $2 Billion Valuation Within a Year

• What made Mistral 7B so successful? What did Arthur learn from the model release?

• What are the biggest barriers at Mistral today?

• How does Arthur balance the sales and research teams at Mistral?

• What does Arthur know now that he wishes he had known when he started Mistral?

3. How to Win in AI: Open Source, Cost, & Adoption

• Why did Arthur open-source some models? Why did he close some?

• How quickly will the cost of compute go down? Why does Arthur believe marginal costs will not go to zero?

• How will open-sourcing LLMs affect the marginal cost?

• Does Arthur think open source is ready for enterprise adoption?

• What questions should enterprises be asking about AI adoption today?

• What are the biggest challenges to AI adoption today?

4. The Future of LLMs

• What does Arthur think are the largest bottlenecks of model quality today?

• Does Arthur think future models will be more generalized or vertical-focused?

• What does Arthur think about the future of commoditization in models?

• Why is Arthur optimistic about the profitability of the application layer of AI?

• How should models differentiate themselves today?