Want to learn the best business insights from remarkable speakers at Slush 2023, but don’t have the time to watch the full speeches on YouTube?
You’ve come to the right place. Below is a summary of a speech by Arthur Mensch, Co-founder and CEO of Mistral, in conversation with Paul Murphy, Partner at Lightspeed Venture Partners.
(psst: the notes were automatically generated with Wudpecker's AI notetaking tool. If you'd like to check these notes + transcript + audio recording without needing to log in, press here.)
<p class="h1-rich">🤏 TL;DR</p>
Mistral, a company that aims to make foundational AI models more accessible to developers, has built the 7B model in just 3 months and has seen thousands of derivative works from developers who fine-tuned the model for their own tasks and datasets. Mistral plans to announce new models, new techniques, and the beginning of a platform offering hosting capacities for their models with fast inference capabilities. The company believes in a pragmatic approach that focuses on enabling useful applications today rather than pursuing multi-usage, large models and AGI.
<p class="h1-rich">✨ Summary</p>
Introduction
- Light Speed is a Silicon Valley-based fund that has been investing in Europe since 2007.
- They have invested over a billion dollars in AI and have a portfolio of 50 AI companies.
- Mistral, founded by Arthur, Guillaume, and Timote, aims to make foundational AI models more accessible to developers.
Building the 7B Model
- The Mistral team built the 7B model in just 3 months by working intensively and focusing on creating a good training code base, inference code base, and high-quality datasets.
- They have seen thousands of derivative works from developers who fine-tuned the 7B model for their own tasks and datasets.
- Developers have created new capabilities, such as longer context and better instruction following, using the 7B model.
Future Plans
- Mistral plans to announce new models, new techniques, and the beginning of a platform offering hosting capacities for their models with fast inference capabilities.
Challenges
- Hiring top talent is a major challenge for Mistral.
- Creating and engaging with the developer community is another challenge.
- Policy matters, such as hardware regulation and safety, keep Mistral up at night.
Differentiating Philosophy
- Mistral differentiates itself by targeting the developer space and enabling developers to create specialized and cost-efficient models for their applications.
- They believe in a pragmatic approach that focuses on enabling useful applications today rather than pursuing multi-usage, large models and AGI.
- Mistral aims to empower developers to align models with their own values and control biases.
Open Source Approach
- Mistral provides open weights for their models, allowing modification and customization.
- While they support open source principles, they keep some aspects proprietary for competitive advantage and due to the difficulty and cost of obtaining datasets.
- The open weight approach helps with modifying biases, improving interpretability, and conducting red teaming.
Regulation and Safety
- Mistral believes that regulation should focus on product safety at the application layer, which will create market pressure on foundational model developers to provide controllable models and evaluation tools.
- They emphasize the importance of empirical evidence and caution against regulating the foundational model layer directly, as it may favor big actors and hinder competition.
- Independent organizations should monitor product safety and ensure regulation standards are not influenced by industry pressure.
Utopian Future with AI
- AI can revolutionize healthcare by empowering physicians to make better decisions.
- Personalized education can be achieved through AI, especially in regions like the global south.
- AI can free up time for creative thinking and help address challenges like climate change through breakthroughs in fields like chemistry and material science.
Importance of a European Champion
- Having a strong technological actor in Europe is crucial for shaping AI technology according to European values and driving policy and technological advancements.
.svg)
.jpg)
.svg)
