Mistral AI, a French AI startup, has announced the upcoming release of an open-source GPT-4 level model in 2024. During a recent interview on French national radio, Arthur Mensch, CEO of Mistral AI, unveiled the company’s ambitious plans for the future of artificial intelligence. Mensch hinted at a groundbreaking development—an anticipated release of an open-source GPT-4 level model in 2024.
Table of Contents
About Mistral AI
Mistral AI is a French startup founded in May 2022 by three AI experts – Arthur Mensch, Guillaume Lample, and Timothee Lacroix. These founders have impressive academic backgrounds and gained experience working at AI labs operated by major American tech companies in France before starting their own company, “Mistral AI.”
The company has quickly gained recognition in the field of AI. They specialize in developing and releasing open-source language models that can be used for commercial purposes. Mistral’s models are trained using publicly available data and offer extensive customization options for developers to adapt them to different applications.
Mistral’s Fundraising Success
Arthur Mensch, the 31-year-old founder and CEO of Mistral AI, revealed the news on French national radio. Mensch was interviewed on station France Inter by anchor Sonia De Villers about Mistral’s remarkable ascent in the AI sector.
In only seven months, Mistral got €385 million in funding. First, €105 million in June, then €280 million this week. This is amazing because Mistral only has 22 workers but has become one of Europe’s top two AI companies.
Arthur Mensch stated, “A funding round is a promise, not a success. It’s something that commits us.” According to him, the goal of this funding was to accelerate the capacity of models, which is closely related to the computing power used to train them.
Surpassing GPT-3.5 Capabilities
Mistral AI provides open-source natural language models to corporate customers by training algorithms on publicly available data. Mensch stated their current models, Mistral 7B and Mixtral 8x7B, outperform the OpenAI GPT-3.5 capabilities. Furthermore, these models also surpass many LLMs, including Llama 2 models (7B to 70B).
Anticipated Release of GPT-4 Level Model
The key announcement from Mistral AI and Arthur Mensch is their goal to release an open-source AI model rivalling GPT-4 by the year 2024. Mensch expressed confidence that Mistral AI can match and even exceed the GPT-4 model. The goal is to achieve parity with GPT-4 or better within months, not years.
| Also Read: Has OpenAI Really Released GPT-4.5-Turbo?
Mistral AI Open-Source Approach
Mensch believes focusing on transparency gives Mistral an edge compared to closed competitors. Unlike OpenAI’s closed approach, Mistral AI openly shares access to its underlying models.
When asked whether Mistral AI considers itself a true rival to OpenAI, valued at $29 billion, Mensch added, “The difference in size is so significant.” Moreover, he said that Mistral AI models differ considerably by empowering developers to customize functionality for unique use cases extensively.
This contrasts with ChatGPT, which OpenAI wholly controls. Mensch argues that the ability to modify and augment capabilities makes Mistral’s models far more dynamic and attractive to potential commercial buyers.
| Also Read: Make any open-source LLM model uncensored with this simple trick!
The European Regulation of AI
The European Union recently reached an unprecedented legislation agreement to regulate artificial intelligence (AI). Mistral AI supports regulation, believing that for secure deployment of AI in applications, it is essential to have product safety. However, the company wants to maintain a competitive advantage over its American competitors and doesn’t see the regulation as a hindrance to innovation.
Mistral’s Vision for the Future
Mistral AI has a clear vision for the future. The company recently released the 8X7B, which is based on the MoE architecture and comparable to other popular models such as GPT 3.5 and Llama 2 70B. Licensed under Apache 2.0, Mistral surpasses Llama 2 70B on most benchmarks with 6x faster inference. Mistral’s goal is to go beyond 4, indeed. That’s why they raised funds. And so, this milestone is counted more in months than in years. The focus is more on next year.
Conclusion
Arthur Mensch’s announcement regarding Mistral AI’s intention to introduce an open-source GPT-4 level model in 2024 has sparked tremendous interest within the tech community. The company’s rapid growth, visionary approach, and commitment to pioneering advancements underscore its pivotal role in shaping the future landscape of artificial intelligence. If Mistral AI unveils GPT-4 capabilities to global partners in little more than a year, it may force established American AI dynasties to look over their shoulders.
| Read More from Mistral AI:
