The Unconventional Release of Mistral's MoE 8x7B Model
Introducing Mistral “Mixture of Experts”!
12/9/20232 min read
Mistral, a prominent player in the AI community, recently took a different approach to releasing their highly anticipated new model, MoE 8x7B. In stark contrast to Google's Gemini release, which garnered mixed reviews, Mistral opted for a simple yet impactful strategy.
Instead of an elaborate professional video, Mistral chose to share a link to a large torrent file, allowing users to directly download their new model. This unconventional method immediately generated a buzz within the AI community, drawing attention to the stark differences between Mistral and Google's approaches.
Google's Gemini release, as described by OpenAI's Andrej Karpathy, featured an over-rehearsed professional video that promised a revolution in AI. However, one of the demo videos from the launch came under heavy criticism for its apparent editing and staging, which made the AI system appear more capable than it actually was.
Mistral's approach, on the other hand, was refreshingly straightforward. By providing a direct download link, they allowed users to experience the MoE 8x7B model firsthand, without any embellishments or misleading representations. This transparency resonated with many in the AI community, who appreciated Mistral's commitment to delivering an authentic and unfiltered experience.
The decision to share the new model via a torrent file also sparked discussions about accessibility and inclusivity. Torrents, often associated with piracy, have both positive and negative connotations. While they can facilitate the distribution of large files efficiently, they have also been used for illegal purposes.
Mistral's use of a torrent file raised questions about the democratization of AI and the potential for more open-source models. By making the MoE 8x7B model available through a torrent, Mistral challenged the traditional model of controlled releases, inviting the AI community to explore and experiment with their creation.
While Mistral's approach may have been unconventional, it undoubtedly achieved its goal of creating a community buzz. The simplicity and transparency of their release stood in stark contrast to Google's Gemini, prompting discussions about the best way to introduce new AI models to the world.
As the AI landscape continues to evolve, it is essential to consider alternative methods of sharing and distributing models. Mistral's bold move with the MoE 8x7B model serves as a reminder that innovation can come in unexpected forms, challenging established norms and sparking valuable conversations within the AI community.
Edited and written by David J Ritchie