According to the latest news and tweets, the Mistral Medium leak has revealed Miqu 70b. What is even more interesting is that the company’s CEO, Arthur Mensch, confirmed it on X with a fairly funny tweet.
The recent leak of an AI model called “miqu-1–70b” has caught the attention of tech enthusiasts and experts alike. This event started with a simple file upload by someone named “Miqu Dev” on a platform called HuggingFace, sparking widespread interest and conversations about Mistral, a leading AI company in Paris. Arthur Mensch, the CEO of Mistral, later confirmed that the leaked model was indeed an older version from their company, accidentally shared by someone.
Mistral Medium leak reveals Miqu 70b
Leaks aren’t uncommon in the bustling world of technology, but they seldom fail to stir excitement and curiosity. Such is the case with the recent buzz surrounding the AI community following the unexpected appearance of a new language model dubbed “Miqu 70b” on HuggingFace, a leading platform for open-source AI technologies. This leak has ignited discussions across various social media platforms, with everyone eager to understand what “Miqu 70b” is and what it signifies for the future of AI development.
Might be late but I am now 100% convinced that Miqu is the same model that's accessible as Mistral-Medium on Perplexity Labs. It was plausible that it knows standard puzzles, but there ain't no way in Hell a pranker has tuned it to identically phrase the responses in Russian too. pic.twitter.com/zZMcpspXch
— Teortaxes▶️ (@teortaxesTex) January 30, 2024
The story began to unfold when a user named “Miqu Dev” uploaded files to HuggingFace, allegedly showcasing a new Large Language Model (LLM) closely related to the technology developed by Mistral, a Paris-based front-runner in open-source AI. The situation took a dramatic turn when an anonymous post appeared on 4chan, possibly made by “Miqu Dev,” leading to a frenzy of online discussions about the nature and origin of this model.
The AI community, spread across platforms like X and LinkedIn, dove into analyzing the potential of this mysterious model, with speculations flying about its origins and capabilities.
Comparison: Can Mistral 7B really beat GPT-3.5 Turbo?
Mistral Medium Leak confirmed by company officials
These speculations were soon addressed by none other than Arthur Mensch, the CEO of Mistral, confirming the leak’s authenticity. In a surprising twist, it was revealed that “Miqu 70b” is indeed a quantized version of an older Mistral model, accidentally leaked by an “over-enthusiastic employee” of an early access customer. This model, which some believed to be an internal leak or a rogue move, turned out to be a crucial piece of Mistral’s development puzzle, showcasing the firm’s progress and ambition in the competitive field of AI.
An over-enthusiastic employee of one of our early access customers leaked a quantised (and watermarked) version of an old model we trained and distributed quite openly.
To quickly start working with a few selected customers, we retrained this model from Llama 2 the minute we got…
— Arthur Mensch (@arthurmensch) January 31, 2024
What Is Miqu 70b?
The intrigue around “Miqu 70b” largely stems from its performance. Early tests by the AI community suggest that it’s not just any model; it performs as well as, or even better than, Mistral’s best open-source models. This includes surpassing Mistral’s top models in some tests and ranking just below GPT-4 in others. Such achievements have understandably caused a stir, pointing to “Miqu 70b” as a notable advancement in AI technology.
Arthur Mensch’s insights reveal that “Miqu 70b” is a modified version of an older model that Mistral had developed. It was based on another technology called Llama 2, and it was finalized right when Mistral was releasing another major model. This background offers a glimpse into Mistral’s approach to developing cutting-edge AI technologies.
A new approach to leaks?
What’s particularly interesting about this situation is how Mistral responded. Instead of demanding the removal of the leaked model from HuggingFace, Mensch playfully suggested that the uploader might want to give proper credit next time. This reaction suggests a shift in how tech companies might deal with leaks in the future, focusing more on acknowledgment and potential collaboration rather than legal action.
Mistral AI’s Mixtral 8x7B surpasses GPT-3.5, shaking up the AI world
What’s next?
The leak has spotlighted “Miqu 70b” and brought attention to Mistral’s innovative strategies and their progress in the AI field. The community’s response has been largely positive, viewing this as an exciting development in artificial intelligence. With Mistral’s promise of more advancements on the horizon, the tech world is eagerly awaiting what they’ll come up with next.
In conclusion, the leak of “Miqu 70b” has provided a unique insight into the fast-paced world of AI development. It underscores the importance of innovation, collaboration, and perhaps a new way of navigating the unpredictable waters of technology leaks. As we watch Mistral’s next moves, it’s clear that the future of AI is bright, with unexpected turns that can lead to groundbreaking advancements.
Featured image credit: Mistral AI
