A French artificial intelligence company called Mistral has announced the first multimodal model, Pixtral 12B. This powerful fresh new model which is equipped with astonishing 12 billion parameters can be considered as one of the Milestones of the Mistral’s journey to the development of the new AI technologies.
Pixtral 12B has been launched in the backdrop of Mistral’s latest $ 645 million funding round, thus taking the valuation of the company to $ 6 billion. Mistral’s rise is one to watch in the AI space, they are already establishing themselves as Europes answer to OpenAI with a focus on next gen, open source models.
A Versatile Multimodal Powerhouse
What makes the Pixtral 12B unique is that it has been trained as a multimodal model so it can naturally work with images and text. Pixtral 12B is based on Mistral’s text-based Nemo 12B model and, leveraging URL or base64 input, can process the unlimited number of images of any size.
Such flexibility allows expanding the usage of AI to a broad spectrum of problems which involves image captioning, object counting, augmented object detection, recognizing the context, and understanding the textual data. Thus, with Pixtral 12B, Mistral wants to help developers and researchers make the most of multimodal AI for the development of various solutions.
Accessible and Open for All
This is not to mention the availability of Pixtral 12B – the anticipation for the new release is easily one of the most thrilling features of Pixtral. Mistral has shared the model with the public through the repository site GitHub, in form of a torrent link and here on the popular and comprehensive Artificial Intelligence development platform, the Hugging Face under Apache 2. 0 license. In other words, this allows developers to download, fine-tune and use Pixtral 12B without any restrictions thus encouraging collaboration as well as innovation in the development of AI systems.
However, a working web demo of Pixtral 12B was not available at the time of the announcement, but Sophia Yang, who is the head of Mistral developer relations, has since confirmed the availability of Pixtral 12B for testing at Misra l’s chatbot and API-serving platforms: Le Chat and Le Plateforme. It will give the developers an ideal chance of getting a feel of the model and get an opportunity to contribute to the enhancement of the model.
Addressing the Data Dilemma
Like most of the generative AI systems, the creation of Pixtral 12B has left questions concerning the data utilized for training. Most existing AI models use large dataset that was scraped from the public web which is most of the times of copyrighted content. While some model vendors insist such usage falls under the “fair use” rights, copyright holders are gradually responding to this by proceeding to file lawsuits against major vendors including Open AI, Midjourney among others.
Mistral has not yet provided information about what image data Pixtral 12B was trained with and there is still space for guessing and future lawsuits. Such issues are yet to be resolved with high sensitivity in the AI industry, therefore, it will be pivotal for the firms like Mistral to balance between the innovation and infringement of the intellectual property rights.
Mistral’s Rising Star
The launch of Pixtral 12B is the evidence of growth of Mistral as a tech giant in the sphere of AI. This achievement was marked within a year and a half since the establishment of the company, today it has attracted considerable investment, cooperation with investors and partners, and has become an important player in the European market of artificial intelligence.
While dedicating its efforts to developing new horizons in the implementation of AI Mistral is steadily generating a strong upswing in becoming a global player. The new Pixtral 12B release proves once again that Mistral is a company ready to deliver and bring change to the AI world by making it easily accessible and turn it into a collaboration platform.
