Large language models (LLMs), one of the main technologies supporting the explosion in generative AI tools, may have a geographical future. The Paris-based AI business Mistral, which is competing with companies like the U.S.-based Anthropic and OpenAI, is launching a model on Monday that differs slightly from its typical LLM.
The new custom-trained model, called Mistral Saba, is intended to target a particular geographic area. nations that speak Arabic. Mistral Saba wants to be the best at interacting in Arabic.
Mistral Saba has 24 billion parameters, making it a rather tiny model. Recall that lower latency and higher performance are typically achieved with fewer settings. Although there isn’t a direct correlation, more parameters typically translate into more intelligent responses.
Mistral Saba is around the same size as Mistral Small 3, which is a small, all-purpose model. However, Mistral’s own tests show that when it comes to handling Arabic content, Mistral Saba outperforms Mistral Small 3.
According to Mistral, Saba also functions well with Indian-origin languages, particularly South Indian-origin languages like Tamil and Malayalam, as an intriguing byproduct of cultural exchange between the Middle East and South Asia.
The French AI behemoth’s new model demonstrates a greater emphasis on the Middle East, which is an intriguing strategic decision. According to Mistral, the company anticipates that the model will help it attract clients in the area.
Mistral Saba is an off-the-shelf model that could be used to generate more relevant and natural-sounding Arabic content or to augment discussion. According to the company, it can also serve as the foundation for some refined models for internal use cases.
Mistral showed that it is taking business seriously last week at the AI Action Summit. Even while the business has already raised a significant sum of money from foreign investors, a big number of its foreign backers, including Salesforce Ventures, Andreessen Horowitz, and Lightspeed Venture Partners, are based in the United States.
Given the dynamic geopolitical environment, Mistral may be able to attract Middle Eastern investors to its next investment round.Â
It would be a means of raising additional funds to maintain technological relevance in the AI competition while establishing itself as the global substitute for Chinese and American AI firms. Thus, Saba, Mistral’s newest model, might help with that possible fundraising initiative.
