Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
French developer Mistral AI It launches a new set of language models designed to bring cutting-edge AI capabilities to more people, regardless of where they are, how reliable their internet access is, or what language they speak.
The company on Tuesday announced a new large language model called Mistral Large 3, intended for large-scale, general-purpose uses. He thinks ChatGPT or twin. Other models come in a range of sizes and capacities and are designed for use On devices themselves. These smaller models can run on laptops, smartphones, in cars or on robots, and can be tuned to perform specific tasks.
All models are Open source And open weight, which means developers who use it can see how it works and modify it to suit their needs. “We strongly believe that this will make AI more accessible to everyone, and essentially put AI in their hands,” Guillaume Lampel, co-founder and chief scientist at AI company Mistral, said in an interview.
Mistral AI, founded by former Google DeepMind and Meta researchers, isn’t as big a name in the U.S. as rivals like OpenAI and Anthropic, but it’s better known in Europe. Besides the templates available to researchers and companies, it offers a chatbot called Le Chat, which is available Via browser Or in app stores.
AI models are designed to be multilingual
Lampel said the company has a goal with its new set of models to provide cutting-edge AI capabilities that are open source and accessible. Part of that has to do with language. Most popular AI models in the United States are designed primarily for use in English, as well as benchmarking tools that compare the models’ capabilities. Although these models are capable of working in other languages and translating, they may not be quite as good as the standards suggest when used in languages other than English, Lampel said.
Watch this: Can artificial intelligence develop gambling addiction? AI-Fueled Browser Wars, and the Future of Work with ZDNET’s Jason Hainer | Technology today
Mistral AI wanted its new models to perform better for speakers of all languages, so it increased the amount of non-English training data in proportion to the English data. “I think people usually don’t put too much pressure on multilingual abilities because if they do, performance will also deteriorate a little bit against the common standards that everyone sees,” Lampel said. “So, if you want your model to really shine on common benchmarks, you have to sacrifice multilingual (performance). Conversely, if you want your model to really do well in multilingual, you have to sacrifice common benchmarks, basically.”
A variety of sizes for a variety of uses
In addition to the general-purpose Large Mistral 3 model, with a total of 675 billion parameters, there are three smaller models called Ministral 3 – 3 billion, 8 billion and 14 billion parameters – each coming in three variants, for a total of nine. (A parameter is a weight or function that tells a model how to handle its input data. Larger models are better and more capable, but they also need more computing power and run slower.)
The three types of smaller models are divided in this way: a basic model that can be modified and tweaked by the user, a model that is fine-tuned by Mistral to perform well, and a model designed for inference that spends more time repeating and processing the query to get a better answer.
Read more: AI basics: 29 ways you can make AGI work for you, according to our experts
Smaller models are especially important because many AI users want something that does one or two tasks well and efficiently versus large, expensive generic models, according to Lampl. Developers can customize these models for those specific jobs, and any person or company can host them on their own servers, saving the cost of running them in a data center somewhere.
Smaller models can also run on specific hardware. A small device can run on your smartphone, and a slightly larger device on your laptop. This has privacy and security benefits – your data never leaves your device – as well as cost and energy savings.
The little model running on the device itself doesn’t need internet access to work either, which is vital when you think about using AI in things like robots and cars, where relying on reliable Wi-Fi for things to work properly isn’t the case.