A French Startup called Mistral released its new AI model on February 26th. The LLM language is called Mistral-Large. Compared to GPT-4 it is a smaller language apparently with fewer parameters. Yet the founders claim that it has capabilities quite similar to American AI pioneers. Number of parameters has been used as a gauge in predicting model strength. However many scientists believe that the number of parameters is only one factor, model creativity is equivalently important for large language models.
Although the exact number of parameters are not divulged either by Mistral or OpenAI. It is perceived that Mistral-Large model parameters count in billions whereas ChatGPT model contains something around 1.8 trillion variables. The Mistral model is much smaller, it can run on private computers, on the other hand GPT-4 can only run on the strong cloud datacenters.
The French Startup is also betting on political issues. As the AI models are getting more important, many European firms may prefer to use Europe-based AI models rather than being dependent on American ones. Politicians and governments also play a major role here. They obviously prefer to become “Self-sufficient’ in terms of AI models and a Europe-based AI model may allure them significantly. One of the founders of Mistral is indeed a former French digital minister with close ties to Emanuel Macron. Last year when the European Union AI act was forcing Mistral to divulge its model. Mistral through its ties with president Macron developed a successful French-German coalition to veto such a requirement.
Mistral should benefit from “Second mover advantage”, now that all the world is obsessed with AI and LLM models. It must be much easier for them to penetrate into the market. Yet end users need more time in order to compare the two models and confirm if the the Mistral-Large is true competitor for GPT4 or not?
Source: Economist
