French startup Mistral launches an AI model that outperforms GPT-3.5

While most AI companies unveil their latest algorithms through press tours and blog posts, other companies seem content to fling their latest products into the digital ether like a pirate ship throwing its dead weight. One company that fits this last category is Mistral, a French AI startup that released its latest big language model, without any explanation, in a nondescript manner. Torrent link posted on X during the Weekend.

Mistral, which recently It raised $415 million In the financing round series, and now estimated Worth $2 billion, it impressed people with its fast, efficient LLM courses and fun, carefree hacker attitude. The company’s decision is made informally He falls Its latest show is inspired by memes and compliments on X, with One commenter Noting: “No blog, no hype, no description – just a torrent with model files… Mistral realizes that its primary audience is engineers and knows their cultural erogenous zones.”

On Monday, the company finally followed up its initial release with Blog post Which shared more details about the program, which is just called Mixtral-8x7B. According to the benchmarks in this blog post, the Mistral algorithm outperforms some of its US competitors, including Meta’s Llama 2 family and OpenAI’s GPT-3.5. People online Seems to agree The new Mistral algorithm is very good. A lot of people are currently bragging about how fast and fun the program is.

An added bonus is that Mixtral-8x7B is open source, unlike the ridiculously named OpenAI, which has kept the latest LLM certifications closed source and inspired A certain amount of backlash as a result of. In fact, Mistral It focuses on open sourcing all of its AI software, which puts it firmly on one side of the AI ​​industry’s growing culture war. Mistral AI co-founder and CEO Arthur Mensch Recently noticed On this decision, he noted that his company is committed to an “open, responsible, and decentralized approach to technology.”