- The Paris-based firm made headlines in June when it raised USD 113 million in a venture fundraising round just four weeks after it was formed.
- Like its competitors OpenAI and Google, Mistral AI creates large language models, a type of artificial intelligence that drives chatbots like ChatGPT and Gemini.
The renowned venture capital firms Andreessen Horowitz and Lightspeed Venture Partners spearheaded the deal that raised 385 million euros, or around USD 415 million, for the French generative artificial intelligence startup Mistral AI, which announced that it had closed on its second record-breaking investment round of the year.
The most recent funding round, which is estimated to have increased Mistral AI’s valuation to almost USD two billion, is another indication that investors are unlikely to find enough potential generative AI businesses to satisfy their hunt anytime soon.
The Paris-based firm made headlines in June when it raised USD 113 million in a venture fundraising round just four weeks after it was formed. Since then, its valuation has increased more than seven times in six months. Each of its three co-founders has a long history in the AI sector. While Arthur Mensch formerly worked at DeepMind, an AI research lab that has been a part of Google LLC since 2014, Timothée Lacroix and Guillaume Lample were researchers at Meta Platforms Inc.’s Paris AI Lab.
Like its competitors OpenAI and Google, Mistral AI creates large language models, a type of artificial intelligence that drives chatbots like ChatGPT and Gemini. With its remarkable capacity to have human-like discussions with users on nearly any topic, ChatGPT famously grabbed the internet by storm towards the end of last year.
Mistral AI takes a different tack, though. It is adamant that generative AI technologies ought to be open source, which implies that anybody is welcome to use and alter the code that powers its LLMs. The business hopes to provide other users with the means to construct their personalized chatbots rapidly by adopting this open-source methodology.
In the past, OpenAI and Google have warned against doing things this way, claiming that the underlying LLMs might be misused and turned into instruments that propagate malware and other harmful content.
Anjney Midha, partner at Andreessen Horowitz, stated, “We just believe A.I. should be open.” He contends that practically every other technology field, such as databases, computer operating systems, programming languages, and more, has adopted an open-source model.
It is thought that OpenAI, Google, and Microsoft Corp. are at the forefront of the generative AI development race, having invested billions of dollars in training models like GPT-4 and Gemini. These models can compose tests and poems, build original computer code, and respond to queries. They are trained on vast volumes of material and data from the internet.
Some contend that making such potent models available in open-source space is extremely risky. Notably, Google and OpenAI have dedicated months to creating safety nets around their LLMs, making sure they can’t be used, for instance, to produce biased search results or propagate hate speech and misinformation.
Some, on the other hand, think that firms that provide the technology for free will eventually win the crown of generative AI, even if it means taking the chance of open sourcing it without many of the same restrictions.
Holger Mueller, Analyst of Constellation Research Inc., mentioned this as one of the detailed discussions related to AI development. “Microsoft, Google, and OpenAI all believe AI development should be closed, whereas the likes of Mistral AI believe it should be open source. There are pros and cons to both approaches, but in most other aspects of the technology industry, such as infrastructure-as-a-service and platform-as-a-service, open source has emerged as the clear winner,” he said.
However, this will only be the case for a while because, in the extent of SaaS, most of the leading platforms continue to be proprietary software. “So, we can’t be sure which approach will win for AI,” Mueller added.
With the release of its well-liked Llama 2 model earlier this year, Meta may be the leading developer of open-source LLMs at present. Anyone can build chatbots tailored to perform distinct jobs by fine-tuning the basic Llama 2 model using their own data.
Mistral AI’s offering is comparable. It announced that Mistral 7B, its first-ever LLM, was available in September. With only seven billion parameters, it’s a comparatively tiny LLM. However, the business says it can outperform many of its larger competitors by processing and producing responses more cheaply and efficiently.
According to Mistral AI’s ambitious agenda, further frontier models will be developed to answer queries and summarize data.
Chief Executive Officer of Mistral AI, Mensch, claims that the business has created a more effective and economical training program for its LLMs. He claims its models can function for less than half the price of the greatest LLMs from Google or OpenAI.