In artificial intelligence, the race for the most efficient models to catch up with OpenAI

In artificial intelligence, the race for the most efficient models to catch up with OpenAI
In artificial intelligence, the race for the most efficient models to catch up with OpenAI

When we lift the hood of an AI to check the mechanics that drive it, we sometimes have surprises. Microsoft’s Copilot, for example, does not run with Phy, the in-house machinery, but with ChatGPT, developed by OpenAI. At the end of 2022, the shattering arrival of this chatbot pushed even Google, whose laboratories nevertheless contributed enormously to the advances in generative AI in the 2010s.

Read also | Article reserved for our subscribers AI: how technology is gradually integrating our personal and professional lives

Add to your selections

That Microsoft trusts OpenAI is not surprising: the software giant has invested 13 billion dollars (around 12.1 billion euros) in it. ChatGPT has other prestigious partners who demonstrate the esteem of the sector, starting with Apple, which chose it to co-lead its future Apple Intelligence, alongside other in-house AI.

At this time, the gap is large between the best generative mechanics, which bear the scholarly name of Large Language Models (LLM), and their pursuers. The performance of these LLMs is scrutinized, dozens of tests measure their comprehension, logic, translation abilities, their speed, etc. Not to mention their error rates.

A more empathetic tone

In the tech world, a race is underway to catch up with OpenAI. Google has worked hard and its Gemini AI is now one of the best. Meta recently revealed that in 2023 its investments in AI amounted to approximately $35 billion. And start-ups, which are increasing their fundraising, are not left out.

Many of the LLM publishing companies that stand out in the tests are based in California. Antropic, Claude’s publisher, raised $4 billion from Amazon in 2023. Reka, which created Core, received $60 million in funding the same year when Palmyra, published by Writer, raised $100 million. dollars. Inflection, the creator of Pi, an AI with a more empathetic tone and personalized responses based on the interests of each user, raised $1.3 billion a year ago.

In France, the young company Mistral recently raised 600 million euros from French and American investors, valuing the company at 5.8 billion euros. Its LLMs are well recognized by international AI researchers and rank high in rankings.

Write drafts

We also find one of the Mistral models under the hood of the chatbot popular American Perplexity, which uses it alternating with Meta’s LLM, named Llama. Two models chosen because they are open source: certain companies can reuse and modify them freely. Llama has also made dozens of babies, even as far as China, reveals the New York Times.

You have 18.36% of this article left to read. The rest is reserved for subscribers.



PREV Google Pixel 9 Pro Fold officially arrives in India
NEXT a carbon fiber architecture to be lighter than a Galaxy S24 Ultra