Microsoft prepares its own LLM-type AI, MAI-1! It could become the engine of the chatbot co-pilotso replace it GPT-4 ofOpenAI with whom the company has established a partnership since the launch of ChatGPT ! Is there water in the gas between the two companies?
OpenAI’s main partner, Microsoft is also one of the first investors from the Californian startup. This made it possible to integrate AI such as ChatGPT Or WEAK to Windows, Bing or even Office 365.
The American company enjoys a preview or even exclusive access, to the various OpenAI models. Also, a recently leaked email revealed that the purpose of this alliance was to dethrone Google.
This is the main reason for that Microsoft has invested tens of billions of dollars in the beginning Sam Altmanhoping regain his crown as digital lord.
All different products from the Microsoft Copilot range are based on GPT-4, especially the chatbot incorporated in Windows 11. However, this deal could finally end in the near future…
MAY-1: an AI created directly by Microsoft
According to The information, Microsoft has started training its own AI type LLM (large language model) titled MAI-1 that could be launched during the Build conference on May 16, 2024.
More precisely, two distinct models are currently under construction. The project is the supervision of Mustafa Suleymanthe former director by Google DeepMind and founder of the laboratory Inflection HAVE.
The latter joined the company last March with their team, through a The $625 million contract. Training data and other elements from Inflection AI will be used for MAI-1.
The dataset contains other types of information, including text generated by GPT-4 and content from the web. To carry out the training, Microsoft will use a large cluster of equipped servers graphics cards Nvidia trendy.
If the source is not wrong, MAI-1 would have 500 billion parameters. This makes it significantly larger than GPT-3.5 and its 175 billion parameters, but less than GPT-4 Or Claude anthropic 3 which are close to 2 billion.
However, we do not know if it is the larger of the two models in preparation, or on the contrary the smaller. At the end of April 2024, Microsoft presented Phi-3 Mini: an LLM of only 3.8 billion settings, but able to exceed models 10 times wider.
A change of strategy after the agreement with OpenAI?
Regardless, create your own LLM MAI-1 for Copilot offers more control of Microsoft over the technology. It would therefore be a wise strategic choice.
This would allow in particular to reduce massive costs related to Data Centers on which GPT is based, and to avoid being immersed in OpenAI controversies such as respect for copyright, confidentiality or creating a dangerous AGI…
These controversies pushed Microsoft do you want to break your agreement with OpenAI? There is no information to be sure at the moment, but everyone can draw their own conclusions…
And what do you think? Is this a good strategic choice on Microsoft’s part? Or should I continue to rely on OpenAI’s AI? Express your opinion in comments!
- Share the article:
Our blog is powered by readers. When you shop through links on our site, we may earn an affiliate commission.