Microsoft creates new AI model to compete with Google, OpenAI
According to an article, Microsoft is developing a new internal AI language model that is big enough to take on those from Alphabet's Google and OpenAI. According to the story, which cited two Microsoft workers with knowledge of the project, Mustafa Suleyman, the newly hired co-founder of Google DeepMind and former CEO of AI firm Inflection, is in charge of the new model, internally known as MAI-1.
Highlights:
- Microsoft is competing with Google and OpenAI by creating the MAI-1 AI language model.
- MAI-1, led by Mustafa Suleyman, is expected to be significantly larger than previous models, with around 500 billion parameters.
- This move reflects Microsoft's ambition to strengthen its position in the AI market and leverage its investment in OpenAI.
The specific purpose of the model remains unclear and will be determined by its effectiveness. The source claims that Microsoft could introduce the new gadget later this month at its Build programming conference.
Because MAI-1 will be "far larger" than the prior, smaller, free software models that Microsoft had earlier trained, the study asserts that it will cost more.
Microsoft released Phi-3-mini, a more reasonably priced AI model, just recently in an attempt to appeal to a wider audience.
Due to its billion-dollar stake in OpenAI and the integration of ChatGPT maker's software throughout its productivity application portfolio, the firm has jumped ahead in the race for generative AI.
According to the source, Microsoft has been allocating a sizable cluster of computers with Nvidia graphics processing units and a sizable quantity of data to enhance the model.
According to the study, MAI-1 is expected to contain about 500 billion parameters, whilst GPT-4 from OpenAI is supposed to have one trillion parameters, and Phi-3 tiny measures 3.8 billion parameters.
Microsoft hired Suleyman to head its newly formed consumer AI branch in March and added other Inflection employees.
The study also said that while the new model could draw from starting training data, it is not carried over from Inflection.