Microsoft Prepares New AI Model to Take on Google and OpenAI, Reports The Information
Microsoft is training a new AI language model, MAI-1, to rival Google and OpenAI. The purpose of MAI-1 is yet to be determined, but a preview may be showcased at the Build conference. MAI-1 will be larger and more expensive than previous models, with around 500 billion parameters.
The project, internally known as MAI-1, is under the supervision of Mustafa Suleyman, the co-founder of Google DeepMind and former CEO of AI startup Inflection. Two Microsoft employees familiar with the initiative provided this information.
The specific purpose of MAI-1 has yet to be determined and will depend on its performance. However, there are indications that Microsoft may unveil a preview of the model at its upcoming Build developer conference later this month. When contacted by Reuters, Microsoft declined to comment on the matter.
MAI-1 is expected to be significantly larger than the smaller open-source models previously trained by Microsoft. As a result, it will also be more expensive to develop, as per The Information's report. Last month, Microsoft launched a smaller AI model called Phi-3-mini, aiming to attract a broader range of clients with cost-effective options.
Microsoft has made substantial investments in OpenAI and has integrated the ChatGPT maker's technology into its suite of productivity software, giving it an early advantage in the generative AI race. The company has dedicated a significant number of servers equipped with Nvidia's graphic processing units and vast amounts of data to enhance the model, according to The Information.
Reportedly, MAI-1 will have approximately 500 billion parameters, while OpenAI's GPT-4 is said to have one trillion parameters. In comparison, Phi-3 mini measures 3.8 billion parameters.
In March, Microsoft appointed Mustafa Suleyman as the head of its newly established consumer AI unit and recruited several employees from Inflection. Although the new model is not a direct continuation of Inflection's work, it may leverage training data from the startup, the report added.
Microsoft is training a new AI language model, MAI-1, to rival Google and OpenAI.
The purpose of MAI-1 is yet to be determined, but a preview may be showcased at the Build conference.
MAI-1 will be larger and more expensive than previous models, with around 500 billion parameters.
Source: REUTERS