According to the report, Microsoft is said to be training an in-house AI language model that can compete with large-scale models developed by Alphabet's Google and OpenAI. The model is said to be overseen by former Googler Mustafa Suleyman, who joined Microsoft earlier this year from Inflection AI as CEO of its AI division.
The new model is being internally referred to as MAI-1, and the exact purpose of the model has not yet been determined, The Information reports. Microsoft may also preview the new model at the Build developer conference later this month, the report said.
“This is a Microsoft model and is not inherited from Inflection. However, it may be built on training data, or it may not be a model inherited from Inflection, but may be built on training data or It may also be built on other technologies from startups,” the report states. This is different from the model Inflection previously released, the publication said, citing two Microsoft employees with knowledge of the effort.
Why this development matters
This is reportedly the first time Microsoft has trained a model in-house since investing more than $10 billion in OpenAI.
“For the first time since investing more than $10 billion in OpenAl in exchange for the right to reuse the startup’s Al models, Microsoft is training a new in-house AI model large enough to compete with state-of-the-art models. “From Google, Anthropic, and OpenAl itself,” the report said.
Microsoft has been using the Large Language Model (LLM) developed by OpenAI.
The report also states that MAI-1 will be “much larger than the smaller open source models that Microsoft has previously trained.” This means AI models require more computing power and training data, which can be costly.
“MAI-1 has approximately 500 billion parameters or settings that can be adjusted to determine what the model learns during training,” the report states.
Meanwhile, OpenAI's GPT-4 has over 1 trillion parameters, and smaller open source models released by companies like Meta Platforms and Mistral have 70 billion parameters.
The new model is being internally referred to as MAI-1, and the exact purpose of the model has not yet been determined, The Information reports. Microsoft may also preview the new model at the Build developer conference later this month, the report said.
“This is a Microsoft model and is not inherited from Inflection. However, it may be built on training data, or it may not be a model inherited from Inflection, but may be built on training data or It may also be built on other technologies from startups,” the report states. This is different from the model Inflection previously released, the publication said, citing two Microsoft employees with knowledge of the effort.
Why this development matters
This is reportedly the first time Microsoft has trained a model in-house since investing more than $10 billion in OpenAI.
Expanding
“For the first time since investing more than $10 billion in OpenAl in exchange for the right to reuse the startup’s Al models, Microsoft is training a new in-house AI model large enough to compete with state-of-the-art models. “From Google, Anthropic, and OpenAl itself,” the report said.
Microsoft has been using the Large Language Model (LLM) developed by OpenAI.
The report also states that MAI-1 will be “much larger than the smaller open source models that Microsoft has previously trained.” This means AI models require more computing power and training data, which can be costly.
“MAI-1 has approximately 500 billion parameters or settings that can be adjusted to determine what the model learns during training,” the report states.
Meanwhile, OpenAI's GPT-4 has over 1 trillion parameters, and smaller open source models released by companies like Meta Platforms and Mistral have 70 billion parameters.
