A new large-scale AI language model is being developed by Microsoft to compete with cutting-edge technology from OpenAI and Google.
The model, informally named MAI-1, is much larger than its predecessor, according to a report published by The Information, according to Mustafa Suleyman, co-founder of Google DeepMind and former CEO of AI startup Inflection. It is said to be under the direction of.
Microsoft has not officially commented on the report, but Microsoft CTO Kevin Scott posted on LinkedIn to discuss the work the tech company is doing with AI models.
“Our partner OpenAI uses these supercomputers to train models that define the frontier, and we both use these supercomputers so that many people can then benefit from them. We're making the model available for use in products and services. We rather like this arrangement,” Scott said.
It is not yet clear what this model will be used for or when it will be fully deployed. Microsoft is reportedly working on its own large-scale AI model, with the tech company investing around £10 billion ($13 billion) in AI company OpenAI.
In recent months, Microsoft has launched a relatively small AI model called the Phi-3 mini, but the MAI-1 model is expected to be much larger and more expensive. According to the report, MAI-1 has about 500 billion parameters, compared to nearly 1 trillion parameters in OpenAI's GPT-4. Phi-3 mini has 3.8 billion parameters.
“Each supercomputer we build for Open AI is much larger than the previous supercomputer, and each frontier model that is trained is much more powerful than the previous supercomputer,” Scott said. Masu.
“We will continue on this path, building increasingly powerful supercomputers for Open AI and training models that will lead the entire field into the future. The impact of our collaboration There is no end in sight to the increase.”
Microsoft announced last month that it plans to more than double the capacity of its existing data centers this year to meet the expected surge in demand due to cross-industry adoption of artificial intelligence (AI). According to a leaked presentation from Microsoft, the company is gearing up to double its new construction capacity in the second half of 2024, and aims for triple growth in the first half of 2025.
