By Reuters | Updated: May 07, 2024
May 6 (Reuters) – Microsoft (MSFT.O) is training a new, in-house AI language model large enough to compete with those from Alphabet’s Google (GOOGL.O) and OpenAI, the Information reported on Monday.
The new model, internally referred to as MAI-1, is being overseen by recently hired Mustafa Suleyman, the Google DeepMind co-founder and former CEO of AI startup Inflection, the report said, citing two Microsoft employees with knowledge of the effort.
The exact purpose of the model has not been determined yet and will depend on how well it performs. Microsoft could preview the new model as soon as its Build developer conference later this month, the report said.
Microsoft declined to comment when contacted by Reuters.
MAI-1 will be “far larger” than the previous smaller, open source models Microsoft had previously trained which means it will be more expensive, according to the report.
Microsoft last month launched a smaller artificial intelligence model called Phi-3-mini as it looks to attract a wider client base with cost-effective options.
The company has invested billions of dollars in OpenAI and deployed the ChatGPT maker’s technology across its suite of productivity software, allowing it to take an early lead in the generative AI race.
Microsoft has been setting aside large cluster of servers equipped with Nvidia’s (NVDA.O) graphic processing units along with large amounts of data to improve the model, according to the report.
MAI-1 will have roughly 500 billion parameters, the report said, while OpenAI’s GPT-4 is reported to have one trillion parameters and Phi-3 mini measures 3.8 billion parameters.
Microsoft tapped Suleyman in March as the head of its newly created consumer AI unit and hired several employees of Inflection.
The new model is not carried over from Inflection, although it may build on training data from the startup, the report added.
@ Thomson Reuters 2024
