Microsoft is reportedly training an in-house large language model called MAI-1. This is a big departure for the company, having previously focused on investing in OpenAI and using its models including the massive GPT-4 family of AIs to power the various Microsoft Copilot products.
The move would make sense as it will give Microsoft more control over the technology, and follows the hiring of Mustafa Suleyman, a former Google AI chief and founder of the Inflection AI lab. Despite Suleyman and his team joining Microsoft, the MAI-1 LLM is expected to be an entirely new model.
The report comes via The Information and follows Microsoft's sizeable investment of over $10 billion in OpenAI for the rights to use the company's AI models. This could, however, mean GPT-4 will no longer power Copilot, Microsoft's consumer-facing chatbot that's found on the Windows 11 desktop.
Microsoft hedges its bets on AI future
If the report is correct, Microsoft may be testing the waters by building two large language models at once.
While chatbot-style interfaces have risen in popularity, they have sizeable running costs because of the amount of data being transferred. There's also a question mark over the privacy of queries that are sent to external servers for processing.
While many, including Apple for its upcoming iOS 18 release, are working on processing the data on-device, the sheer size of LLMs (large is in the name, after all), makes this a tricky process.
By splitting its focus between a smaller-scale LLM and one more ambitious overarching one, Microsoft may be able to offer the best of both worlds.
