Microsoft is developing a new AI language model called MAI-1, potentially challenging leading models from Google, Anthropic, and OpenAI, reports The Information.
This venture into in-house AI development follows Microsoft’s $10 billion investment in OpenAI, whose GPT-4 model underpins Microsoft’s products like ChatGPT and Copilot. Led by Mustafa Suleyman, former Google AI head and CEO of Inflection, Microsoft took a significant step by acquiring most of Inflection’s team and IP for $650 million. MAI-1, an original large language model distinct from previous projects, incorporates techniques from the Inflection team and is expected to have around 500 billion parameters.
This size positions it close to GPT-4 and far exceeds smaller models like those from Meta and Mistral. Microsoft’s strategy includes smaller models for mobile devices and advanced cloud-based models, similar to Apple’s approach. The use of MAI-1 will be decided based on its performance, with Microsoft currently preparing it using a substantial server cluster and diverse training data. A preview of MAI-1 could be unveiled at Microsoft’s Build conference later this month, according to The Information.
