Microsoft is engaged on a brand new large-scale AI language model known as MAI-1, which might probably rival state-of-the-art fashions from Google, Anthropic, and OpenAI, based on a report by The Information. This marks the primary time Microsoft has developed an in-house AI model of this magnitude since investing over $10 billion in OpenAI for the rights to reuse the startup’s AI fashions. OpenAI’s GPT-4 powers not solely ChatGPT but additionally Microsoft Copilot.
The growth of MAI-1 is being led by Mustafa Suleyman, the previous Google AI chief who just lately served as CEO of the AI startup Inflection earlier than Microsoft acquired nearly all of the startup’s workers and mental property for $650 million in March. Although MAI-1 may construct on methods introduced over by former Inflection workers, it’s reportedly a completely new massive language model (LLM), as confirmed by two Microsoft workers aware of the undertaking.
With roughly 500 billion parameters, MAI-1 shall be considerably bigger than Microsoft’s earlier open supply fashions (similar to Phi-3, which we lined final month), requiring extra computing energy and coaching information. This reportedly locations MAI-1 in an identical league as OpenAI’s GPT-4, which is rumored to have over 1 trillion parameters (in a mixture-of-experts configuration) and effectively above smaller fashions like Meta and Mistral’s 70 billion parameter fashions.
The growth of MAI-1 suggests a twin strategy to AI inside Microsoft, specializing in each small domestically run language fashions for cellular gadgets and bigger state-of-the-art fashions which are powered by the cloud. Apple is reportedly exploring an identical strategy. It additionally highlights the corporate’s willingness to discover AI growth independently from OpenAI, whose know-how presently powers Microsoft’s most bold generative AI options, together with a chatbot baked into Windows.
Reportedly, the precise objective of MAI-1 has not been decided (even inside Microsoft), and its most supreme use will rely upon its efficiency, based on one in all The Information’s sources. To prepare the model, Microsoft has been allocating a big cluster of servers with Nvidia GPUs and compiling coaching information from numerous sources, together with textual content generated by OpenAI’s GPT-4 and public Internet information.
Depending on the progress made within the coming weeks, The Information experiences that Microsoft may preview MAI-1 as early as its Build developer convention later this month, as reported by one of many sources cited by the publication.
