According to The Information Beijing time tonight,MicrosoftA new AI model is being trained in-house at the company, at a scale sufficient to compete with state-of-the-art models from Google, Anthropic, and even OpenAI itself.
The new model is reportedly codenamed "MAI-1”, led by Mustafa Suleyman, the former Google AI leader. Before Mustafa joined Microsoft, he served as CEO of the startup Inflection.Until March this year, Microsoft paid $650 million (IT Home Note: currently about 4.687 billion yuan) to purchase the company's property rights and hired most of its employees..
Image source: Pixabay
Microsoft's "MAI-1" may be based on Inflection's training data and other technologies, and according to two Microsoft employees familiar with the matter,Microsoft's new model is a different project from Inflection's original model Pi..
Relevant sources revealed that the scale of MAI-1 is "far larger" than any small open source model Microsoft has previously trained, which means it will require more computing power and training data, and will also have higher costs.MAI-1 will have approximately 500 billion parameters or settings, which users can adjust to determine what the model learns during training. For comparison, GPT-4 has over 1 trillion parameters, and smaller open source models released by companies like Meta and Mistral have 70 billion parameters.