Microsoft is reportedly building a large new AI model that could potentially compete with OpenAI and Google’s most advanced tech.
As first reported by The Information, Microsoft is developing an in-house AI language model internally known as MAI-1. This model is allegedly sophisticated enough to rival those produced by Google, Anthropic, and Microsoft’s prominent AI partner, OpenAI.
The Information’s sources said that MAI-1 will be overseen by the recently hired Mustafa Suleyman, the Google DeepMind co-founder and former CEO of AI startup Inflection, who is now known as Microsoft AI CEO.
The precise plan behind the model has not yet been determined and will seemingly depend on its initial reception and performance. The Information added that Microsoft may provide a preview of the new model as early as its Build developer conference this month.
Last month, Microsoft introduced a smaller AI model named Phi-3-mini, aspiring to appeal to a broader client base with cost-effective options. In contrast, the MAI-1 will be “considerably larger” than the smaller, open-source models Microsoft previously trained, meaning it will likely also be more costly.
According to The Information, MAI-1 will consist of approximately 500 billion parameters, while OpenAI’s GPT-4 is reported to have one trillion parameters, and Phi-3 mini measures 3.8 billion parameters.
Microsoft AI and Suleyman
Suleyman took the helm as Microsoft AI’s CEO in March.
Suleyman oversees Microsoft’s consumer—and business-focused AI products and services, including Copilot, Bing, and Edge. As an Executive Vice President (EVP) at Microsoft, Suleyman joins the senior leadership team and reports directly to CEO Satya Nadella.
In addition to hiring Suleyman, Microsoft has recruited several employees from Inflection AI, including Hoffman and co-founder Karén Simonyan. Nadella praised Simonyan as “a renowned AI researcher and thought leader” and appointed him as the chief scientist of Microsoft AI.
Last month, Microsoft also hired former Meta executive Jason Taylor for its AI supercomputing team. Taylor will assume the role of Corporate Vice President and Deputy CEO at Microsoft.
Also in April, Microsoft announced it was opening an AI hub in London. AI scientist Jordan Hoffmann, formerly of AI organisations Deepmind and Inflection, will lead the new AI hub and will be joined by a group of Microsoft AI team members based in its London Paddington office.
Despite several prominent hires from Inflection, the new MAI-1 model is not inherited from Inflection, though, as The Information reported, it may leverage training data from the startup.
Microsoft and OpenAI
Given Microsoft has invested roughly $13 billion in AI vendor OpenAI over the last five years and has a nonvoting position on the organisation’s board, there was inevitably some speculation about what it means that Microsoft is working on its own in-house models that could meaningfully compete with its partner’s.
Although Microsoft declined to comment to Reuters on the story, Microsoft CTO Kevin Scott took to LinkedIn to explain the tech giant’s position.
“I’m not sure why this is news, but just to summarise the obvious: we build big supercomputers to train AI models; our partner Open AI uses these supercomputers to train frontier-defining models; and then we both make these models available in products and services so that lots of people can benefit from them,” Scott said. “We rather like this arrangement.”
We also, for years and years and years, have built AI models in MSR and in our product groups. AI models turn out to be interesting things to work on, and our researchers do great work studying and building them.”
Last week, emails published in the US Justice Department’s antitrust case against Google suggested that Microsoft‘s investment in OpenAI was motivated by concerns about Google’s superior progress in AI.
As first reported by Business Insider, the Justice Department’s investigation triggered the release of an internal email between Microsoft Co-Founder Bill Gates, CEO Satya Nadella, and Scott.
In the June 2019 email titled “Thoughts on OpenAI”, the investment opportunity in the AI organisation was outlined, emphasising the areas where Google had a significant edge over Microsoft in AI research and models.
While Scott initially admits to being “dismissive” of OpenAI and Google DeepMind’s “game-playing stunts”, potentially referring to the latter’s AlphaGo Zero demonstrations showcasing AI’s potential, he became concerned upon realising the advanced state of competitors’ natural language processing (NLP) models.
from UC Today https://ift.tt/j6ZSgvT
0 Comments