Source: Gelonghui
Microsoft's ambition to expand
Although it has invested more than $10 billion in OpenAI to become the “biggest financier,” Microsoft's ambitions in artificial intelligence are not satisfied with this.
According to reports, Microsoft is preparing to launch a new self-developed AI model, the “MAI-1,” which will become the largest Microsoft AI model so far. The new model will have about 500 billion parameters, which is large enough to compete with Google, Anthropic, and OpenAI.
Currently, the model has not been officially announced, and Microsoft may reveal it at the Build developer conference later this month.
MAI-1, Microsoft's ambition?
The development of the new model MAI-1 was led by Mustafa Suleyman (Mustafa Suleyman), CEO of Microsoft AI.
This Suleiman is quite talented. He was the co-founder of Google DeepMind. After DeepMind was acquired by Google, he left Google in 2022 and founded the artificial intelligence startup Inflection.
In March of this year, Inflection was bought by Microsoft for $650 million in most of its employees and intellectual property.
So the market speculates that MAI-1 may be based on Inflection's technology.
However, according to media quoting insiders, MAI-1 is a new large-scale language model (LLM) built by Microsoft, although it may use some of Inflection's training data and technology.
According to reports, MAI-1 has about 500 billion parameters, which is much larger than Microsoft's previous open source model, and is currently ranked among the top in the market.
Microsoft's Phi-3 Mini launched in March has only 3.8 billion parameters, while Meta's Llama 2 model has 70 billion parameters, and OpenAI's GPT-4 has 1 trillion parameters.
Although the parameters are lower than ChatGPT-4, this will also mean that Microsoft's reasoning costs are lower.
Additionally, Microsoft also has huge data resources and computing power, including large server clusters equipped with Nvidia graphics processing units.
These should play a critical role in training and supporting the development of MAI-1.
Crazy “throwing money”
Right now, the battle for dominance of the big AI model is in full swing.
Faced with “competition” from tech giants such as Google, Microsoft has also been frantically “throwing money” in the field of artificial intelligence.
In addition to investing heavily in OpenAI, Microsoft is also “casting a wide net” of AI unicorns.
Looking back, since 2019, Microsoft's investment in OpenAI has exceeded 13 billion US dollars. In February of this year, Microsoft reached an in-depth cooperation with the French open source AI startup Mistral AI, and in March it also acquired Inflection AI, etc.
Currently, Microsoft is a big investor in OpenAI and the French startup Mistral, so why create a new model from scratch?
Microsoft is probably betting on both sides.
Currently, Microsoft is now pursuing the dual-track development of AI, aiming to develop both small language models that focus on mobile devices running locally and large-scale, cutting-edge models supported by the cloud.
This also means that Microsoft is willing to reinvent a new path in the field of AI independently of OpenAI.
Although the exact use of MAI-1 has yet to be determined, Microsoft has not responded to this.
However, Microsoft Chief Technology Officer Kevin Scott (Kevin Scott) said on social media on Monday that OpenAI uses supercomputers built by Microsoft to train artificial intelligence models, then both companies applied them to products and services, and Microsoft Research and the company's product group also built artificial intelligence models.
He said that almost every product, service, and operation process of Microsoft uses artificial intelligence models, and teams that make and operate products sometimes need to do their own customization work, whether it's training a model from scratch or fine-tuning the model that someone else has already built. There will be more things like this in the future. The names of some of these models include Turing and MAI.
What is clear is that the imminent launch of MAI-1 represents another step forward for Microsoft and marks a critical moment in the ongoing artificial intelligence arms race.