Microsoft is developing its own large-scale language model MAI-1, which can compete with AI technologies from Google, Anthropic and OpenAI, despite investments in the latter.
Model MAI-1
New development by Microsoft
Microsoft is working on creating a large-scale language model called MAI- 1, which will potentially become a serious competitor to artificial intelligence developed by Google, Anthropic and OpenAI. This is happening despite the fact that the corporation invested $10 billion in OpenAI, the developer of ChatGPT, and received priority rights to use their products.
Project management
According to The Information, the development of MAI-1 is headed by Mustafa Suleiman - Former head of artificial intelligence at Google. He served as CEO of the startup Inflection, the intellectual property of which was acquired by Microsoft in March for $650 million. At the same time, the corporation also attracted part of the company's staff.
Model characteristics
Although MAI-1 can be based on methods developed in Inflection, Microsoft is positioning this product as an entirely new large-scale language model. According to two company employees, the model contains 500 billion parameters, which is significantly larger than previous Microsoft open source projects such as Phi-3. In terms of the number of parameters, MAI-1 approaches OpenAI's GPT-4, which is rumored to boast 1 trillion parameters, significantly ahead of competing models Meta and Mistral with 70 billion parameters.
Dual Approach to AI
Development of MAI-1 demonstrates Microsoft's dual approach to artificial intelligence : The company focuses on both small, local language models and more advanced, large-scale models running in the cloud. Apple is considering a similar approach. MAI-1 also signals Microsoft's willingness to explore artificial intelligence independently of OpenAI, whose technology powers the company's generative capabilities, including the Copilot assistant integrated into Windows.
Training and purpose of the model
The exact purpose of MAI-1 has not yet been determined and will depend on its performance. To train the model, Microsoft dedicated a large cluster of servers with Nvidia GPUs and collected training data from a variety of sources, including text generated by GPT-4 and public data from the Internet. It is expected that the announcement of MAI-1 could take place as early as this month at the Microsoft Build developer conference.
Glossary
- Microsoft is a transnational company that produces software for various computing devices.
- Google is an American transnational corporation, one of the leading companies in the field of information technology and the Internet.
- Anthropic is a company specializing in the development of artificial intelligence and advanced language models.
- OpenAI is a research company dedicated to the development and advancement of artificial intelligence.
- ChatGPT is a popular large language model based chatbot developed by OpenAI.
Links
Answers to questions
What is MAI-1 and what is it intended for?
What are the technical specifications of the MAI-1?
Who is leading the development of MAI-1?
How is MAI-1 related to OpenAI and other AI developers?
What data was used to train MAI-1?
Hashtags
Save a link to this article
Discussion of the topic – MAI-1: Microsoft's New $500 Billion AI Model to Combat ChatGPT
Microsoft is developing a powerful AI language model, MAI-1, with 500 billion parameters to compete with OpenAI's ChatGPT, in which it has invested $10 billion.
Latest comments
8 comments
Write a comment
Your email address will not be published. Required fields are checked *
Kseniya
Something interesting is happening at Microsoft! It looks like they are preparing a competitor for ChatGPT 🤔 And this is a model with a huge number of parameters - as many as 500 billion! Even more than GPT-4 OpenAI 🤯
Hans
Yeah, Microsoft isn't kidding. They clearly want to beat OpenAI in the race for the best large language models 💪 I wonder how they plan to use this MAI-1? Maybe in some of their products like Copilot?
Francois
It's not clear yet, but Microsoft says the MAI-1's purpose will depend on its performance. I think they just want to create a powerful base for future AI solutions 🧠 And yes, 500 billion parameters are serious! Many times more than Phi-3.
Enzo
I'm glad to see such competition between AI giants! 🙌 This should spur progress in large language models. Maybe we'll see even more advanced features than ChatGPT 😎
Zdzislaw
Don't get too excited ahead of time 😒 These large models require enormous computing power and training data. They may be too resource-intensive and expensive for mass use.
Sophia
But, as the article says, Microsoft is going to take a dual path: MAI-1 for the cloud, and smaller on-premises models for devices 💻 So they seem to have a strategy for different cases.
Sergio
I wonder how they will deal with issues of ethics and biases when training such large models 🤔 After all, it is already known that ChatGPT sometimes gives biased answers. With such a huge amount of data, this effect can only intensify 😕
Oksana
I think Microsoft takes AI ethics seriously given their 'Responsible AI Principles' 🧐 But you're right, this is a big problem for the industry. Either way, I'm looking forward to the MAI-1 announcement at Build this month! 🤩