– Microsoft is developing a 500B parameter LLM called MAI-1 to rival GPT-4 and Google’s Gemini models.
– MAI-1 is the largest model Microsoft has deployed, led by Mustafa Suleyman, co-founder of DeepMind.
– Microsoft’s CTO Kevin Scott downplays competition with OpenAI, stating they both build big supercomputers to train AI models.
Microsoft is reportedly working on a 500B parameter LLM called MAI-1 in order to compete with GPT-4 and Google’s Gemini models. This model is significantly larger than Microsoft’s previous small language models, with the potential to rival some of the biggest models in the industry. Although details about the architecture of MAI-1 are still scarce, it is expected to be a powerful model if it is a dense model rather than a MoE model like GPT-4.
The development of MAI-1 is being led by Mustafa Suleyman, who left his position as the head of applied AI at DeepMind to co-found AI startup Inflection in 2022. Microsoft later acquired Inflection for $650 million, making MAI-1 a new project led by Suleyman and his team at Microsoft. The release date for MAI-1 is currently unknown, but Microsoft may provide a preview at its Build developer conference on 16 May.
Despite being the biggest investor in OpenAI, Microsoft’s development of MAI-1 has raised questions about the company’s commitment to its partnership with OpenAI. Microsoft’s CTO, Kevin Scott, downplayed concerns about competition with OpenAI, emphasizing the collaborative nature of their relationship. However, the potential release of MAI-1 could put Microsoft in direct competition with OpenAI, particularly if OpenAI were to release GPT-5 in the near future.
With the ongoing developments in the AI industry, including the postponed event by OpenAI and the mysterious appearances of GPT-2 chatbots, the competition between Microsoft and OpenAI continues to intensify. The release of MAI-1 and potential future models by both companies will likely shape the future of AI technology.