Microsoft’s partnership with OpenAI may have been a strategic one to subvert regulatory scrutiny but it’s also one that frees the company up to explore its own large language model (LLM) development while competing with its own barrowed pre-generative technology.
According to a report from The Information, Microsoft is in the midst of developing its own homegrown LLM that could rival the performance of industry competitors such as Google’s Gemin model or even its partner OpenAI’s offering.
The Information cites to Microsoft employees that are familiar with the internal model dubbed MAI-1, claim that the company’s model could host around five hundred billion parameters, which would make it “far larger than any of the smaller, open-source models that Microsoft has previously trained.”
Over the past year, Microsoft has been spotted developing several Small Language Models (SLM) as less resource intensive and hyper-focused AI offerings to customers, but MAI-1 appears to be a direct attempt at creating its own ChatGPT or Llama 3 AI competitor to serve to developers soon.
Currently, Meta’s Llama 3 AI model utilizes seventy billion parameters while OpenAI’s GPT-4 model leverages over one trillion which would put Microsoft’s in-house endeavor squarely in the middle of the pack out of the gate if it were to release in its current form.
MAI-1 is being developed solely in-house and led by the recent company re-org with Mustafa Suleyman at the helm of the newly formed Microsoft AI business. While Microsoft may have considered current OpenAI CEO Sam Altman for the role of Microsoft AI CEO, it was DeepMind and Inflection co-founder Suleyman who landed the position.
Suleyman was quick to address possible “intellectual property” issues that could arise with the MAI-1 project by saying “it may build on training data and other tech from the startup,” but that’s as far as cross-development would go.
Where does the existence of MAI-1 leave the $12 billion investment in OpenAI?
Perfectly intact according to a recent statement from Microsoft CEO Satya Nadella who reiterated his confidence with OpenAI’s future with the company.
We will continue to build AI infrastructure inclusive of custom systems and silicon work in support of OpenAI’s foundation model roadmap, and also innovate and build products on top of their foundation models
Microsoft CEO, Satya Nadella
Further squelching the presumed zero-sum argument of AI development at Microsoft, the company continues to invest and partner with other large AI platforms in international regions to grow its AI presence while skirting those nasty little regulatory investigations.
