Microsoft would be developing a smaller and cheaper artificial intelligence model. The company seeks to offer an alternative to GPT, but without the hardware and maintenance requirements that this technology implies. The long term goal is build an AI that is at the level of what OpenAI offers and other companies.
According to a report from The Information, Microsoft formed a team to develop conversational artificial intelligence which requires less computing power than ChatGPT. The new division, known as GenAI, will be led by Misha Bilenko, former director of Azure AI. The tech giant will tap into its most experienced AI developers and redeploy them to GenAI starting this month.
The task of Bilenko and his team will be to build a small language model (SLM) optimized to consume fewer resources. Microsoft wants to create an AI that equals (or surpasses) that of OpenAI, although without the hardware requirements it represents.
Unlike large language models (LLMs), such as GPT-4 or Gemini, SLMs have fewer parameters and can run on mobile devices. Another characteristic of small language models is that they require less data to train them. Currently, Microsoft has two models of this type: Phi 2 and Orca 2, the latter based on Llama 2.
The technology company wants to reduce the carbon footprint represented by large models. A study conducted by researchers at the University of California found that LLM training can cost up to $100 billion and produces the same amount of CO₂ emissions as New York City for a month.
The first evidence that Microsoft was working on this model emerged in September 2023.
Microsoft wants to reduce its dependence on OpenAI
The idea of developing a small language model is another step by Microsoft in its strategy for consolidate its commitment to artificial intelligence. Although it made a multi-million dollar investment in OpenAI, the technology giant does not want to depend exclusively on this technology.
In an interview with Bloomberg During the Davos forum, Satya Nadella stated that Their products are not based on a single model. Microsoft’s CEO mentioned that they will invest in models with different capabilities, although he clarified that there is a very deep link with OpenAI.
“I feel like we are very capable of controlling our own destiny. We care about having the best frontier model, which today is GPT-4,” said Nadella. “But we also have Mixtral on Azure as an as-a-service model. We use Llama in some places. We have Phi, which is the best SLM from Microsoft.”
Partnership between Microsoft and OpenAI has come under scrutiny by regulators in various countries. The UK Competition and Markets Authority (CMA) said in December that it closely monitors the impact of alliances on the artificial intelligence industry. The CMA fears that this alliance will weaken competition in the development of language models.
On the other hand, the FTC opened the door to an investigation to determine whether Microsoft’s investment in OpenAI is subject to the United States Merger Act. Like the CMA, the US authority considers that we would be facing a merger scenario, although Sam Altman’s company is a non-profit organization, so it is not subject to these laws.
The latest to join the table is the European Commission, which called on all those involved to comment on whether the association between OpenAI and Microsoft threatens competition.