AI plans at Microsoft: smaller models and faster research

Microsoft's AI plans look set for a rosy future. And no wonder. The technology giant is currently at the top of the food chain, recently overtook its competitor Apple on the stock market for the first time and is always a talking point thanks to the constant further development of its software and, above all, artificial intelligence. Microsoft still has big plans for this artificial intelligence.

AI plans to accelerate research

Research is a constantly evolving process. Progress is moving faster and faster. In a way, this is also logical. Every invention, every discovery and every development accelerates further inventions, discoveries and developments. The discovery of fire opened up endless possibilities. When our ancestors invented the wheel, everything was quicker and easier. Today we have the internet, AI, infinite data sets, tools, machines and so on to help us develop things further.

Microsoft is currently fully committed to artificial intelligence and wants to use it to accelerate research many times over. 250 years of research are to be compressed into 25 years.

This is to be achieved with Azure Quantum Elements, a supercomputer paired with generative AI. With this combination of high-performance computing and machine learning, Microsoft wants to revolutionize science - "chemistry and materials science" to be precise, as the company itself writes.

These AI plans have already produced their first success. Within just one week, a new material for improving rechargeable batteries was found. A process that normally takes around 20 years. The AI suggested a lithium-sodium compound that could reduce lithium consumption in rechargeable batteries by up to 70 %.

Of course, all of this still needs to be developed and extensively tested, but so far Azure Quantum Elements has already saved 20 years of research. It remains exciting to see what else the supercomputer will produce in the future.

AI plans for smaller models with the same performance

Up to now, large language models (LLMs) have been the most popular in the field of generative AI. Microsoft has now set up its own team to bring the performance of LLMs to smaller AI models. In other words, smaller models with the same performance, as we already know from numerous areas of technology.

The Small Language Models (SLMs) are Microsoft's first step towards becoming more independent in AI development. To date, all AI-based applications and functions at Microsoft have been based on OpenAI and its LLM ChatGPT or the generative image AI DALL-E.

The idea behind Small Language Models is also to reduce energy consumption. LLMs consume a lot of energy in the form of water and electricity because the computing power is very high. ChatGPT consumes around one liter of water for 40 responses. This requires the most powerful processors on the market and there is a kind of battle for the valuable chips, which are not available in infinite numbers. SLMs would get by with less computing power and reduce the need for the powerful chips.

A few months ago, Microsoft showed with Phi-2 that an SLM can certainly keep up with or even outperform major competitors such as Llama 2-7B. So the way seems to be paved.

According to rumors, one reason for the development could also be that both the EU and the US authorities are not very enthusiastic about the cooperation between Microsoft and OpenAIs, as there are fears of a cartel emerging.

Microsoft Azure seminars from skill it

Microsoft Azure already has numerous AI-based applications in the cloud that you can familiarize yourself with via our courses.

In our training Develop Generative AI Solutions with Azure Open AI Services you will directly program solutions with generative AI.

Our three-day training goes into more detail Designing and Implementing a Microsoft Azure AI Solution in which you develop applications with AI in Azure using numerous practical exercises.

The courses require programming knowledge. You should also already be familiar with Azure. You will get the basics in Microsoft Azure AI Fundamentals.