Microsoft introduces a series of "small language models" with Phi-3

The Phi-3 consists of a series of "small language models," or SLMs. These models offer many of the same capabilities as large language models, but with one important difference: they require less computational power and data.

The Phi-3-mini model, for example, outperforms models twice its size, Microsoft said.

SLMs are not meant to replace large models, but offer new opportunities for businesses and devices where size matters. 

Large models cost hundreds of millions to billions to train and are expensive to run. SLMs could significantly reduce these costs and provide new options for AI developers.

It also enables local AI on mobile devices. Apple researched that small, well-trained models can run on devices and perform as well as large models.

This means that AI may start to become an invisible part in tools and devices we use every day, which will start to accelerate wider adoption of AI.

Graphic illustration showing how the quality of the new Phi-3 models as measured by the Massive Multitask Language Understanding (MMLU) benchmark compares to other similarly sized models. (Image provided by Microsoft)