Microsoft has recently released a new lightweight AI language model called Phi-3 mini, following the successful launch of their Phi-2 model in late 2023. This latest edition boasts 3.8 billion parameters trained on 3.3 trillion tokens, providing similar performance to competitor products.

To develop the Phi-3 mini model, Microsoft employed a scaled-up dataset training approach with heavily filtered web and synthetic data. The company used this approach to build on the foundation set by Phi-2, resulting in a powerful language model that can handle massive multitask language understanding (MMLU) tasks and multi-turn benchmarks with ease.

The Phi-3 mini model achieved a rating of 69 per cent in MMLU and 8.38 in MT benchmarks, demonstrating its advanced capabilities. Despite its size limitations for certain tasks and lack of capacity to store extensive factual knowledge, the model is still an impressive achievement for Microsoft’s responsible AI development efforts.

Microsoft emphasized that Phi-3 mini was developed in alignment with their responsible AI principles, with input from internal teams helping to curate additional datasets tailored to address specific insights. This approach resulted in significantly reduced harmful response rates, further demonstrating Microsoft’s commitment to ethical AI development.

The Phi-3 mini model is currently available through Microsoft’s Azure cloud platform, providing users with access to its advanced features and capabilities. With its powerful language processing capabilities and ethical development practices, it is sure to be a valuable tool for businesses looking to improve their AI applications.