Microsoft says its open-source Phi-4 AI model outperforms Gemini 1.5 Pro
Microsoft released its Phi-4 model for artificial intelligence (AI) on Friday. The company’s latest small language model (SLM) joins the open-source Phi family of foundational models. The AI model comes eight months after the release of Phi-3 and four months after the introduction of the Phi-3.5 series of AI models. The tech giant claims that the SLM is better able to solve complex, reasoning-based questions in areas such as mathematics. Moreover, it is also said to excel at conventional language processing.
Microsoft’s Phi-4 AI model will be available through Hugging Face
So far, every Phi series has been launched with a mini variant, but the Phi-4 model was not accompanied by a mini model. Microsoft, in one blog posthighlighted that Phi-4 is currently available on Azure AI Foundry under a Microsoft Research License Agreement (MSRLA). The company plans to make it available on Hugging Face next week as well.
The company also shared benchmark scores from its internal testing. Based on this, the new AI model significantly improves the capabilities of the older generation model. The tech giant claimed that Phi-4 outperforms Gemini Pro 1.5, a much larger model, on the benchmark for mathematical competitiveness problems. It also published detailed benchmark performance in a technical document published in the online journal arXiv.
On the security side, Microsoft stated that the Azure AI Foundry comes with a suite of capabilities to help organizations measure, mitigate, and manage AI risks across the entire development lifecycle of traditional machine learning and generative AI applications. Additionally, business users can use Azure AI Content Safety features such as prompt shields, orientation detection, and others as a content filter.
Developers can also add these security capabilities to their applications through a single application programming interface (API). The platform can monitor applications for quality and security, adversarial prompt attacks and data integrity and provide developers with real-time alerts. This will be available to the Phi users who access it through Azure.
Smaller language models in particular are often trained on synthetic data after implementation, allowing them to quickly acquire more knowledge and higher efficiency. However, the results after training are not always consistent in real-world situations.
