Microsoft puts its energy into its new mini AI model Phi-2.

Microsoft Research has announced the release of Phi-2, a Small Language Model (SLM) that demonstrates impressive capabilities in terms of scale. Launched today, the model was first revealed at Microsoft's Ignite 2023 event, where Satya Nadella revealed that it was able to achieve state-of-the-art performance with a fraction of the training data.

Like GPT, Gemini, and other large language models (LLMs), SLM is trained on a limited dataset, using fewer parameters but requiring less computation to run. As a result, a model may not generalize as a large language model, but may be very good and efficient at specific tasks—such as mathematics and computations in Fig.

Phi-2, with 2.7 billion measurements, shows good reasoning and language comprehension, competing models up to 25 times, according to Microsoft. This is due to Microsoft's research focus on high-quality training data and advanced modeling techniques, resulting in a model that outperforms its predecessors on a number of criteria, including math, coding, and common sense.

“At just 2.7 billion parameters, the Phi-2 outperforms the Mistral and Lama-2 models by 7B and 13B parameters in various integrated benchmarks,” says Microsoft, giving a lowdown on Google's new AI model. Despite its smaller size, the 2 matches or surpasses the recently announced Google Gemini Nano 2.”

Gemini Nano 2 is Google's latest bet that a multi-modal LLM can run domestically. It was announced as part of the Gemini LLM family, which is expected to replace PaLM-2 in most Google services.

Microsoft's approach to AI, however, goes beyond model development. As reported by Decrypt, the introduction of custom chips, Maia and Cobalt, shows that the company is moving towards full integration of AI and cloud computing. Computer chips optimized for AI tasks support Microsoft's ambitious vision of combining hardware and software capabilities and are in direct competition with Google's Tensor and Apple's new M-series chips.

It should be noted that Phi-2 is a small language model that can be run on domestic low-end devices, even smartphones, opening the way for new applications and use cases.

As Phi-2 enters AI research and development, its presence in Azure AI Studio's model catalog is a step towards democratizing AI research. Microsoft is one of the most active companies contributing to open source AI development.

As the AI ​​landscape continues to evolve, Microsoft's Phi-2 is proof that the world of AI isn't always about thinking big. Sometimes, the greatest power lies in being small and smart.

Edited by Ryan Ozawa.