How to get Phi-3-Mini? Explained by Digimagg

Microsoft has introduced the Phi-3-Mini, a compact AI model designed for tasks like content creation, running locally on smartphones and laptops, catering to resource-constrained companies. Despite its small size and modest dataset compared to GPT-4, it promises superior performance, prioritizing data quality over quantity. With other tech giants also venturing into compact AI models, Phi-3-Mini is part of Microsoft's plan to release three such models, offering users affordability and functionality. Discover what users can anticipate from this smaller model and how to test it firsthand.

Microsoft unveils latest AI model: Phi-3-Mini

For a significant period, the prevailing notion was that bigger equated to better in the realm of artificial intelligence. However, while AI chatbots trained on massive datasets like Gemini and ChatGPT boast extensive applications, several major tech companies are deviating from this trend by introducing smaller, specialized models. Among them is Microsoft, which recently unveiled its most compact AI model to date.

Although its name may evoke thoughts of a sorority house, Phi-3 Mini is one of three compact large language models (LLMs) developed by Microsoft, with Phi-3 Small and Phi-3 Medium reportedly set for release in the near future. Despite being trained on a considerably smaller dataset compared to other LLMs, and with only 3.8 billion parameters—far fewer than GPT-4's 1.76 trillion—Microsoft asserts that Phi-3 Mini outperforms its predecessor Phi-2, released in December. The model's success can be largely attributed to its unique training approach, which heavily relies on filtered web data and synthetic data rather than real-world data acquired through processes like web crawling. Synthetic data, being more cost-effective, diverse, and easier to fine-tune than real-world data, has emerged as a logical input for many small AI models.

How to obtain Microsoft's Phi-3-Mini AI model

While traditional chatbots remain prevalent, Microsoft's latest micro-chatbot, Phi-3-Mini, is now accessible for public use.

Microsoft has recently integrated Phi-3 into its Azure model gallery on its cloud service platform, and the bot is also accessible on open-source model repositories like Hugging Face and Olloma. Future releases, such as Phi-3-Small and Phi-3-Medium, will also be available on these platforms.

If Microsoft's compact chatbot doesn't meet your needs or if you wish to compare it with competitors, refer to our guide on the top AI chatbots in 2024.

Is the future dominated by compact AI models?

Due to their reduced computational requirements, smaller models like Phi-3 Mini are typically more cost-effective to operate and demonstrate superior performance on devices such as phones and laptops compared to their larger counterparts. This makes them particularly attractive for companies with limited AI budgets, leveraging chatbots for specific applications such as content generation or problem-solving.

As compact AI models showcase distinct advantages over general-purpose ones, Microsoft is not the sole player prioritizing targeted chatbots at lower costs. Google recently introduced two language-focused chatbots, Gemma 2B and 7B, while Anthropic launched Claude 3 Haiku, designed to summarize complex research papers succinctly. Despite the growing popularity of smaller models, Microsoft isn't abandoning larger models entirely. The software giant emphasizes that models like Phi-3 are not intended to replace larger ones but rather to fulfill tasks that models like GPT-4 or Gemini cannot, such as local device execution.