Bigger is not always better: Microsoft is also launching a ‘mini-AI’ model

More and more tech companies are producing ‘light’ versions of their AI models. Now Microsoft is also jumping on the bandwagon with a mini-AI, Phi-3.

In the news: Phi-3 Mini will be the first of those smaller AI models that Microsoft launches.

• According to Eric Boyd, vice president at Azure, Microsoft’s AI platform, Phi-3 Mini will be “as powerful as Large Language Models such as ChatGPT-3.5 (the current free version of ChatGPT, ed.), but in a smaller size,” he explains The Verge.
• In addition to Mini, there will also be Small and Medium models. Each model has a different number of ‘parameters’, which indicate how many complex instructions it can handle. For the mini version this is relatively limited to 3.8 billion parameters. Phi-3 Mini is available for download via Azure, Hugging Face and Ollama.
• The Phi models were put together with a certain “curriculum,” according to Boyd. Thus, they were trained to produce children’s stories. “There aren’t enough children’s books, so we took a list of over 3,000 words and asked an LLM to create ‘children’s books’ to teach to Phi.”

Zoomed in: Why are tech companies bringing smaller AI models to the market?

• Boyd believes that smaller models like Phi-3 work better for specific enterprise applications because their internal data sets are smaller. And they are often more affordable, because these models use less computing power.
• Other companies preceded Microsoft. Google launched Gemma, a slimmed-down version of Gemini. Meta also released a small version of its Llama model and Claude 3 Haiku can quickly summarize scientific documents.
• In this way, tech companies are diversifying their AI models, and the perception that AI is cumbersome and heavy is being eliminated. Accessibility increases, because as an end user you know what you can use a particular AI model for.