We might think that from the role of Open AI partner, Microsoft does not need to compete with the creator of GPTs, those specialized AI models that do not require cloud processing. But Phi-3 Mini is no rival, this Mini AI is actually an extra-simplified version, adapted for low-power devices such as smartphones and tablets.
Google has Gemini and OpenAI has Chat GPT. We’d be tempted to think that’s the whole story, except that AIs are a whole species, where Chat GPT falls into the category of “top predators”, GPTs can switch to “domesticated animals” and Phi-3… is kind of hamster that you can train. In other words, the level of “intelligence” varies depending on the application you need that AI for, with Microsoft’s announced solution filling a potentially very lucrative niche of general-purpose AI processing for things you might not you would have thought you needed AI.
Even though it could be described as an AI model brought to an almost embryonic stage, Phi-3 Mini still runs a language model with 3.8 billion parameters, being only the extra-simplified version of the Phi-3 Small model (7 billion parameters), respectively Phi-3 Medium (14 billion parameters).
As you go down the complexity scale the disadvantages are easy to understand: the AI model can perform less complex tasks, its scope becoming narrower. In return, you gain portability, the received tasks being completed with less consumption of resources (processor, memory, battery).
By comparison, the current cloud-processed version of GPT-4 has over a trillion parameters, and Meta’s small-scale model Llama 3 comes with 8 billion parameters.
According to Microsoft Azure AI Platform vice president Eric Boyd, the Phi-3 Mini is as efficient as models like the GPT-3.5, “just in a smaller form factor.” As a compact AI model, it should be cheaper and should also offer better performance on devices like laptops and phones.
To improve its performance, an LLM (large language model) created the AI equivalent of a “children’s book” from a list of 3,000 words to “educate” Phi-3, which has already improved its performance by becoming better at coding and reasoning.
In other words, the Phi-3 Mini is a mini AI developed/optimized with the help of another much more sophisticated AI so that it can work as efficiently as possible with as few resources as possible.
