There are many AI tools out there, and each of them are great and powerful.
However, many of those AI tools require an internet connection, so they send users' request to their massive computing datacenter to be processed. But according to Microsoft, not all AI tools should run in the cloud because they should also run locally on phones and PCs.
In an attempt to pursuit this, Microsoft introduces Phi-3 mini.
According to Microsoft, this AI can compete with other AIs that are 10x more expensive to power and run.
It can even rival the popular web-based AIs, like OpenAI's ChatGPT-3.5.
We're excited to announce the launch of Phi-3, a groundbreaking family of small language models that outperform larger models on a range of benchmarks. Learn how these small language models trained on high-quality data are doing more with less: https://t.co/dFfyktuEUL
— Microsoft (@Microsoft) April 23, 2024
Microsoft's efforts to build better AIs is part of a growing trend that happens following the hype created by generative AIs.
With tech companies racing to create more powerful AIs, their focus tend to be on brute power, like the number of parameters and such. Less efforts were spent on developing AIs that are more efficient.
After all, in the the dizzying but lucrative race to build generative AI systems, the tech industry’s mantra has been bigger is better, no matter the price tag.
If a generative AI can be squeezed down in size to fit inside phones and PCs, and require no internet connection, the AI will definitely be a lot less powerful.
Smaller systems are less powerful, which means they are also less accurate, and may sound more awkward.
But Microsoft thinks that it may be a good trade-off, if customers can have them with a lot cheaper prices.
Customers imagine many ways to use AI, but with the biggest systems, "they’re like, ‘Oh, but you know, they can get kind of expensive,’" said Eric Boyd, a Microsoft executive.
Smaller models, almost by definition, are cheaper to deploy, he said.
To redefine what's possible with small language models (SLMs), Microsoft announced that it has built three different small models: Phi-3-mini, Phi-3-small and Phi-3-medium. Phi-3-mini.
The smallest of the Phi-3 models, the Phi-3 mini, can fit and run on the kinds of chips that power regular smartphones and computers, rather than more expensive processors made by Nvidia.
Microsoft hopes that with AIs that can run offline, more customers can use AIs in more places where bigger, more advanced models are too expensive to use, or cannot deploy its full capacity.
Whereas big projects could justify the costs of the larger, more precise A.I. systems, smaller projects may not need the same level of accuracy.
People who wish to write an essay, or summarize a long story, may not need a fully-fledged AI that connects to the internet whenever it needs to work, for example.
"Phi-3 is not slightly cheaper, it's dramatically cheaper," said Sébastien Bubeck, Microsoft's vice president of GenAI research.
Microsoft is betting that customers will be willing to forgo some performance if it means they can finally afford AI.
Building these models are a trade-off between power and size.