Why is Artificial Intelligence Getting Smaller? (And Why That's Good News)

When we think of Artificial Intelligence (AI), we often imagine giant supercomputers and processes that consume massive amounts of energy. Famous programs like ChatGPT work this way: they are “Large” models that live in the cloud.

However, this year, a new trend is emerging: Small Language Models (SLMs). Essentially, this is about shrinking the “brain” of the AI to make it more efficient, faster, and more private.

1. What is a “Small” Model?

Imagine a large model is a 20-volume encyclopedia that knows a little bit about everything, but to consult it, you have to travel to a distant library. A small model is like a pocket handbook: it doesn’t know everything in the world, but it is an expert in specific topics and you can carry it with you everywhere.

These models are designed to perform specific tasks very well, rather than trying to know everything at once.

2. The Advantages of “Miniature” AI

Why would we want a smaller AI? Here are the three main reasons:

• It works on your phone: Because they are small, these models can run directly on your smartphone or laptop without needing an internet connection.
• Total Privacy: Since the AI lives inside your device, your questions and data never leave. They aren’t sent to any company or travel through the cloud.
• Speed: Because the data doesn’t have to “travel” over the internet to a distant server, the answers are almost instantaneous.

3. Where will we see them?

We are already starting to use them without realizing it. Some examples include:

• Personal Assistants: That can help you organize your schedule without reading your emails in the cloud.
• Real-time Translators: That work even if you are in the middle of a forest with no signal.
• Document Summaries: You can ask your computer to summarize a private file with the peace of mind that no one else will see that content.

4. The Future is Hybrid

This doesn’t mean large models are going to disappear. In the future, we will use giant models to solve complex scientific problems, but we will use small models for our daily lives, looking for speed and, above all, ensuring our personal information stays with us.

To Learn More

If you want to explore these “mini” models further, you can look for these names:

• Phi-3: A family of small models created by Microsoft that are surprisingly intelligent.
• Gemma: The lightweight version of Google’s AI technology.
• Hugging Face: A community where you can try many of these models for free.