Understanding the Value of AI

People often talk about artificial intelligence as a revolution capable of changing society. Some see it as a once-in-a-generation breakthrough, while others dismiss it as an overhyped fad and point to its frequent mistakes.

Lately I’ve been experimenting with small language models (LLMs) that can run directly on the device. I built an Android app in WebAssembly that executes a local model and tried it on a five-year-old phone. When I bought it, the phone was meant only for calls and installing apps. Today it manages to answer my questions.

Here’s a short screencast of the experiment:

In the video I ask the model: “When was Italy born?”
The answer is wrong, yes — but what’s interesting is that the model understands the question, recognizes that it’s about history, and connects coherent concepts. That’s not trivial for a system that runs entirely in your pocket.

The lesson, to me, is clear: AI works, as long as you understand how to use it. Tools need to be known and evaluated for what they are, not for what we expect them to be. Even the smallest models have enormous potential, especially when paired with tools like MCPs, which let them access real information.

Maybe it isn’t an intelligence on par with the human one, but it’s certainly not something to downplay. We’re holding a powerful, fascinating tool, and it’s worth getting to know it, understand it, and use it well so that we can capture everything it has to offer.