what is artificial intelligence?

AI means that a computer has learned by itself how do something.

Traditionally, when we use a software for a task, it relies on an algorithm. That is, humans have written a set of instructions for the computer to follow. if this then do that; else do whatever; while at it continue; or break; etc. In contrast, AI used to be called machine learning (ML) because, instead, the computer is given an objective and has to learn and come up with a solution by itself.

How does this learning work? Through trial and error. It uses a set of training examples and adjusts many "parameters" These parameters are usually called weights (and biases) which control how to go from an input to the expected output.

From the A.I. movie in 2001 (an eternity ago). AI is very good at very complex things, where writing the rules would be far too cumbersome. But it's also not an algorithm, and humans don't know what it does internally. AI is a mini computer brain; and like our own brains, you can't just look at it to know what it does.

large language models

Because of this "black box" effect, I used to be skeptical of AI in physics. But in early 2023, ChatGPT came out and I was intrigued. I kept hearing: 'no one understands how it works'.

ChatGPT is what we call a Large Language Model, or LLM. It means that it's a model (the term for a trained little computer brain), for language, and that it is large. It's "large" because it contains billions of parameters, whose values have been acquired through training.

The insides of an LLMs are like a giant machine made of billions of Lego bricks. In fact, LLMs are so large that they become human-level smart. But also unpredictable — because they are so complex. Rather than "black boxes", I prefer to refer to them as "Lego boxes": we can look inside, but all we see is a giant machine made of billions of bricks, far beyond human comprehension. But can we still learn something?