I remember that when I was a kid, whenever a train passed by, I would try to pay as much attention as possible to see how they worked. Today, when I want to know how something works, I simply ask Claude or ChatGPT. In many cases, the responses are good and accurate. In some, they are very off. But the data comes from things we create . If we want better results, we need to put an effort in generating better data. And to make sure the models we train get trained on this improved data.