What 'Trained On' Means for Your Students

AI companies say their models are "trained on" vast amounts of text — books, websites, articles. In plain English: the model studied patterns in that material the way a student studies for an exam, except the exam was the entire public internet.

Here's why that matters for your classroom. The internet contains brilliance and bias in roughly the proportions you'd expect. A model trained on it will write a decent essay about the Civil War and also reproduce the stereotypes, confident errors, and outdated facts it absorbed along the way. "Trained on the internet" means "inherited the internet's blind spots."

Three practical consequences. First, AI output needs a teacher's review the way a Wikipedia article needs a second source — useful starting point, never the final word. Second, the model has no idea what's true; it predicts what sounds plausible. Confidence is not accuracy. Third, anything unusual about your students — their culture, their context, their way of speaking — is underrepresented in that training data, so the model's defaults won't fit them.

You don't need to understand neural networks. You need this one sentence: the machine learned from the world's text, and the world's text is flawed. Teach students to read AI output the way you'd teach them to read any source — with curiosity and a raised eyebrow.

I explain frameworks; I don't give legal counsel.