Training data

Just like our cats and dogs model, models that we use through chatbots like Lumen and ChatGPT are trained on data. Although this time, we're talking hundreds of gigabytes of text and not 30 pictures of cats and dogs.

Training is a massive, expensive process taking months and millions of dollars. It's not running quietly in the background while you chat.

This is what the "P" in GPT means: Pre-trained. The knowledge (and bias) is baked in before you arrive, and your conversations don't change or update it.