From a recent Economist article https://www.economist.com/schools-brief/2024/07/23/ai-firms-will-soon-exhaust-most-of-the-internets-data?
Model capabilities can also be improved when the version produced by self-supervised learning, known as the pre-trained version, is refined through additional data in post-training. “Supervised fine-tuning”, for example, involves feeding a model question-and-answer pairs collected or handcrafted by humans. This teaches models what good answers look like.