Introduction to Large Language Models by Andrej Karpathy is one of the clearest explanations of LLMs. He covers the three stages of training – Pre-training, Assistant Model Training and RLHF (reinforcement through human feedback).
According to him, pre-raining producers, large foundation models (he does not use the term foundation). It is characterized by large quantity but low-quality data.
Assistant model training is data from people is used to refine the foundation (basic) model but of small quantity and high quality.
RLHF is the stage of refinement where human feedback is used to improve the model further.
If you are new to LLMs, this is certainly a good place to start with. Our methods were for professional products. So I will share them. We have no experience with consumer product.