Pre-training, fine-tuning and in-context learning in Large Language Models (LLMs)