Fine-tuning LLMs is a waste of time