Fine-Tuning LLMs Is a Waste of Time