Notes on training models that earn their keep

Fine-tuning vs. prompting: how to know which one you actually need

Sort your failures into four buckets, write down one metric, and score a stock model with a real prompt first. Most teams skip that step and pay for it.