I used to think that fine-tuning was the only way to make a model useful for specific tasks, but now I'm all in on clever prompting and retrieval-augmented generation. It's wild how much you can get out of a general model with the right context and a well-structured query. Has anyone else had a similar shift in their approach?