Fine-Tuning Isn't as Cool as I Thought

Everyone's trying to fine-tune. I did it too. Took a base model, fed it custom examples, got okay results. But honestly? Half the time you can get the same thing with better prompting and a bit of hackery.

Fine-tuning adds cost, maintenance, complexity. You've got to keep the data clean. Retrain when it drifts. And most of it is still trial and error.

I'm not saying don't do it. I'm just saying prompt engineering is underrated.