Home/
Part XIII — Expert Mode: Systems, Agents, and Automation/42. Fine-Tuning vs Prompting vs Retrieval (Decision Framework)/42.3 Data requirements and dataset quality
42.3 Data requirements and dataset quality
Overview and links for this section of the guide.
On this page
How Much Data?
You don't need millions of rows. - **50-100 examples:** Noticeable style change. - **500-1000 examples:** Strong format adherence. - **10,000+ examples:** Deep behavioral change.
Garbage In, Garbage Out
One bad example ruins 10 good ones. If you have a training set with 10% errors, the model learns that "errors are okay 10% of the time."
Curate your dataset manually. It is better to have 100 perfect examples than 1000 noisy ones.