Fine-Tuned Models

Learn behavioral regression testing for fine-tuned models — systematically verifying that capabilities you didn't touch weren't silently degraded by training.

Learn to build a representative benchmark, compare fine-tuned models against their base models and frontier alternatives, and avoid the eval-set overfitting trap that makes numbers look better than they are.

Learn alignment verification for fine-tuned models — detecting reward hacking, sycophancy, and unintended behavioral shortcuts that make models look good on metrics while failing on actual goals.