Improving language model behavior by training on a curated dataset

In three linesOpenAI shows that fine-tuning on a small curated dataset improves language model behavior with respect to specific behavioral values. Pragmatic approach to align models without requiring massive data volumes.

Summary generated by Claude — human-verified