H
Member
We're data-constrained in a niche domain and considering synthetic data generation with a strong teacher model. The risk of mode collapse and compounding errors worries me. Has anyone made synthetic data work for fine-tuning in a specialized domain? What validation convinced you it was safe?

