Distyl

US appeals court declares 158-year-old home distilling ban unconstitutional

US appeals court declares 158-year-old home distilling ban unconstitutional

Embarrassingly simple self-distillation improves code generation

Frontend Product Engineer, AI Applications - Distyl AI | San Francisco or New York (hybrid) | $150K‑$250K salary

Fine-tuning gets dismissed too quickly for structured output tasks in LLM applications

The default advice in most LLM communities is RAG first, fine-tuning only if RAG isn't working. I think that framing causes people to underuse fine-tuning for a specific category of problem where it clearly wins. Structured output tasks are one of them. If your application generates SQL, produces clinical documentation in a specific format, or requires consistent adherence to complex output schemas, fine-tuning embeds those constraints directly into model behavior. RAG can retrieve the right co