When would you use RAG vs fine-tuning? Can you use both?

Strong answer

Use RAG for knowledge: private/frequently-updated data, when you need citations, when data changes faster than you can retrain. Use fine-tuning for behaviour: consistent output style, domain-specific reasoning patterns, custom formats, private languages. They're complementary — fine-tune the model's behaviour/style, use RAG for the knowledge. Most production systems start with RAG + prompting (cheaper, updatable) and add fine-tuning only when prompting can't achieve the required consistency. Never fine-tune for knowledge updates — retraining is too slow and expensive compared to re-indexing.

Weak-answer indicators

• Says 'fine-tuning is better' without qualifying
• Doesn't mention they're complementary
• Ignores update frequency as a factor
• Doesn't mention citations