Fine-Tuning (AI)

What Is Fine-Tuning (AI)?

Fine-tuning takes a general-purpose LLM and specializes it for your domain. For example, fine-tuning GPT-4 on your customer support transcripts creates a model that speaks in your brand voice and understands your product terminology.

When to fine-tune: your domain has specialized language (legal, medical, financial), you need consistent output format or tone, RAG alone does not achieve sufficient accuracy, or you want to reduce prompt length (and therefore cost).

When NOT to fine-tune: your data changes frequently (use RAG instead), you have less than 1,000 training examples, your budget is under $5K, or your use case works well with prompt engineering alone.