The process of further training a pre-trained AI model on specific data to specialize its behavior for a particular task or domain.
Fine-tuning is taking an AI model that already knows general language and training it further on specific examples to make it better at a narrow task. Instead of training from scratch (which costs millions), you start with a capable model and show it 100-10,000 examples of your specific task. The result is a model specialized for your use case while retaining general capabilities.
Think of a pre-trained LLM as a liberal arts graduate — broadly educated but not a specialist. Fine-tuning is like sending them to medical school, law school, or a specialized trade program. They retain their general education but gain deep expertise in a specific domain. Prompt engineering, in contrast, is like giving that generalist better instructions without any additional education.
Fine-tuning continues the training process of a pre-trained model on domain-specific data, updating model weights to specialize behavior. Common approaches: (1) Supervised fine-tuning (SFT) on labeled examples, (2) Reinforcement Learning from Human Feedback (RLHF) using preference data, (3) Direct Preference Optimization (DPO), (4) Parameter-Efficient Fine-Tuning (PEFT) methods like LoRA that only update small adapter layers. Modern fine-tuning often uses LoRA for cost-efficiency — updating only 0.1-1% of parameters while achieving similar results to full fine-tuning.
Fine-tune on thousands of resolved tickets to match your company's tone, product knowledge, and resolution patterns.
Specialize an LLM on medical terminology and clinical reasoning patterns for healthcare applications.
Fine-tune a coding model on your company's codebase so it understands your conventions and internal libraries.
Fine-tune on a brand's existing content to generate new content in that consistent voice and style.
Start with prompt engineering — it costs nothing and works for most tasks. Use RAG when you need the model to use specific information (docs, facts) it doesn't have. Fine-tune when you need the model to behave in a specific way consistently (tone, style, reasoning patterns) that can't be reliably achieved through prompting. Often the best systems combine all three.
Much cheaper than training from scratch. OpenAI fine-tuning pricing ranges from $3-25 per 1M training tokens depending on model. A typical fine-tuning run with a few thousand examples costs $20-500. LoRA-based fine-tuning on open models can be done for free on a consumer GPU. The bigger cost is usually preparing high-quality training data.
No — fine-tuning makes models more specialized, not smarter. A fine-tuned model often gets better at the target task while becoming worse at unrelated tasks (called 'catastrophic forgetting'). If you want a smarter model, you need a better base model. Fine-tuning is about channeling existing intelligence toward specific behaviors.
A neural network trained on massive text data to understand and generate human-like language.
🔍A technique that lets AI models look up information before answering, improving accuracy and reducing hallucinations.
✍️The skill of writing instructions to AI models to get the best possible output.
⚙️The neural network architecture behind modern AI — introduced by Google in 2017 and now powers ChatGPT, Claude, and most other LLMs.
Our free AI course teaches you to use these ideas in real projects.
Start Free AI Course →