ClarityAI LogoClarityAI
Training & Optimization/fine-tuning

Fine-tuning

Further training a pre-trained AI model on specific data to specialise its behaviour.

What it actually means

Fine-tuning takes a general-purpose model that has already been trained on vast amounts of text and trains it further on a smaller, specific dataset. This shifts the model's behaviour toward a particular style, domain, or task without starting from scratch.

Real-world analogy

Think of a medical student who has finished general education and now does a specialisation residency. They already know medicine broadly — fine-tuning is the residency that makes them an expert cardiologist. The base knowledge stays, but specialist knowledge is layered on top.

Common misconception

Fine-tuning is not the same as giving the model new instructions in a prompt. It permanently changes the model's weights. It's also not always better than good prompting — for many tasks, a well-written prompt on a strong base model outperforms a poorly fine-tuned one.

Related terms