What is fine-tuning used for?#
By training the model on a curated set of your own data, fine-tuning makes the AI more accurate, relevant, and effective for your specific needs. This results in better performance for a target function, such as an AI chatbot that can answer detailed questions about your products using the correct brand voice.
How does fine-tuning work?#
The process starts with a large, pre-trained AI model, which has already learned language patterns, facts, and reasoning skills from being trained on vast amounts of public internet data. This foundational model can be thought of as a highly educated generalist.
To fine-tune it, a smaller, high-quality dataset specific to a business is prepared. This could be a collection of past customer support chats, internal technical documents, or marketing copy. The model is then trained further on this new data, slightly adjusting its internal connections to better reflect the patterns, style, and information in the specialised dataset. It isn't learning from scratch; it is refining its existing knowledge for a specialised purpose.
How is fine-tuning different from prompt engineering?#
Fine-tuning creates a specialised version of an AI model by permanently altering its internal knowledge. This differs from "prompt engineering," which involves carefully crafting instructions to guide a general-purpose model's response in real-time, without changing the model itself. Think of fine-tuning as permanently teaching an employee a new skill, whereas prompting is like giving that employee very specific instructions for a single task.