What is it?
Fine-tuning
Further training of a model on a narrower dataset so it specializes in a task or style. It can boost performance now, but the articles warn it can lock you into a "local optimum": when a better general model arrives, your fine-tuned setup may already be obsolete. See also the Bitter Lesson.
From the AI agent glossary
As the Foundational Modelling glossary puts it
Fine-tuning means taking an already pre-trained model and training it a bit more on a narrower dataset to specialize it. Pre-training gives broad general ability; fine-tuning shapes that ability toward a specific goal, such as following instructions, writing in a company's voice, or answering medical questions.
Fine-tuning is far cheaper than pre-training because the model already knows language; you are only nudging it. Chapter 4 (LoRA) is about making fine-tuning cheaper still.
Where Fine-tuning is explained
- LoRA, Fine-Tuning Giant Models on a BudgetFoundational Modelling
- Alignment, Turning a Text Predictor Into a Helpful AssistantFoundational Modelling · Paper 1: InstructGPT and the RLHF recipe
- Practical Lessons from Building ManusContext Engineering · Do not train your own model yet, and respect the Bitter Lesson
- The Case Against: Cognition's "Don't Build Multi-Agents"Multi-Agent Systems · What to build instead
- Deep Research Agents and How They Are TrainedBuilding Coding Agents and Harnesses · How they are trained