DictionaryArchitecture
Fine-Tuning
Giving an AI extra training on your own examples so it permanently picks up a style or specialty.
Definition
Further training a pretrained model on additional data so it adopts that data's patterns, domain, or style. Unlike prompting or RAG, it changes the model's weights. Anthropic's glossary notes that the Claude API does not currently offer fine-tuning; for most coding work, context engineering and retrieval are the practical alternatives.
Example
A company trains an open-weight model on 10,000 past support replies so it answers in their tone without a long style guide in every prompt.
Appears in
- TRACE: TRajectory Attribution for Automated Context EngineeringarXiv
- OpenGame: Open Agentic Coding for GamesarXiv
- Ref-Long: Benchmarking the Long-context Referencing Capability of Long-context Language ModelsarXiv
- Token-Efficient RL for LLM ReasoningarXiv
- Token-Efficient Leverage Learning in Large Language ModelsarXiv
- How to work with large language modelsOpenAI Cookbook