Which customization approach usually costs least?
Why D is the answer
In-context learning through the prompt
Costs rise from prompts to RAG, fine-tuning and pre-training.
In detail
- Examples in a prompt change no model, so it usually costs least.
- Move up the ladder only if the cheaper step fails.
Use case
An illustrative scenario, not a real incident.
A team adds a few examples to its prompt and gets what it needs, so it avoids any tuning cost.
See it in different ways
flowchart LR A["In-context learning: prompts"] --> B["RAG"] B --> C["Fine-tuning"] C --> D["Pre-training"]
stateDiagram-v2 [*] --> InContext: prompt examples InContext --> Rag: needs your data Rag --> FineTuning: needs your style FineTuning --> PreTraining: rarely, a new model
flowchart LR S0["In-context learning"] --> S1["RAG"] --> S2["Fine-tuning"] --> S3["Pre-training"]
Key point: Prompting is usually the cheapest customization.
Why the other options are wrong
- A. Pre-training a model from scratch
- Pre-training is the most costly.
- B. Fine-tuning on a large dataset
- Fine-tuning costs more than prompting.
- C. Training a new foundation model each month
- Monthly training is very costly.