Retrieval vs few-shot prompting vs fine-tuning vs a custom model
This sheet compares Retrieval over a knowledge base, Few-shot examples in the prompt, Fine-tuning, Custom model from scratch.
| Criterion | Retrieval over a knowledge base | Few-shot examples in the prompt | Fine-tuning | Custom model from scratch |
|---|---|---|---|---|
| Fixes | Missing or changing content | Format and behaviour; cannot carry a whole changing library | Long few-shot prompts, rising tokens and latency, a fixed output schema, wrong tool choice | Nothing the scenarios needed; heaviest option and still goes stale |
The verdict, the full comparison, 6 rules and 4 traps are part of AB-100 access. Unlock AB-100.