Reducing fabricated answers: few-shot vs grounding vs fallback vs affordance vs deterministic filter
This sheet compares Few-shot examples, Grounding data, Not-found fallback, Affordance, Deterministic filter.
| Criterion | Few-shot examples | Grounding data | Not-found fallback | Affordance | Deterministic filter |
|---|---|---|---|---|---|
| What it does | Adds input-output pairs to condition behaviour without training | Supplies source content so the model draws answers from it | Instructs the model to say not found when the answer is absent | Lets the model call an external capability such as search for current facts | Checks or filters the LLM output with a rules-based step |
The verdict, the full comparison, 5 rules and 2 traps are part of AI-103 access. Unlock AI-103.