A hallucination is incorrect or nonsensical information generated by a language model. Insufficient context and vague prompts make this more likely because the model produces probabilistic responses. section 3.1
Too little or too much context
A very short prompt may leave the model without enough context, causing it to invent information or wander off topic. An overloaded prompt can also reduce quality when important instructions become unclear. section 3.1
Reduce hallucinations by:
- Providing necessary context with clear, structured instructions. section 3.1
- Adding verification layers or integrated external checks. section 3.1
- Avoiding unnecessary prompt length, since excessively long prompts can degrade response quality. section 3.1
Figure 3.1 Like pirates, GenAI can also suffer from hallucinations.
Use the illustration as a memorable visual metaphor for generated misinformation. section 3.1
