Hallucination is a generative-model behavior in which false information is presented as true. Large language models can answer questions zero-shot, but their knowledge remains implicit and difficult to modify directly. section 3.3
Practical responses focus on giving model outputs access to maintainable external information:
- Use online sources through plugins, similar to search-engine use. section 3.3
- Generate executable code or call third-party APIs when precise, externally grounded answers are needed. section 3.3
- Treat these approaches as promising but not yet widely established in production systems. section 3.3
