GCP-PMLE · Drill mode

To control LLM hallucinations, the most generally effective combination is:

• AA. Retrieval-augmented generation (RAG) over authoritative sources, plus structured prompt patterns and post-generation verification
• BB. Higher temperature
• CC. Larger response token limit
• DD. Removing all system prompts