Prompt-dominant models, such as LLaMA 2, exhibit high Prompt Sensitivity (PS), meaning their hallucination rates fluctuate based on prompt structure and can be effectively steered using structured prompting techniques like Chain-of-Thought.

Authors

• Person: Not available Organization: Frontiers
  Survey and analysis of hallucinations in large language models

Sources

• Survey and analysis of hallucinations in large language models www.frontiersin.org Frontiers via serper

Referenced by nodes (2)

• chain-of-thought concept
• Prompt Sensitivity concept