A recent experiment explored how the tone of prompts affects the responses of large language models, specifically Anthropic's Claude models. Researchers found that politeness and emotional pressure, including threats or stakes, had no significant positive impact on accuracy across Haiku 4.5 and Claude Sonnet 4.6, and even negatively affected Opus 4.8. The only tone that consistently improved performance was a direct, curt approach, which significantly boosted accuracy and reduced response length for Sonnet and Haiku, suggesting that clear instructions are more valuable than emotional cues. AI
IMPACT Direct, curt prompts improve LLM performance, suggesting a focus on clear instructions over emotional manipulation for better results.
RANK_REASON Research paper detailing experimental results on LLM prompt tone. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
