Li et al. (2023), "Large Language Models Understand and Can Be Enhanced by Emotional Stimuli," test appending short psychologically-motivated phrases, such as "This is very important for my career" or "You'd better be sure and think carefully," to the end of otherwise-unchanged task instructions, an approach the paper calls EmotionPrompt. What did the paper report as the effect of adding these phrases, and did the technique require retraining the model?
- The phrases had no measurable effect on any tested model, confirming that language models are insensitive to wording intended to convey urgency or stakes
- The phrases improved performance only after the model was fine-tuned on a dataset of emotionally-annotated examples paired with correct answers, so the effect required retraining rather than being a pure prompting technique
- The phrases decreased benchmark performance by making the model overly cautious and more likely to refuse to answer, though human raters still preferred the more cautious tone
- Adding these appended emotional-stimulus phrases to otherwise-unchanged instructions produced measurable improvements in benchmark performance and in human ratings of the resulting text across multiple LLMs, and the technique required no retraining or fine-tuning of any kind, since it works purely by changing the wording of the prompt
Why D? And why not the others?
Correct answer: D. Adding these appended emotional-stimulus phrases to otherwise-unchanged instructions produced measurable improvements in benchmark performance and in human ratings of the resulting text across multiple LLMs, and the technique required no retraining or fine-tuning of any kind, since it works purely by changing the wording of the prompt
EmotionPrompt appends a short psychologically-motivated phrase to an existing instruction without touching anything else about the prompt or the model, and the paper reports that this simple wording change produced measurable improvements in benchmark performance and in human ratings of the resulting text across the multiple LLMs tested, with no retraining or fine-tuning involved at any point; the improvement comes purely from how the request is phrased. The option claiming no measurable effect is wrong because the paper's core finding is precisely the opposite, that these phrases reliably shifted performance and human ratings. The option claiming the benefit only appears after fine-tuning on emotionally-annotated examples is wrong because EmotionPrompt is tested and reported as a prompting-only technique applied to already-trained, frozen models. The option claiming performance dropped due to excessive caution is wrong because the paper's reported outcome is an improvement in both benchmark scores and human preference, not a refusal-driven decline.
Source: Li, Jiang, Zhang, Chen, Lv, Zhao, Wu & Lin, 'Large Language Models Understand and Can Be Enhanced by Emotional Stimuli' (arXiv:2307.11760, 2023)