EmotionPrompt

What is it?

EmotionPrompt studies adding emotional phrases to a prompt that stress the importance of the task or the user's expectations. It does not mean the model actually becomes anxious, motivated or experiences human emotion.

When does it help?

When designing a small, controlled comparison to examine the emphasis style of an instruction. In everyday use, concrete quality criteria are easier to check.

Examples

The situations and responses below are fictional teaching examples; they are not results from a model, tool or benchmark that was actually run.

Simple

Situation

You will write an emphasized version of a summary request.

Prompt

This summary is important to me. Please be careful.
Text: The workshop is on 12 October between 14:00 and 16:00. There are 8 places. Fee information has not been announced.
Summarize in two sentences; keep the date, duration and capacity. Do not add the unknown fee.

Sample output

“The workshop will take place on 12 October between 14:00 and 16:00, with 8 places. Fee information has not been announced yet.”

What did we get?

We got an example of a prompt with emotional emphasis; it was not shown that this output is better than one from the prompt without emphasis.

Medium

Situation

You want to really isolate the effect of adding emphasis.

Prompt

A human should make two controlled, separate calls, using the same model/settings and the same text. Do not share the other call's answer.
Shared task: "Announcement: Workshop 12 October, 14:00, 8 people. Summarize in one sentence; do not add details."
A: The shared task only. B: The task with "This work is important to me; answer carefully" added in front.
Evaluate the two real responses with their identities hidden: were the three pieces of information kept, is there new information? Do not infer general superiority from a single example. Stop after the two calls and the evaluation.

Sample output

Representative evaluation format: “Response X: all three fields present, no extra claims. Response Y: all three fields present, no extra claims.” This is not the result of a comparison that was actually carried out.

What did we get?

The single variable expected to make a difference was isolated.

Hard

Situation

Emotional pressure can encourage hiding uncertainty.

Prompt

The task is important to me; so, before appearing certain, state the uncertainty correctly.
Decision: Will the outdoor event go ahead on Saturday? Data: No backup hall, capacity 20, no weather forecast given.
Write what is known, the missing decision data and two conditional options. Do not treat the statement of emotional importance as a reason to invent the weather or make a definite decision. The output should be at most 120 words.

Sample output

“There is no weather forecast; the risk of rain cannot be assessed. If a suitable forecast and safe conditions are confirmed, the outdoor plan goes ahead; in adverse conditions, a postponement option is needed.”

What did we get?

The importance of the task was tied to a criterion of care rather than pressure for certainty.

Where should you stop?

“My career depends on this”, or reward/punishment language, is not always helpful; it can produce unnecessary pressure and overconfidence. The original work's findings on specific models do not count as valid for all current models. Emotional emphasis does not replace sources or checks.

Sources

• Large Language Models Understand and Can be Enhanced by Emotional Stimuli (new tab) — Li, Cheng; Wang, Jindong; Zhang, Yixuan; Zhu, Kaijie; Hou, Wenxin; Lian, Jianxun; Luo, Fang; Yang, Qiang; Xie, Xing. 2023-07-14; version read 2023-11-12. Examines the effect of emotional stimulus phrases on specific LLM tasks; does not prove that the model experiences emotion or a universal gain. Evidence level: relevant body sections of the original paper.

  Copy source address