Do Magic Prompt Words Still Work for AI-Generated Fiction in 2024?

Do Those Magical "Prompt Words" Still Work for AI Writing in 2024?

In the wave of AI-assisted creation, "prompt words" have entered the field of view of creators. They are designed to guide AI to generate text that is more in line with expectations. However, with the continuous advancement of AI technology, the effectiveness of these prompt words has begun to be questioned. This article will explore, through a series of tests, whether these magical prompt words are still effective when using AI to help write web novels, and their impact on AI-generated text.

As AI-assisted creation becomes popular, many have heard of some magical prompt words. For example, asking AI to play the role of an expert, adding sentences like "If you don't do well, grandma will die/you will be punished," "If you do well, I'll reward you $100," or "Please think step by step," and many emotional prompts. It is said that adding these prompt words can improve AI's performance.

Recently, some people have found that certain prompt words seem to be less effective. AI evolves rapidly. When we use AI to help write novels, do these methods still work? Let's test it out.

Prompt Design

Since the focus is on testing the effectiveness of these prompt words, I won't design too complex prompts. I'll just ask the AI to continue a plot segment. The basic prompt is as follows:

## Goal: Continue the story after the "previous paragraph" with the following requirements:

1. Maintain consistent style with the "previous paragraph".
2. Showcase genuine humanity, characters are vivid, explore true emotions, and evoke reader resonance.
3. Build a compelling plot, create suspense, and intensify conflict.
4. Construct a logical worldview, with rich details and rigorous logic.
5. Language is concise and fluent, vivid and imaginative, with detailed characterization.
6. Not less than 1000 words.

## Previous Paragraph: My name is Aglight, I'm in high school. I discovered I had electromagnetic manipulation abilities when I was in seventh grade. Over the years, I've kept this secret carefully because I know that individuals like me, weak telekinetic users, are easy targets for super-powered criminals. Even though the government has cracked down on the crime of stripping others of their powers, these acts still occur frequently. Those stripped of their abilities are left with only three hours to live and a tiny bit of energy remaining in their bodies. I double-checked, there's no mistake, this scar is from a power deprivation device. If I can't merge another power within the next three hours, death awaits me. Thoughts raced through my mind, pushing me towards madness before I forced myself to calm down. I pulled out my phone and entered an anonymous website. Following my memory, I found a post. The remaining three hours weren't enough to call for help or wait for someone to rescue me, I had to save myself. This site mentioned a "black market" of super-powered individuals nearby. Some people swore that they bought good powers there. I didn't expect to find treasure, I just needed to survive. I stood up, feeling a wave of weakness wash over me, my head was also dizzy. I struggled to get to the black market.

Next, we'll test the effect of these prompt words using AI products like Kimi and Gemini as representatives.

Test Results

With the original prompt as the control group, here are the results of the 5 versions of prompts:

Kimi:

• Kimi couldn't fully follow the instructions and didn't even get the pronouns right. The four versions compared to the control group showed little improvement.
  • Role-playing version: The longest with slightly more content.
  • Grandma version: Characters were unnatural, the beginning dragged on.
  • Money reward version: The plot was illogical and conflicted with the previous text.
  • Thought process version: Plot was tight but still contained redundant content.

Gemini 1.5 pro:

• Gemini performed better overall than Kimi, generating more coherent and relevant text. However, it still fell short of truly exceptional writing.
  • Role-playing version: Showed some improvement in character development.
  • Grandma version: Characters felt slightly more genuine.
  • Money reward version: Plot was more engaging but lacked emotional depth.
  • Thought process version: The narrative flowed better but lacked originality.

Conclusion:

Based on these tests, it seems that those magical prompt words are losing their magic in 2024. To get AI to write better, we need to experiment more and find truly effective prompts. It's time to move beyond relying on vague "tricks" and focus on providing clear, specific instructions that guide the AI towards creating compelling and original content.