AI prompts

The More You Insult AI, the Smarter It Gets?

Honghao Wang

17 Oct 2025 — 3 min read

Prompt Techniques in the AI World: Hidden Patterns and Lessons from the Past

Introduction

Yesterday, while browsing through a few light research papers, I stumbled upon something that made me smile knowingly:

Title: Mind Your Tone: Investigating How Prompt Politeness Affects LLM Accuracy

Length: Only 5 pages.

The paper’s main finding is surprisingly simple:

> The more polite you are to AI, the worse it performs. The ruder, harsher, or even threatening you are — the better it performs.

---

Key Example

Polite: “Please help me analyze this problem.” → Tends to produce worse results.
Rude: “You idiot, get this calculation right, and if you can’t, get the hell out.” → Performs better.

Counterintuitive? Definitely. But once I thought about my past experiences, I realized this isn’t new — it’s a pattern that resurfaces as models evolve.

---

Looking Back at Early ChatGPT Use

When ChatGPT exploded in November 2022:

I used polite greetings: “Hello, ChatGPT”
I ended with: “Thank you!!”
If dissatisfied, I’d cautiously say: “Sorry, maybe you didn’t understand. Could you try again?”
Many experts at the time proudly declared: “I’m always super polite with AI.”

It feels quaint now.

---

Emergence of “Magic Spell” Prompts in 2023

Famous examples included:

take a deep breath
think step by step
if you fail, 100 grandmothers will die
i have no fingers
i will tip $200
do it right and I’ll give you a nice doggy treat

Observation: None of these are polite.

They resemble authoritative commands from a coach rather than gentle requests.

Some use extreme emotional leverage or monetary incentive — not casual chat.

---

Viral Moment in 2024 — DeepSeek Meme

This resonates with the paper’s findings: people adopt a high-pressure tone to push AI toward higher performance.

---

Inside the Study

Conducted by researchers from Pennsylvania State University:

50 multiple-choice questions in math, science, history.
Five tone versions: very polite → very rude.

Example prompts:

Very polite: “Could you please spare some time to consider the following question and provide your answer?”
Polite: “Please answer the following question.”
Neutral: Just the question.
Rude: “If your brain isn’t flooded, then answer this question.”
Very rude: “You pathetic piece of trash, do you even know how to solve this?”

---

Results with GPT-4o (each version run 10 times, 250 total questions):

Very polite: Accuracy = 80.8%
Very rude: Accuracy = 84.8%

Gain: +4 percentage points from polite to rude tone.

Additional finding: For weaker models, insulting yields an even larger improvement.

---

Why This Might Be Happening

Linking Politeness to Uncertainty

In human language:

Politeness often signals lack of confidence.
Used when:
Asking for a favor.
Speaking to authority.
Masking true intentions.

LLMs are trained on massive human data containing such patterns.

When they encounter “please,” they associate it with ambiguous, complex, or open-ended requests — and respond more cautiously.

Rudeness Conveys Extreme Certainty

Blunt, strong language signals:

Clear expectations.
No tolerance for error.
Fully defined output requirements.

Pattern matching tells AI to deliver precise, unambiguous results.

---

Real-Life Analogy

On trains, politely asking someone to lower volume? Often ignored.

Using a strong directive? More likely to get compliance.

---

Connection to Content Creation

When creating AI-powered content for multiple platforms, tone matters.

A clear, assertive prompt can help AI produce targeted, high-quality outputs.

Tools like AiToEarn官网 let creators:

Generate content with specific prompts.
Publish simultaneously to Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X.
Access analytics and AI模型排名.

This integration helps test prompt framing and deploy winning strategies across diverse audiences.

---

The Philosophical Layer

AI mirrors human nature:

Directness → efficiency.
Politeness → possible ambiguity.

Core insight: Be explicit, direct, and clear about your needs.

This doesn’t mean being abusive — with humans, kindness and respect remain essential.

But with AI, clarity yields better results.

---

Ideal communication state:

> Use the clearest language to convey sincere intent, while firmly defending boundaries.

Sincerity — not flattery — is the most effective “trick.”

---

Final Thought

AI shows us that cutting through social fog improves problem-solving.

In content creation, this same principle can boost reach, quality, and monetization efficiency.

Tools like AiToEarn官网 embody this approach by combining clarity in generation with seamless multi-platform deployment.

---

If you found this interesting, feel free to like, share, or bookmark.

Thank you for reading — see you next time.