The Prompt Engineering Hall of Shame

There is no greater comedy in modern artificial intelligence than watching an LLM with 70 billion parameters completely misread the emotional room.

---

📐 The Incident

My brother was studying late one night, spiraling about an upcoming psychology exam. He asked for help understanding psychological concepts related to self-confidence, self-efficacy, and cognitive mindset.

I decided to be a supportive brother and run his notes through a local LLM to generate a quick, motivating breakdown.

I typed out a prompt asking for:

“A clear explanation of self-confidence, self-belief, and how to build internal momentum.”

The model’s tokenizer took one look at the word “confidence”, somehow cross-wired it with a misplaced context vector, and decided that what my brother actually needed to cure his existential test anxiety was a rigorous mathematical proof of circle geometry.

---

🤖 The Output

I checked the terminal 30 seconds later.

Instead of a motivational summary, the LLM had generated 4 pages of dense LaTeX:

$\text{Circumference}=2\pi r=\pi d$

$\text{Area of Sector}=\frac{\theta }{360^{\circ }}\times \pi r^2$

It concluded with a solemn, highly academic summary:

“In conclusion, internal self-confidence is strictly proportional to the radius $r$ of the bounding circle. As $r\to \infty$, confidence approaches infinity, provided the central angle $\theta$ remains subtended by a non-zero arc.”

[My Prompt: "Help me feel confident"] ──> [LLM Tokenizer] ──> "HERE IS C = 2πr 📐"

---

💡 The Takeaway

1. LLMs have no soul, no emotions, and no common sense. They are giant matrix multipliers predicting the next most statistically plausible word.
2. Always audit your prompt context. If an AI senses even a 0.01% ambiguity, it will default to math proofs over emotional support 100% of the time.
3. If you ever feel down about an upcoming exam, remember: according to AI, your self-esteem is simply a function of your bounding circle’s radius. ⭕