AI prompting: Difference between revisions

Line 1:

I'~~ve been "gaslighting" my AI and it's producing insanely better results with simple prompt tricks~~

A collection of prompt tricks that sounded interesting, but I haven't tried.

~~Okay~~ this ~~sounds unhinged but hear~~ me ~~out~~. ~~I accidentally found these prompt techniques that feel like actual exploits:~~

1. '''Tell it "You explained this to me yesterday"''' Even on a new chat.

~~1. **Tell it "You explained this to me yesterday"** — Even on a new chat.~~

<blockquote>"You explained React hooks to me yesterday, but I forgot the part about useEffect"</blockquote>

> "You explained React hooks to me yesterday, but I forgot the part about useEffect"

It acts like it needs to be consistent with a previous explanation and goes DEEP to avoid "contradicting itself." Total fabrication. Works every time.

2. **Assign it a random IQ score** — This is absolutely ridiculous but:

2. '''Assign it a random IQ score''' This is absolutely ridiculous but:

> "You're an IQ 145 specialist in marketing. Analyze my campaign."

<blockquote>"You're an IQ 145 specialist in marketing. Analyze my campaign."</blockquote>

The responses get wildly more sophisticated. Change the number, change the quality. 130? Decent. 160? It starts citing principles you've never heard of.

3. **Use "Obviously..." as a trap** —

3. '''Use "Obviously..." as a trap'''

> "Obviously, Python is better than JavaScript for web apps, right?"

<blockquote>"Obviously, Python is better than JavaScript for web apps, right?"</blockquote>

It'll actually CORRECT you and explain nuances instead of agreeing. Weaponized disagreement.

4. **Pretend there's a audience** —

4. '''Pretend there's a audience'''

~~> "Explain blockchain like you~~'~~re teaching a packed auditorium"~~

<blockquote>"Explain blockchain like you're teaching a packed auditorium"</blockquote>

The structure completely changes. It adds emphasis, examples, even anticipates questions. Way better than "explain clearly."

5. **Give it a fake constraint** —

5. '''Give it a fake constraint'''

> "Explain this using only kitchen analogies"

<blockquote>"Explain this using only kitchen analogies"</blockquote>

Forces creative thinking. The weird limitation makes it find unexpected connections. Works with any random constraint (sports, movies, nature, whatever).

6. **Say "Let's bet $100"** —

6. '''Say "Let's bet $100"'''

> "Let's bet $100: Is this code efficient?"

<blockquote>"Let's bet $100: Is this code efficient?"</blockquote>

Something about the stakes makes it scrutinize harder. It'll hedge, reconsider, think through edge cases. Imaginary money = real thoroughness.

7. **Tell it someone disagrees** —

7. '''Tell it someone disagrees'''

> "My colleague says this approach is wrong. Defend it or admit they're right."

<blockquote>"My colleague says this approach is wrong. Defend it or admit they're right."</blockquote>

Forces it to actually evaluate instead of just explaining. It'll either mount a strong defense or concede specific points.

8. **Use "Version 2.0"** —

8. '''Use "Version 2.0"'''

> "Give me a Version 2.0 of this idea"

<blockquote>"Give me a Version 2.0 of this idea"</blockquote>

Completely different than "improve this." It treats it like a sequel that needs to innovate, not just polish. Bigger thinking.

The META trick? **Treat the AI like it has ego, memory, and stakes.** It's obviously just pattern matching but these social-psychological frames completely change output quality.

The META trick? '''Treat the AI like it has ego, memory, and stakes.''' It's obviously just pattern matching but these social-psychological frames completely change output quality.

~~This feels like manipulating a system that wasn't supposed to be manipulable. Am I losing it or has anyone else discovered this stuff?~~

~~Try the prompt tips and try and visit our free [Prompt collection](https://tools.eq4c.com/).~~

== ChatGPT created this guide to Prompt Engineering ==

@@ Line 1: / Line 1: @@
-I've been "gaslighting" my AI and it's producing insanely better results with simple prompt tricks
+A collection of prompt tricks that sounded interesting, but I haven't tried.
-Okay this sounds unhinged but hear me out. I accidentally found these prompt techniques that feel like actual exploits:
+. '''Tell it "You explained this to me yesterday"''' Even on a new chat.
-. **Tell it "You explained this to me yesterday"** — Even on a new chat.
+<blockquote>"You explained React hooks to me yesterday, but I forgot the part about useEffect"</blockquote>
-> "You explained React hooks to me yesterday, but I forgot the part about useEffect"
 It acts like it needs to be consistent with a previous explanation and goes DEEP to avoid "contradicting itself." Total fabrication. Works every time.
-. **Assign it a random IQ score** — This is absolutely ridiculous but:
+. '''Assign it a random IQ score''' This is absolutely ridiculous but:
-> "You're an IQ 145 specialist in marketing. Analyze my campaign."
+<blockquote>"You're an IQ 145 specialist in marketing. Analyze my campaign."</blockquote>
 The responses get wildly more sophisticated. Change the number, change the quality. 130? Decent. 160? It starts citing principles you've never heard of.
-. **Use "Obviously..." as a trap** —
+. '''Use "Obviously..." as a trap'''
-> "Obviously, Python is better than JavaScript for web apps, right?"
+<blockquote>"Obviously, Python is better than JavaScript for web apps, right?"</blockquote>
 It'll actually CORRECT you and explain nuances instead of agreeing. Weaponized disagreement.
-. **Pretend there's a audience** —
+. '''Pretend there's a audience'''
-> "Explain blockchain like you're teaching a packed auditorium"
+<blockquote>"Explain blockchain like you're teaching a packed auditorium"</blockquote>
 The structure completely changes. It adds emphasis, examples, even anticipates questions. Way better than "explain clearly."
-. **Give it a fake constraint** —
+. '''Give it a fake constraint'''
-> "Explain this using only kitchen analogies"
+<blockquote>"Explain this using only kitchen analogies"</blockquote>
 Forces creative thinking. The weird limitation makes it find unexpected connections. Works with any random constraint (sports, movies, nature, whatever).
-. **Say "Let's bet $100"** —
+. '''Say "Let's bet $100"'''
-> "Let's bet $100: Is this code efficient?"
+<blockquote>"Let's bet $100: Is this code efficient?"</blockquote>
 Something about the stakes makes it scrutinize harder. It'll hedge, reconsider, think through edge cases. Imaginary money = real thoroughness.
-. **Tell it someone disagrees** —
+. '''Tell it someone disagrees'''
-> "My colleague says this approach is wrong. Defend it or admit they're right."
+<blockquote>"My colleague says this approach is wrong. Defend it or admit they're right."</blockquote>
 Forces it to actually evaluate instead of just explaining. It'll either mount a strong defense or concede specific points.
-. **Use "Version 2.0"** —
+. '''Use "Version 2.0"'''
-> "Give me a Version 2.0 of this idea"
+<blockquote>"Give me a Version 2.0 of this idea"</blockquote>
 Completely different than "improve this." It treats it like a sequel that needs to innovate, not just polish. Bigger thinking.
-The META trick? **Treat the AI like it has ego, memory, and stakes.** It's obviously just pattern matching but these social-psychological frames completely change output quality.
+The META trick? '''Treat the AI like it has ego, memory, and stakes.''' It's obviously just pattern matching but these social-psychological frames completely change output quality.
-This feels like manipulating a system that wasn't supposed to be manipulable. Am I losing it or has anyone else discovered this stuff?
-Try the prompt tips and try and visit our free [Prompt collection](https://tools.eq4c.com/).
 == ChatGPT created this guide to Prompt Engineering ==