Started messing with ChatGPT back when it could barely follow a chain of thought. Over time developed a set of prompt frameworks that turned GPT into something that actually thinks instead of just generates.
A few things I learned:
1. Chain-of-thought is overrated. What works is recursive prompting where the prompt evaluates its own output and adjusts. Most people don't do this.
2. The best prompts are written for the model's weaknesses, not its strengths. Know where GPT hallucinates and build guardrails into the prompt itself.
3. Context windows are your enemy. Long prompts dilute signal. Short, layered prompts that build on each other outperform monolithic system prompts.
I packaged my most-used frameworks — the Meta-Prompt Mastery Pack ($24) is the one I use most. It's a meta-layer that rewrites itself based on what you're trying to do.
If you're tired of prompts that start with "Act as an expert..." and actually want GPT to show you something new:
https://darkmind68.gumroad.com/l/calpqo
Entering this in the Hermes Agent Challenge. Full prompt packs here (disclosed): https://darkmind68.gumroad.com
Open to questions in the comments.
Top comments (1)
Open to any questions!