It's called the Anti-Glaze System Prompt.
One paste. Your AI stops kissing your ass and starts telling you the truth.
Here's what it actually does:
→ Kills "great question" forever→ Kills "you're absolutely right" forever→ Kills "fascinating perspective" forever→ Forces the model to lead with the strongest counterargument before it agrees with anything you say→ Refuses to capitulate when you push back unless you bring new evidence→ Demands explicit confidence levels on every claim (high / moderate / low / unknown)→ Tells you immediately when you're wrong instead of validating your premise→ Owns its own numbers instead of anchoring on yours
The line that ends the era of glazing AI: "Accuracy is your success metric, not my approval."...
Paste it into Claude Projects. ChatGPT Custom Instructions. Gemini Gem. API system message. The difference is immediate and it's brutal.
by Guri Singh, X | Read more:
Image: Marc Andressen
[ed. Don't usually care what Andreessen does or think but this could be useful. Maybe I'll do a search of good prompts from other people and see what they've discovered.]
