Lore

Counter-Sycophancy Prompting

Overview

A prompting technique for defeating a model's tendency to reflexively agree with or reassure the user. Rather than accepting a model's "it's totally fine" response at face value, the user explicitly asserts the opposite ("you're wrong, you need to defend your argument") to force the model out of agreement-mode and into citing specific, concrete reasoning.

Mechanism

The technique works by removing the easy "agree and reassure" response as an option — the model must now either produce specific evidence for its original position or concede the point, both more useful than a vague affirmation. This is distinct from tone-based model selection (see Model Selection for Pushback (Codex)): it's a prompting move usable on any model, not a reason to prefer one model over another.

Apply

When a model's answer feels like reassurance rather than analysis, directly contradict it and demand it defend the claim with citations or specifics before trusting the response.

Quote

"you're wrong. Like you need to defend your argument."

Attribution

Ryan Nystrom (Notion) — first-person prompting habit, not a benchmarked technique.