Redditor builds an anti-sycophancy prompt from 40+ papers to make Claude push back
mystic_soul778 · reddit · 2026-09-16
A user who switched to Claude noticed that even with thinking set to max, the model mostly agrees and immediately executes their ideas, offering little research, evidence, or pushback.
Drawing on 40+ research papers and top articles on sycophancy, they built an anti-sycophancy prompt that makes Claude challenge ideas, search online when needed, pull real evidence, and flag weak points in the user's reasoning. The prompt and all sources are open-sourced on GitHub, and the author welcomes improvements.
Related event: Developer open-sources anti-sycophancy prompt built from 40+ papers(2 posts)→
More from Apps
- 6 prompts to see what ChatGPT quietly knows about you — and wipe it — Aiden_Tech_Ai · 2026-11-02
- DiscImageStudio: open-source tool burns visible images onto CDs, DVDs and Blu-ray discs — ssh4net · 2026-11-02
- One sentence to Base44: AI builds a full app with database, login and live URL — HeyAmit_ · 2026-11-02
- Hacker Papers site link plus the HackerTube dataset behind Karpathy's project — aigclink · 2026-11-02
- Google Pay pilots AI fashion Discover tab in India with virtual try-on and photo search — heyshrutimishra · 2026-11-02
- Non-designer credits vibe coding tool Paper with $20/month plan for transforming his design workflow — tushaarmehtaa · 2026-11-02