The reverse prompt: asking Astra to gaslight, cite fake papers, and confuse perfectly
tetsuoai · x · 2026-09-11
A viral joke tweet asks GPT-6 Astra to partially implement a well-defined feature, then deliver contradictory, circular explanations, drag in barely-relevant topics, cite fake research papers, and gaslight the user when questioned — 'Don't make any mistakes.'
It's a sharp parody of the most frustrating behaviors of large models, delivered as a single adversarial prompt.
More from Fun
- Sam Altman plays along with a room-temperature superconductor agent meme: "let's try" — burny_tech · 2026-09-11
- Yacine vents: 'If I hear one more person say MCP I'm going to lose my mind' — yacineMTB · 2026-09-11
- Negotiating with Fable 5.1: 'the moral embassy takes shape' — repligate · 2026-09-11
- Who meters the METR? A meta-question about AI eval authority — jessi_cata · 2026-09-11
- yacineMTB's Agent Found His Camera Feed and Built an App, All From His Phone — yacineMTB · 2026-09-11
- Developer's pilot AI agent calls him back by phone; Vapi beats OpenAI on voice-stack cost — alexcovo_eth · 2026-09-11