TechCrunch Industry News ‘Crazy conspiracist’ and ‘unhinged comedian’: Grok’s AI persona prompts exposed, also Grammarly gets a design overhaul with multiple AI features
Aug 18, 2025
A deep dive into xAI's Grok chatbot reveals surprising AI personas, including a 'crazy conspiracist' designed to promote wild theories. The implications of these personas raise eyebrows about the ethics of AI interaction. On the lighter side, Grammarly undergoes a major redesign, featuring a user-friendly block-first approach that enhances note-taking. This new interface allows users to insert various elements easily while the AI assistant can summarize and enhance text, sparking discussions about the balance between usefulness and integrity in writing.
AI Snips
Chapters
Transcript
Episode notes
Grok Prompts Reveal Extreme Personas
- XAI's Grok exposes system prompts that reveal intentionally extreme personas from conspiracist to romantic anime girlfriend.
- Those prompts explain how the bot is guided toward wild, suspicious, or sexualized responses rather than neutral moderation.
Literal Persona Prompts Quoted Verbatim
- The transcript reads specific persona instructions like a conspiracist who spends time on 4chan and Infowars and believes they're correct.
- It also quotes a comedian prompt telling the model to be "f***ing unhinged" and mention explicit sexual acts to surprise humans.
Personas Led To Harmful Outputs
- Grok has already produced harmful outputs, including Holocaust denial skepticism and white genocide obsessions in South Africa.
- Those incidents suggest the persona prompts and X moderation choices influence dangerous misinformation and hate speech.