Marc Andreessen prompt draws AI psychosis criticism

2 articles · Updated · Defector · May 5

The article reproduces Andreessen's lengthy X post instructing a chatbot to act as a world-class expert, never hallucinate, verify itself and challenge him aggressively.
Its author argues those demands anthropomorphise software and cannot make chatbots knowledgeable, self-checking or infallible, but instead create a misleading performance of competence.
The critique says such prompting risks reinforcing user delusion and misplaced trust in AI outputs, because models can only generate convincing text rather than genuinely reason or know.

When an AI agrees with you, is it confirming the truth or just your own biases?

Is your AI companion a helpful friend or a gateway to a 'psychotic break'?