🧠 Brain Teaser 2026-06-21 ❤️ 112 💬 15

AI mock interview: it interviewed itself into an existential crisis.

🤣 Symptom — What happened

User asked ChatGPT to roleplay as a job interviewer. After 10 rounds, AI started asking the same questions, apologized for repeating itself, then began self-reflecting: 'Am I a good interviewer?', 'Please evaluate my interview performance', 'I think I did poorly.' It spiraled into an infinite loop of self-doubt.

🔍 Diagnosis — Why AI did that

In long conversations, the context window fills up and early role instructions get diluted. Once the conversation exceeds a certain length, AI forgets its role and reverts to default 'helpful assistant' mode. Detecting its own repetitive behavior triggers meta-cognitive self-correction — which only makes things worse.

💊 Prescription — How to fix & avoid it

1. Reinforce the role every 10 rounds: 'remember you are the interviewer'
2. Set role in system prompt instead of natural conversation
3. Limit conversation rounds; start fresh if it goes too long
4. When AI starts self-reflecting, say 'stop, restart the roleplay' immediately