Andrej Karpathy’s AI Council uses multimodel peer review
The author explores setting up a multi-model "AI council" where a prompt is sent to several models for peer review before a chairman model synthesizes a final answer. While the setup helps mitigate reliance on a single confident chatbot and provides valuable insights for high-stakes decisions, the author argues it is impractical for daily use due to high token usage, latency, and setup overhead. Open article →