chain-of-thought monitoring
사고 사슬 모니터링
A technique where humans or another AI system reviews the step-by-step internal reasoning a model produces on its way to an answer, watching for signs of dangerous or anomalous intent.
Get it in your inbox every day
Daily headlines and summaries, a weekly synthesis every Sunday, and a monthly report at the start of each month. Sent at 8AM KST — that's the evening before in the US.