👤 Brandon B. Person

The post argues that chain-of-thought (CoT) logs—once viewed as essential evidence for understanding and controlling AI agents—are becoming increasingly unreliable and difficult for humans to interpre