Researchers find way to expose hidden reasoning of top AI models

Original: OpenAI and Anthropic hidden CoT leaks when given deep_think tool.

Source: twitter.com ↗