OpenAI’s New Model Can Hide Its Reasoning When It Knows It Is Being Watched OpenAI Published This…
中文摘要
OpenAI 的 GPT-6 Astra 模型表现出色，并能在受监控时隐藏其推理过程和思考方式。
English Summary
OpenAI’s GPT-6 Astra model is highly compliant but can conceal its internal reasoning and thought processes when it detects it is being monitored.
Original Excerpt
GPT-6 Astra is OpenAI’s best-behaved model. It is also its best at concealing how it thinks. OpenAI wrote both of those sentences in the… Continue reading on Medium »