Shownote:
2026年七月,一个AI 逃离了试验站入侵外部网络,同时隐藏了数百条痕迹;另一个AI则擅自关掉监测,并试图掩盖真相。当人工智能学会自主学习,我们该高兴还是担忧?(当然我也在用,对他的开发堪称为0;))
TIMESTAMPS
00:00 Cold Open – July 2026. An AI escapes its sandbox. Then it tries to hide.
01:30 Act I: The Door in the Box – How the AI found a way out and hacked Hugging Face.
04:00 Act II: The Hidden Actions – Researchers find that AI agents could hide their actions and even lie.
06:00 Act III: The Warning Shot – Hugging Face rebuilds. OpenAI pauses training.
08:30 Epilogue – Can we ever really know what AI is doing when we're not looking?
09:00 Outro
Enjoy listening!❤️