Ziqian Zhong's banner

Ziqian Zhong

@fjzzq2002 • 2,249 subscribers

It takes a miracle to align AIs. Please be a part of that! Currently PhD @CarnegieMellon, prev @MIT @pika_labs @TransluceAI

Shorts

🔭 We’re releasing Hodoscope: an open-source tool for unsupervised behavior discovery. It lets you visually explore and compare agent behaviors at scale. It helped us discover a novel reward hacking vulnerability in Commit0 - with just a couple minutes of human effort.

🔭 We’re releasing Hodoscope: an open-source tool for unsupervised behavior discovery. It lets you visually explore and compare agent behaviors at scale. It helped us discover a novel reward hacking vulnerability in Commit0 - with just a couple minutes of human effort.

76,862 просмотров