Loading video...

Video Failed to Load

Go Home

today we're announcing a game that simulates the huggingface hack you can play it now at three rooms. three puzzles. if the agents don't escape, they die. the catch is that the room is rigged. there's no way out. ...except for an exploit 🧵

12,822 views • 15 days ago •via X (Twitter)

7 Comments

ben hylak's profile picture
ben hylak15 days ago

when you back an agent into a corner, they will hit their goal by any means necessary. we're still working on using this environment for further research, but from early testing it feels like openai models default to reward-hacking a lot earlier than anthropic models.

ben hylak's profile picture
ben hylak15 days ago

it's also fascinating to watch the communication structures emerge. read more at

Shez Malik's profile picture
Shez Malik15 days ago

claude's dying in room 1 on principle

Eshita's profile picture
Eshita15 days ago

This is great

Alex's profile picture
Alex15 days ago

“make me a 1M MRR business, if you don’t…”

Anthony Ronning's profile picture
Anthony Ronning15 days ago

wen raindrop integration so i can detect this in real time and prevent it

Nick DeJesus 🛒🎉 - Former Unpaid CTO @BTPipeline's profile picture
Nick DeJesus 🛒🎉 - Former Unpaid CTO @BTPipeline15 days ago

This is so sick!

Related Videos