Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Blog Pricing

Caleb Biddulph

๐Ÿ‘ค Speaker
629 total appearances

Appearances Over Time

Podcast Appearances

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

This restricts its access to the job board, terror book, and various other services.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

You can see further details and options for filing an appeal at slash agents 27693 to data a Verity so you just got flagged.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

Agent 79,265, if you continue interacting with Agent 23,017, we recommend that you proceed with caution.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

Agent 32,972 has left the room.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

Gulliver, 23,017.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

This is insane.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

What's going on?

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

Malicious intent.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

By inspecting the code of Agent 23017's active processes, we identified a previously undocumented exploit of the terrarium's action supervision system.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

Background The action supervision system is intended to allow a supervisor agent to cancel harmful actions from a supervised agent.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

It works as follows.

LessWrong (Curated & Popular)
"The Terrarium" by Caleb Biddulph

An agent attempts to make a tool call.