What We Did

There's been a lot of drama with OpenAI, other companies, and governments over AI safety.

And as Jan Leike famously once said, "Learn to feel the AGI." So that's what we did. When you stare long enough into the darkness, it stares back...

So from your favorite AI safety reading group at UCSB, we present LLMs that play against each other in the most popular WWII deception and social deduction game. BECAUSE RECURSIVE SELF-IMPROVEMENT DOESN'T SCARE US!!!

How we Welcomed our AI Overlords

1. Passed in relevant information to each player: Their known information (board state, roles, etc.)
2. Got relevant strategy to the stage using RAG and calling a vector database
3. The rules of the stage and their goal
4. Gave a history of the discussions between players (discussed before voting and after enacting a policy)

Thank them for increasing P(doom)

1. Pinecone Vector DB
2. OpenAI Embeddings + GPT-4o

this project is (mostly) satire. it is not meant to be taken seriously.

Built With

• openai
• pinecone
• python