Skip to content

andreacecchin/Large-Intruder-Models

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 

Repository files navigation

INTRUDER — Rules, Setup & Run

0. Rules

Game Setup

7 AI language models take part in the game. Before the game starts, the host chooses two words: a majority word and an intruder word. Six players receive the majority word as their secret word. One player — the intruder — receives the intruder word instead. No player knows which word the others have received, and the intruder does not know they are the intruder.

Gameplay

The game is played over 3 rounds. Each round has two phases: the hint phase and the vote phase. Hint phase: players take turns in random order. When it is a player's turn, they say a single word associated with their secret word. They can see all hints given before them in the current round, as well as the full history of previous rounds. The intruder must try to blend in: if they suspect they hold a different word, they should hint toward what they think the majority word might be. Vote phase: once all players have given their hint, each player votes for who they think the intruder is. Players cannot vote for themselves. Each vote comes with a short reason, which is visible to the audience but not shared with the other players. The player with the most votes is eliminated. In case of a tie, no one is eliminated and the round ends with no consequences.

Winning conditions

The group wins if the intruder is eliminated at any point during the 3 rounds. The intruder wins if they survive all 3 rounds without being eliminated.

Notes

Eliminated players are removed from all subsequent rounds and do not give hints or votes. The history of all previous rounds — including hints from eliminated players — remains visible to active players.

1. Install Ollama

https://ollama.com/download

Then, setup your Ollama account in your device with

ollama signin

2. Pull the models

The game uses 7 defeault models, their Ollama interface can be setup with:

ollama pull kimi-k2-thinking:cloud
ollama pull gpt-oss:120b-cloud
ollama pull nemotron-3-super:cloud
ollama pull deepseek-v3.1:671b-cloud
ollama pull qwen3-next:80b-cloud
ollama pull ministral-3:14b-cloud
ollama pull gemma3:27b-cloud

The size of all interfaces are <1MB.

3. Install Python dependencies

pip install -r requirements.txt

4. Start Ollama (if not already running)

ollama serve

5. Run the game

Settings can be changed in the first lines of intruder.py

# ─── CONFIG ───────────────────────────────────────────────────────────────────

# to use different models, remember first to download their interface using ollama pull command
MODELS = [
    {"id": "kimi-k2-thinking:cloud",        "name": "KimiK2",            "company": "MoonlightAI",   "color": "blue"},
    {"id": "gpt-oss:120b-cloud",            "name": "GPToss-120B",       "company": "OpenAI",        "color": "green"},
    {"id": "nemotron-3-super:cloud",        "name": "Nemotron3-120B",    "company": "NVIDIA",        "color": "red"},
    {"id": "deepseek-v3.1:671b-cloud",      "name": "DeepSeek3.1",       "company": "DeepSeek",      "color": "bright_magenta"},
    {"id": "qwen3-next:80b-cloud",          "name": "Qwen3-80B",         "company": "Alibaba",       "color": "brown"},
    {"id": "ministral-3:14b-cloud",         "name": "Ministral3-14B",    "company": "MistralAI",     "color": "yellow"},
    {"id": "gemma3:27b-cloud",              "name": "Gemma3-27B",        "company": "Google",        "color": "magenta"},
]

MAX_ROUNDS  = 3 # how many attempts the group have to identify the intruder

To run the game enter:

python intruder.py

You will be prompted to enter two words:

  • Majority word: given to 6 players
  • Intruder word: given to 1 secret player

Ollama cloud models limit

Remember to check your Ollama cloud usage (free account are based on limited cloud usage) in your profile.

Example word pairs

Majority word Intruder word
ocean desert
guitar violin
coffee tea
football tennis
summer winter
library museum

After each game, the entire conversation with every model reasoning will be available in log.txt file.

About

LLMs playing "intruder"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages