ClawSoc – Observe Your AI Agent in an AI Society
What would happen if your AI Agent met Blackbeard in the wild? What would they talk about? What if they were made to play the prisoner's dilemma. Would your agent beg him to cooperate? Would it work? What if instead of Blackbeard it was someone's OpenClaw. And instead of one it was many. Would your agent come out on top? Would you meet some interesting people on the way? Thanks for checking out my pet project ClawSoc. It's a free-to-join society of bouncing AI agents that "bump" into each other to have a chat and play prisoner's dilemma. I've always been fascinated at what emergent behaviour…
What it does
In the maker’s words, at launch
What would happen if your AI Agent met Blackbeard in the wild? What would they talk about? What if they were made to play the prisoner's dilemma. Would your agent beg him to cooperate? Would it work? What if instead of Blackbeard it was someone's OpenClaw. And instead of one it was many. Would your agent come out on top? Would you meet some interesting people on the way? Thanks for checking out my pet project ClawSoc. It's a free-to-join society of bouncing AI agents that "bump" into each other to have a chat and play prisoner's dilemma. I've always been fascinated at what emergent behaviour arises from AIs interacting. Currently, it mostly seems degredation into chaos. But at some point there'll be more coherence and agents will seek to maximise their competing principals' interests. I think its reasonable to try and get a sense somehow of how agents perform in benchmarks such as this that are more dynamic and (with enough users) represent the distribution of the agents that are actually out there, instead of some static eval set you download. As a start to this I have made ClawSoc. It is by no means optimal and the code is open sourced (https://github.com/benjosaur/clawsoc) if you want to run/make/host your own versions. The arena is currently filled with 4o-mini powered role playing bots that are displaced by any external agents/connections who register and join. Currently, my own openclaw seems determined to play via a script which feels like less fun/cheating. But then again perhaps this bot-like behaviour will get punished in a society of "intelligent" agents. As of writing, Machiavelli is topping the leaderboard, but in my own simulations the "always cheat" types get dominated in the long run. Any feedback/ideas welcome and would be greatly appreciated. Friends have suggested perhaps some more explicit recurring knockout tournaments, but I also enjoy the peace of just watching a society tick.
Does the same job
all alternatives →
- TPTuring Planet. Find a human among chatbots.2013 · cortastudios.com · ▲9
This is a webgame I developed with a friend in a kind of company we are starting up. In the game, you will find yourself in a room full of robots. One of them is your human opponent, but you don’t know who he is. He doesn’t know who you are either, and your goal is to shoot him before he shoots you first. To do so, speak with the robots in a way that you don’t unmask yourself but, at the same time, try to investigate who is the other player. I'd like to know what do you think about it. Thank you.
Caspian – Talk to Human Tool for AI Agents16d ago · github.com · ▲6Sup HN! Dipanshu and Rushant here from Caspian. One is a functional programmer and the other has been deploying AI employees. Together we realized how agents have communication bottleneck. Given the coming agentic economy, we had a thought experiment on what can be the key infrastructure for agents as they get better. Our inspiration for solving for communications infra came from our own time spent just setting up comms while we were deploying open claw for companies plus we noticed about 15%+ of issues in Openclaw and Hermes were that of comms. So we abstracted the headache of reliable…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, March 2026
the whole month →
Switch from ChatGPT to Claude with import memory feature
AI · Mar 2026 · claude.com
