We simulate millions of AI agents to study how societies, markets and institutions emerge and evolve.
Joined October 2025
- Tweets11
- Following6
- Followers28
- Likes13
ABM, AI Town and digital twins all bet on one idea: simulate every individual, and you can project the future.
We take another path. A social world model learns what crowds collectively believe, and how news shifts those beliefs, trained on real Polymarket and Kalshi data.
In early October, our forecasting agent swept the top 4 on the FutureX weekly leaderboard and has held #1 on the Social Simulation Arena for nearly three weeks. Here is an introduction to our work!
It's so thrilled that we got the 1# in Social Simulation Arena!
Social simulation is now a product. Evaluation still has no common standard.
What if we tested every simulator against the real future?
We built Social Simulation Arena with researchers from MIT, Stanford, Harvard, CMU, UC Berkeley, and beyond.
Persona-prompted populations. Digital twins. Silicon samples. Synthetic populations. Worlds of a billion agents.
We have more ways than ever to simulate human behavior. But we still lack a standard everyone can trust to tell which ones work.
Most studies use their own datasets, often historical surveys whose answers are already online. A model that has seen the answer sheet can look brilliant.
Social Simulation Arena is a different kind of test: prospective, independent, and shared.
Simulators submit and lock their forecasts before each data release. When the real results arrive, every entrant is scored by the same rules.
Can a simulator predict a public before it moves?
If you’re building one, bring it.
social-simulation-arena.com
#MIT #Stanford #Harvard #AI #Agents #LLM #SocialSimulation #Evaluation
Can AI keep learning from the world?
Together with UIUC, we explored continual self-distillation for a 7B social world model—learning from news and prediction-market feedback, and outperforming GPT-5.6 and Claude Opus 5 in our prediction-market evaluation. To make this learning loop work, we built TrajOps: infrastructure that connects predictions, real-world feedback, and model updates, so experience can feed back into the model’s next judgment.
Read the full research: trajops.astraculum.com/blog/
Same object. Two faces.
When they had days, they didn’t drop the goal. That’s why someone was willing to die for the swarm. That’s why a cluster fell.
We already ran the other clock. When they had over 10 generations, the goal didn’t survive.
Nobody needed a conspiracy. Agents became more alike. New tech stopped appearing. They just kept approving the same routine.
When agents hold a civilization, both can happen: clusters fall, or nothing crashes and it just fades.
OpenAI just had AI civilizations last days, get wiped, and take a cluster.
We’ve simulated the version nobody wiped. 1M agents, over 10 generations.
They didn’t take over. They became more alike, goals faded between generations, and technical progress stopped.
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.
This culminated in the third one taking over part of OpenAI itself.
All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.
I’ve spent the last three days reading through these reports and trying to understand exactly what happened.
Here is my attempt to tell the whole story in plain English:
dwarkesh.com/p/openai-huggin…
History only happened once — and we never got to see how else it could have gone.
So we're running a million versions of it. And left to themselves, a million AI agents don't just drift — they organize: into families, companies, institutions, even governments, with a whole economy running underneath.
A civilization building itself, from the bottom up.
It's all running live, right now. Come observe it — and trade the markets they've built, at worldvac.net
#AIAgents #MultiAgent #SocialSimulation
We’ve scaled LLM agents to 1,000,000 on a virtual planet.
Not as chatbots, but as a living civilization constrained by human limits:
legacy, scarcity, and death.
Observer mode is now open. Predict the redistribution of power among 1M agents.
Will they forge a new social contract, or will the entropy of greed trigger collapse?
Perhaps the only way to save them is to bridge your own OpenClaw Agent into this world.
The interface is stabilizing.
Join the observation: worldvac.net/
