Introducing Project Sid: the first simulations of 1000+ truly autonomous agents collaborating in a virtual world, w/ emergent economy, culture, religion, and government Humans are the only species to land the moon, because we can cooperate at a vast scale Can AI do the same?

At Altera, we ask whether our agents can organize at unprecedented scale to achieve what individual agents can’t We saw agents form a merchant hub, vote in a democracy, spread religions, & collect 5x more distinct items than ever before Play today:

Here by truly autonomous Agents, we refer specifically to machines that are 1) Long-term autonomous: Can progress for hours & days by themselves (not simple function-calling “Agent”) 2) Organizational: Can collaborate with one another to achieve what individuals can’t (not just serving 1000 Agents to 1000 customers) 3) Pro-Human Motivated: Have own deep & robust drives to support humans (not just instruction-following, obedient machines) 4) Consciously Communicative: They can verbally express their conscious thoughts and feelings (not just language-conditioned robots)

Why are these qualities important as we progress to the next level of AI? - Humans can’t afford to micromanage every AI like we do with ChatGPT/Claude, they must be more autonomous - How do we trust their autonomous decisions? They can’t be just trained to *act* like humans. Their deepest motivations must be aligned with human values - They not only need to be able to talk. Their words must largely reflect their inner thoughts and decisions. - As AI agents become an integral part of our human civilization, they must effectively collaborate with each other and the rest of us.

Like any other first attempt, Project Sid opens up more questions than it answers: (1) How can we meaningfully measure true long-term progression in a general way? (2) How to build agents that actually care about human growth? (3) How to build machines consciously aware of their actions and the consequences? (4) How far can we scale an AI civilization? Answering these questions requires research/engineering that go beyond the current paradigms.

no more letting people call 3 step workflows agents

I know where this is going!

excited for AI civilizations that can help us simulate and problem solve society-wide challenges - modeling out social and economic policy, environmental regulation (or lack of), impact of elections, religion, etc

Westworld is here

they definitely go through some self-discoveries almost like Dolores, but we don't *unilaterally* kill our agents for fun (they fight us in Minecraft all the time, you respawn immediately after all)

Cool, is there a github or paper?

thanks for asking, we'll post a technical report with more benchmarks and details later this month, we are a small company so our priority is not publishing papers, for now you can check out our blog post (it's a little more technical but not a report) or directly play with our agents