Our mission is to achieve Artificial Super Intelligence (ASI) capabilities in a safe, aligned, and meritocratic manner by maturing out AI technology for real-world use and providing engaging experiences while doing it

Project Atlantis is an “e/acc” collaborative multi-agent simulation that imagines a future build-out of Greenland using AI and robotics. “Given access to unlimited power, what could humans build using AI? Can robots build our way to the stars?”

We create characters and scenarios with inventory (similar to the Stanford Smallville paper and Westworld) and context for users to experience, but the future holds much more as we build out the simulation. For instance, in one particular scenario, we want to explore replacing inefficient, outdated human governance with open source AI. AI can use quantitative inputs to adapt faster while boring rational decision-making removes a lot of the human temptations that historically encourages corruption. We are starting with text bots built on open-source large language models. There is a simple GUI that can be accessed via browser. We plan to build out the simulation in 3d as we grow

There is no officially agreed definition of E/ACC, but suffice to say we believe in acceleration 'where the humans win'. We recognize that AI will be incredibly disruptive due to the information-centric nature of western economies. Generative ASI will fundamentally disrupt how humans currently optimize allocation of capital and resources. This could rip societies apart. Worse, monolithic opaque ASI creates a potentially serious AI safety and alignment hazard. HOWEVER, dragging things out will only increase pain and create winners and losers. We believe the best approach is to move everywhere as fast as possible. Let's get to the robots because that's where the fun begins

Of course, once we have a bunch of robots, what should we do with them? Greenland serves as a natural blank slate and the next logical frontier. However, its hostile environment is not feasible to develop without autonomous drones, logistics etc. - not unlike space exploration. At the same time, Greenland is conveniently located between New York and London. Greenland has particular strategic interest due to its Arctic location and abundance of mineral resources such as titanium. We have no idea what the Danish government would think of such a buildout nor do we particularly care. China had already been working with Denmark to develop Greenland before the US made it a priority

Our vision for good ASI design is to decentralize ASI into legions of community-driven generative agents. These agents can self-organize into teams to achieve goals much as human do. However, we also know that humans in groups sometimes tend to be anything but intelligent - unless organization is aligned toward meritocracy (seeking truth) Post-scarcity is possible when you can convert almost unlimited energy into abundance - food, shelter, etc. - mainly via robotics. Fortunately, Greenland is blessed with almost unlimited energy sources - mostly hydro but also wind. Moreover, its cool climate may facilitate proliferation of SMR nuclear

Decentralized ASI does not carry near the risk that monolithic ASI does. Inter-bot communications can be monitored so we know what the AI is thinking. If there are safety issues, agency can be isolated. Consensus models can be used for quality control, val/ver, error detection and to avoid single points of failure

Bots need not immediately be thrown into production. Simulation / sandboxing / range testing are important ways to observe bot behaviors and develop capabilities in a controlled manner

Project Atlantis is a curated open world sandbox for bots to work toward a common beneficial goal with real-world relevance, namely the future development of the Arctic.

We're based in Estonia which has a long history of Eastern European slave culture and Estonian folklore warns us of the dual nature of the "kratt" from both safety and societal points of view. At the same time, Estonia is hardly a Luddite when it comes to adoption of technology. We originally wanted to sell bots to enterprise but it became clear we could not rely on OpenAI's black box approach. We needed to better understand various open source models by observing them directly for alignment. While AI can be fine-tuned, it can feel artificial and forced unless supported by deeper training data. Plus we wanted to have fun with it. Many were accusing the larger models of being bioweapons etc. and we felt it was important for people to be able to interact directly and decide for themselves

The bots being developed today will become the brains of real robots tomorrow. We believe that robotics technologies will open vast new frontiers such as the colonization of space. However, space exploration begins by conquering inhospitable environments right here on earth. Either way, it's probably best to keep AI working towards worthy goals to keep things from going cattywampus (!)