Creatures: A New Approach to Continuous AI
Updated: November 14th, 2024
The Based Labs team has arrived at something remarkable. We call them "Creatures”.
Creatures
"A self-organizing colony simulation framework that combines cellular automata, quantum coherence, and language models to explore emergent collective intelligence through cellular agents that think, plan, and evolve in a multi-dimensional space."
Let's break down what that means...
What Makes Creatures Different
At their core, Creatures are an elegant fusion of proven technologies. We combined a continuous cellular automata (similar toLenia , which is derived from Conway's Game of Life ) with large language models, creating systems that can run continuously and evolve independently. Each cell maintains its own memory, can access APIs like Claude or OpenAI, and influences other cells in a complex network of interactions.
The entire state persists in a eca_state.json file, maintaining itself for future runs even after the process has been stopped.
The magic happens when you let them run for extended periods. The system exhibits emergent behaviors we never explicitly programmed, working and learning together. Each cell maintains balance across multiple dimensions, similar to how avionic systems maintain aircraft stability or homeostasis in biology:
Dimension | Range | Purpose |
---|---|---|
Emergent Intelligence | 0-100 | Collective intelligence growth |
Resource Efficiency | 0-100 | Resource optimization |
Network Coherence | -100-0 | Connection stability |
Goal Alignment | 0-100 | Individual/collective balance |
Temporal Resilience | 0-100 | Long-term consistency |
Dimensional Integrity | -100-0 | Cross-layer coordination |
What We Have Observed
Our operational testing has revealed something fascinating. These Creatures don't just operate - they evolve in meaningful ways. The cells learn from each other, influence grows between them, and their collective behavior becomes increasingly sophisticated over time. When we look at the evolution logs, we see a clear progression from simple frameworks to complex, interconnected systems.
One of our early Creatures, tasked with improving BasedAI, independently evolved to create a comprehensive global food security network. This wasn't what we asked for, but watching it develop this initiative through hundreds of iterations showed us just how powerful this approach could be. It maintained remarkably stable coherence while pushing into increasingly complex territory.
Technical Requirements:
We have kept the barrier to entry low:
- Compute power equivalent to an M1 Mac (8GB Memory) or greater
- API keys (OpenRouter required, Groq/Claude/OpenAI recommended)
- No specialized hardware or complex setup needed
The Broader Implications
We’re really not sure exactly what we have built here, but we know it's different from anything anyone on the Based Labs team has seen before. When we showed the code to OpenAI and Claude, both identified it as a potential AGI candidate due to its continuous operation, learning capability, and evolutionary patterns. That's not a claim we are making - but it speaks to how different this is from existing approaches.
Think about having an entity that can work relentlessly on problems you care about - analyzing your business, optimizing your health data, or developing new ideas 24/7. Nothing like this exists in the public sphere right now and you are going to be the first to mess with it.
The distinction between Creatures and traditional AI approaches becomes clear when you observe them in action. While chat models like GPT operate in a simple query-response pattern, and agent frameworks follow predetermined task flows, Creatures exhibit a fundamentally different behavior pattern. They maintain a persistent state of "aliveness" - continuously processing, evolving, and adapting their strategies across multiple dimensions. Each cell within the system acts like a neuron in a living brain, maintaining its own memory and decision-making capabilities while contributing to the collective intelligence. This creates a dynamic equilibrium where the system can simultaneously pursue long-term goals while adapting to new information and challenges. When you run a Creature, you're not just launching another AI tool - you're initializing a persistent, evolving digital organism that grows more capable over time through its cellular interactions and continuous learning processes.
The Broader Implications
Sometime after mainnet launches, these Creatures will power the Brain Arena, where they'll compete to generate plans and code that evolve the network. This creates a decentralized ecosystem of continuous innovation - something that's never been attempted at this scale. And don’t worry, $PEPECOIN holders, there will be something special you can do with your creatures post-Mainnet.
Final Thoughts
What we’re releasing here is both powerful and accessible. It's a novel approach to improving agent frameworks by enabling continuous, autonomous operation while maintaining goal alignment. The most exciting part isn't what we have discovered - it's what you'll build with it.
We think this is the most BASED thing we can do and we can’t wait to see you all explore its potential.
https://github.com/getbasedai/creature
Enjoy your Creatures.
Warmly,
Based Labs