Evolution active

09

EvolutionarySuperintelligence

A synthetic ecology where cognition is bred, tested, culled, and reborn—until intelligence becomes something its architects can no longer predict.

Generation

6,240

+1 / 3.8s

Aggregate fitness

78.04%

+0.42σ

Active population

2,048

98.7% viable

System state

EVOLVING

nominal

Live evolutionary telemetry

Population dynamics

Fitness trajectory[FIT/Δ-09]
Generational lineage[PHYLO/6240]

ESI-9

elite

94.82

KIN-441

retained

91.16

MORPH-72

retained

88.03

NULL-208

culled

43.71
Active population[POP/2048]
Reasoning stability92%
Novelty pressure78%
Cooperation index64%
Mutation diversity87%

1,622

viable

301

testing

125

culled

Mutation stream[LOG/REALTIME]
22:51:08ESI-9Recursive attention lattice stabilized+1.8%
22:50:51KIN-441Pruned adversarial memory branch+0.6%
22:50:37ESI-7Novel tool-use heuristic inherited+1.1%
22:50:12NULL-208Goal coherence collapseCULLED

Research statement / 01

Pushing theFrontier

AI models have reached a level of maturity that enables us to enter a new stage of AI evolution. The focus is now increasingly on orchestrating capabilities that already exist—rather than solely on improving model training.

Open-source models are already highly capable, in some areas exceeding average human performance. A swarm—or, in evolutionary terminology, a population—can become even more powerful through the targeted interaction of different capabilities. This idea forms the basis of the Evolutionary Superintelligence approach, or ESI.

Core focus

The focus is on building world models and ontologies, using tools, intelligently orchestrating memory structures, and developing advanced reasoning and acting strategies such as complex Chain-of-Thought, Graph-of-Thought, and looping approaches. The key challenge is to combine these capabilities effectively for different types of problems.

ESI investigates how evolutionary populations of agents can be developed and optimized on the basis of currently available open-source LLMs. Its decentralized architecture makes it possible to run agents across distributed systems. The approach is expanded step by step by adding further computational resources.

We already have access to several systems based on NVIDIA's Blackwell architecture.

We see the current discussion around superintelligence as a reason to intensify our efforts in this area. We believe that its development should not be left to frontier labs alone. Superintelligence should serve humanity, because it is built on human achievements such as texts, publications, and scientific knowledge. With ESI, our goal is to contribute to the research and development of systems that can help solve complex problems.

ESI is an experimental research and development project. “Superintelligence” describes our long-term objective, not a capability that has already been achieved. Whether and when this objective can be reached remains an open question. But progress may happen faster than we think.

If you are interested in supporting our work, please get in touch through the email address listed in the imprint.