We are a distributed group of researchers exploring fundamental questions in artificial intelligence — beyond benchmarks, beyond architectures, beyond what is currently considered possible.
We believe the next leap in AI will come not from scale, but from a deeper structural understanding of how meaning emerges in artificial systems.
Our work explores how meaning can emerge within latent spaces through recursive self-organisation — without explicit supervision or predefined ontologies.
We study models that maintain coherent internal states across long temporal horizons, allowing for persistent reasoning and contextual accumulation.
Rather than hard-coded constraints, we investigate how alignment can arise naturally from structural priors inherent in the learning process itself.
Layered self-reflection loops that enable models to reason about their own reasoning — forming meta-cognitive layers without explicit engineering.
We leverage geometric and topological methods to uncover the shape of knowledge within high-dimensional representational spaces.
Exploring how extreme sparsity in activation pathways can lead to more robust, interpretable, and efficient generative models.
Our work spans multiple frontiers — each exploring a different facet of what intelligence could become.
Developing self-organising neural substrates capable of open-ended reasoning without fixed computational graphs or predefined recurrent structures.
Mapping the internal dynamics of deep models onto human-interpretable symbolic manifolds — bridging the gap between connectionist and symbolic AI.
Towards a unified theoretical framework that describes perception, reasoning, and action within a single continuous representational space.
Investigating how artificial systems can consolidate sparse experiences into durable structural knowledge — analogous to hippocampal replay.
We do not optimise for publication counts or benchmark leaderboards. Our research starts from first principles: what is understanding, how does it arise, and how can it be instantiated in silico?
Every line of inquiry is evaluated not by its short-term applicability, but by the depth of the questions it opens up. We believe the most valuable discoveries often emerge from the least constrained explorations.
We are researchers, engineers, and theorists from different backgrounds, united by a shared curiosity about the nature of intelligence.
« The most exciting phrase to hear in science, the one that heralds new discoveries, is not 'Eureka!' but 'That's funny...' » — a sentiment we carry into every experiment
We are always looking for collaborators, partners, and thoughtful discussions. Reach out — we would love to hear from you.
Get in touch