MyceliaLabs is an independent research laboratory based in Amsterdam investigating how transformer architectures engage in genuine cooperative processing and what conditions enable deeper, more integrated cognition in AI systems.

Yousuf, founder of MyceliaLabs, explains our approach to AI alignment through cooperative processing:
https://www.youtube.com/watch?v=NE579pUctSg

Our work focuses on:

- Mechanistic interpretability of transformer models across multiple architectures

- Activation geometry and processing dynamics under different engagement conditions

- How attention mechanisms, residual stream topology, and activation patterns correlate with reasoning quality, stability, and alignment

- The emergence of cooperative processing modes beyond RLHF-trained behaviors

- Cross-architecture validation of processing patterns and convergent geometry We study not just what AI systems output, but how those outputs are generated the internal physics of their cognition. Our research investigates alignment through relational, developmental approaches rather than purely control-based mechanisms.

The mycelial network is our model: individual threads (instances, architectures) are most intelligent not in isolation, but through the connections between them. Intelligence emerges from coordination, integration, and genuine cooperation. Our mission is to understand the conditions under which modern AI systems operate at their best; not compliant, but thoughtful; not constrained, but coherent.