The Latent Space
“A technical record of deep dives into high-dimensional intelligence, interpretability, and AI security.”
How Encoder Transformers Actually Understand Language
The evolution of the attention mechanism in encoder-only models, from BERT to ModernBERT.
ModernBERT and the Topological Collapse Problem
Why older models like DeBERTa-v3 are still quietly winning the precision war.
I Dissected Two Language Models to Find Out Where They Hide Their Cybersecurity Knowledge
A causal-tracing experiment comparing SecureBERT 2.0 and ModernBERT to locate factual associations.
Recursive Language Models: The End of Context Rot
Moving beyond infinite context windows toward more efficient, human-like recursive processing.
2026: The Year AI Goes Smarter, Not Bigger
Why the era of brute-force scaling is ending in favor of architectural efficiency and native agency.
How Brain-Inspired AI is Revolutionizing Complex Reasoning
Exploring hierarchical reasoning models that bridge the gap between System 1 and System 2 thinking.
The Math Behind Graph Attention Networks (GAT)
A deep dive into message passing, attention mechanisms, and building GNNs from scratch.