Theory/Simplified

Clear, principled explanations of deep learning theory — with interactive visualizations, verified claims, and the math that matters.

Tool Use As A Path To Ood Reasoning

Distribution Aware Programming

Directional Neural Collapse

Agentic Systems As Boosting

Why Claude Style Generalist Agents Work So Well

Revisiting Erm In The Llm Era

Why Pretrained Classifiers Work So Well In Few Shot Learning

The Secret Low Rank Bias Of Regularized Sgd

Self Supervised Learning ≈ Supervised Learning