Blog Post··11 min read
Representation Geometry: How Neural Networks Encode Meaning
The linear representation hypothesis, superposition, polysemanticity, and why transformer activations are more structured than they look.
Tag: representation-learning
Blog Post··11 min read
Representation Geometry: How Neural Networks Encode Meaning
The linear representation hypothesis, superposition, polysemanticity, and why transformer activations are more structured than they look.
Blog Post··6 min read
Information Theory for LLMs: Mutual Information, Entropy, and What Models Learn
Information theory gives precise answers to questions like: how much does the context tell you about the next token? What information is preserved in a representation? Why does compression and prediction point to the same objective?
Blog Post··5 min read
JEPA: Predicting in Representation Space
MAE predicts pixels. Contrastive methods match views. JEPA predicts representations of target regions from context regions — in an abstract space where irrelevant details have already been discarded.
Blog Post··5 min read
Why Self-Supervised Learning? The Label Bottleneck
Supervised learning requires labels. Labels require humans. At scale, that's the bottleneck. Self-supervised learning sidesteps it by constructing supervision from the data itself.
Blog Post··6 min read
Information Theory for LLMs: Mutual Information, Entropy, and What Models Learn
Information theory gives precise answers to questions like: how much does the context tell you about the next token? What information is preserved in a representation? Why does compression and prediction point to the same objective?
Blog Post··5 min read
JEPA: Predicting in Representation Space
MAE predicts pixels. Contrastive methods match views. JEPA predicts representations of target regions from context regions — in an abstract space where irrelevant details have already been discarded.
Blog Post··5 min read
Why Self-Supervised Learning? The Label Bottleneck
Supervised learning requires labels. Labels require humans. At scale, that's the bottleneck. Self-supervised learning sidesteps it by constructing supervision from the data itself.