Reading Notes

Notes and summaries from reading ML/AI papers (and some blog posts). All credit to the content in the papers and blog posts goes to the original authors.

Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking

circuits in fine-tunes

7 min read · March 16, 2026

2026 · distillation
Causal Abstractions of Neural Networks

Causal abstractions

4 min read · March 15, 2026

2026 · distillation
End-to-End Test-Time Training for Long Context

TTT-E2E

6 min read · February 26, 2026

2026 · distillation
Interpreting Physics in Video World Models

physics in video models \+ V-JEPA background

2 min read · February 26, 2026

2026 · partial-read · distillation
Language Models use Lookbacks to Track Beliefs

lookback mechanisms

12 min read · February 21, 2026

2026 · distillation