slides
Ideas, papers, and questions from group meetings.
Research talks & reading notes, collected over time.
6 presentations
AI for Mathematics
Prover Series
A tour of formal theorem proving, from Lean and DeepSeek-Prover to Seed-Prover and geometry.
View slides ↗
Latent Reasoning
An Introduction to Latent Reasoning
An introduction to reasoning in continuous states, recurrent Transformers, Coconut, and mechanistic interpretability.
View slides ↗
Interpretability
Open the Black Box: Transformer Circuit Thread
Reading notes on Transformer circuits, circuit tracing, and the biology of language models.
View slides ↗
Generalization
Memorization or Generalization: Some Empirical Results
Empirical studies of compositionality, mathematical reasoning, and number understanding in Transformers.
View slides ↗
Model Architecture
Back to RNN?
From linear attention and state space models to Mamba, Griffin, and recurrent sequence models.
View slides ↗
Generalization
Grokking
Why can generalization emerge long after memorization? A look at grokking and its mechanisms.
View slides ↗