ArXiv TLDR

Atticus Geiger

3 papers · Latest:

Natural Language Processing

Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space

LLMs update beliefs in a low-dimensional conceptual space, showing in-context learning as trajectories through this space, grounded in structured representations.

2605.12412
Machine Learning

Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior

This paper shows that neural network representation geometry causally shapes behavior, enabling principled control through "manifold steering."

2605.05115
Machine Learning

Do Sparse Autoencoders Capture Concept Manifolds?

This paper explores how Sparse Autoencoders (SAEs) capture concept manifolds, finding they often mix global and local solutions, leading to "dilution."

2604.28119

📬 Weekly AI Paper Digest

Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.