20 citations · 95 across the 49 of their papers we have counts for
56 papers
A Phenomenology of Neural Geometry in Vision-Language-Action Models
Mozes Jacobs, Aiden Swann, Mathilde Papillon +16
Robotic manipulation is inherently geometric, but what shape does the world take inside the policy that performs it? Prior work on vision-language-action (VLA) policies establishes…
Multi-Dimensional Featurizers Yield Fine-Grained Steering and Data Shaping in Robot Foundation Models
Aiden Swann, Mozes Jacobs, Mathilde Papillon +12
Can better feature discovery improve control over robot behavior through internal interventions and training data? Recent work in vision and language models reveals multidimensiona…
Causal Analysis of Robot Foundation Models Enables Efficiency Gains
Aiden Swann, Mathilde Papillon, Mozes Jacobs +13
Vision-language-action (VLA) models are a promising approach to general purpose robot manipulation. While the internal computation of vision-language models, which compose the back…
Language Model Activations Inhabit Privileged Error-Correcting Basins
Matthew Finlayson, Francisco Pernice, Eric Todd +16
Language models exhibit remarkable robustness, continuing to produce coherent text even when their activations are perturbed by interventions like linear steering. We hypothesize t…
Forking Fast: Efficiently Estimating Uncertainty Dynamics in Text Generation
Eric Bigelow, Amir Zur, Satchel Grant +7
LLM reasoning is stochastic, and so understanding a model requires grappling with the distribution of reasoning chains that it might produce for a given question, i.e., its uncerta…
Position: It's Time to Optimize LLMs for Self-Consistency
Itamar Pres, Belinda Z. Li, Laura Ruis +6
Despite ever-increasing sophistication in language model (LM) pre- and post-training pipelines, many important failures persist: models overcondition on user framing ("sycophancy")…
Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds
Thomas Fel, Matthew Kowal, Mozes Jacobs +22
What is the geometry of a visual percept? The most widely used protocols for decomposing neural network representations into interpretable parts treat concepts as isolated directio…
Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal
Leon Bergen, Usha Bhalla, Sidharth Baskaran +14
Language-model post-training is the main stage at which model behavior is shaped, yet it still largely involves optimization of scalar rewards that summarize diverse desiderata. Th…
Why Larger Models Learn More: Effects of Capacity, Interference, and Rare-Task Retention
Jing Huang, Daniel Wurgaft, Rachit Bansal +6
Larger models learn tasks smaller models do not. What drives this phenomenon? We develop a simple phenomenological argument that power-law scaling already suggests that a larger mo…
Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space
Eric Bigelow, Raphaël Sarfati, Daniel Wurgaft +5
Large Language Models (LLMs) update their behavior in context, which can be viewed as a form of Bayesian inference. However, the structure of the latent hypothesis space over which…
Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior
Daniel Wurgaft, Can Rager, Matthew Kowal +13
Neural representations carry rich geometric structure; but does that structure causally shape behavior? To address this question, we intervene along paths through activation space…
Arithmetic in the Wild: Llama uses Base-10 Addition to Reason About Cyclic Concepts
Sheridan Feucht, Tal Haklay, Usha Bhalla +9
Does structure in representations imply structure in computation? We study how Llama-3.1-8B reasons over cyclic concepts (e.g., "what month is six months after August?"). Even thou…
Do Sparse Autoencoders Capture Concept Manifolds?
Usha Bhalla, Thomas Fel, Can Rager +9
Sparse autoencoders (SAEs) are widely used to extract interpretable features from neural network representations, often under the implicit assumption that concepts correspond to in…
Features as Rewards: Scalable Supervision for Open-Ended Tasks via Interpretability
Aaditya Vikram Prasad, Connor Watts, Jack Merullo +4
Language models trained on large-scale datasets have been shown to learn features that encode abstract concepts such as factuality or intent. Such features are traditionally used f…
The Shape of Beliefs: Geometry, Dynamics, and Interventions along Representation Manifolds of Language Models' Posteriors
Raphaël Sarfati, Eric Bigelow, Daniel Wurgaft +6
Large language models (LLMs) form implicit beliefs (posteriors over latent variables) from prompts, but we lack a mechanistic account of how these beliefs are encoded in representa…
From Isolation to Entanglement: When Do Interpretability Methods Identify and Disentangle Known Concepts?
Aaron Mueller, Andrew Lee, Shruti Joshi +3
A goal of interpretability is to recover disentangled representations of latent concepts (features) from the activations of neural networks. The quality of features is typically ev…
Priors in Time: Missing Inductive Biases for Language Model Interpretability
Ekdeep Singh Lubana, Can Rager, Sai Sumedh R. Hindupur +13
Recovering meaningful concepts from language model activations is a central aim of interpretability. While existing feature extraction methods aim to identify concepts that are ind…
Are language models aware of the road not taken? Token-level uncertainty and hidden state dynamics
Amir Zur, Atticus Geiger, Ekdeep Singh Lubana +1
When a language model generates text, the selection of individual tokens might lead it down very different reasoning paths, making uncertainty difficult to quantify. In this work,…
The Impact of Off-Policy Training Data on Probe Generalisation
Nathalie Kirch, Samuel Dower, Adrians Skapars +3
Probing has emerged as a promising method for monitoring large language models (LLMs), enabling cheap inference-time detection of concerning behaviours. However, natural examples o…
Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation Steering
Eric Bigelow, Daniel Wurgaft, YingQiao Wang +4
Large language models (LLMs) can be controlled at inference time through prompts (in-context learning) and internal activations (activation steering). Different accounts have been…
Into the Rabbit Hull: From Task-Relevant Concepts in DINO to Minkowski Geometry
Thomas Fel, Binxu Wang, Michael A. Lepori +8
DINOv2 is routinely deployed to recognize objects, scenes, and actions; yet the nature of what it perceives remains unknown. As a working baseline, we adopt the Linear Representati…
How Do LLMs Persuade? Linear Probes Can Uncover Persuasion Dynamics in Multi-Turn Conversations
Brandon Jaipersaud, David Krueger, Ekdeep Singh Lubana
Large Language Models (LLMs) have started to demonstrate the ability to persuade humans, yet our understanding of how this dynamic transpires is limited. Recent work has used linea…
Emergence of Hierarchical Emotion Organization in Large Language Models
Maya Okawa, Bo Zhao, Eric J. Bigelow +4
As large language models (LLMs) increasingly power conversational agents, understanding how they model users' emotional states is critical for ethical deployment. Inspired by emoti…
In-Context Learning Strategies Emerge Rationally
Daniel Wurgaft, Ekdeep Singh Lubana, Core Francisco Park +3
Recent work analyzing in-context learning (ICL) has identified a broad set of strategies that describe model behavior in different experimental conditions. We aim to unify these fi…