43 citations · 62 across the 23 of their papers we have counts for
24 papers
Matryoshka attribution: Learning to attribute language model outputs to representations and weights
Aryaman Arora, Kirill Acharya, Nathan Hu +4
Attributing language model outputs to their internal computations is an open problem in interpretability. Existing methods, which use causal interventions, gradients, or learnable…
Sycophantic Agreement Transfers with Neutral Data via Contrastive Preference Optimization
Camila Blank, Zhuofan Ying, Christopher Potts +2
Sycophantic agreement refers to a behavior in which language models excessively affirm the user, often at the cost of factual accuracy. Although sycophantic agreement is a well-kno…
PRISM Edit: One Vector for All Temporal Answers
Chen Huang, Qi Zheng, Ruiqin Zheng +2
Model editing keeps large language models (LLMs) up to date without retraining, but temporal facts expose a limitation of the prevailing locate-and-edit paradigm: an update is not…
Do Language Models Consistently Encode the Current Year?
Suze van Adrichem, Aditi Bhaskar, Diyi Yang +2
A consistent concept of the current time is important for temporal reasoning, yet how language models represent the current time is not well understood. We contribute two tasks tha…
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…
PreFT: Prefill-only finetuning for efficient inference
Andrew Lanpouthakoun, Aryaman Arora, Zhengxuan Wu +4
Large language models can now be personalised efficiently at scale using parameter efficient finetuning methods (PEFTs), but serving user-specific PEFTs harms throughput, even with…
Outcome Rewards Do Not Guarantee Verifiable or Causally Important Reasoning
Qinan Yu, Alexa Tartaglini, Peter Hase +2
Reinforcement Learning from Verifiable Rewards (RLVR) on chain-of-thought reasoning has become a standard part of language model post-training recipes. A common assumption is that…
Blackbox Model Provenance via Palimpsestic Membership Inference
Rohith Kuditipudi, Jing Huang, Sally Zhu +3
Suppose Alice trains an open-weight language model and Bob uses a blackbox derivative of Alice's model to produce text. Can Alice prove that Bob is using her model, either by query…
HyperSteer: Activation Steering at Scale with Hypernetworks
Jiuding Sun, Sidharth Baskaran, Zhengxuan Wu +3
Steering language models (LMs) by modifying internal activations is a popular approach for controlling text generation. Unsupervised dictionary learning methods, e.g., sparse autoe…
Mechanistic evaluation of Transformers and state space models
Aryaman Arora, Neil Rathi, Nikil Roashan Selvam +3
State space models (SSMs) for language modelling promise an efficient and performant alternative to quadratic-attention Transformers, yet show variable performance on recalling bas…
Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors
Jing Huang, Junyi Tao, Thomas Icard +2
Interpretability research now offers a variety of techniques for identifying abstract internal mechanisms in neural networks. Can such techniques be used to predict how models will…
HyperDAS: Towards Automating Mechanistic Interpretability with Hypernetworks
Jiuding Sun, Jing Huang, Sidharth Baskaran +4
Mechanistic interpretability has made great strides in identifying neural network features (e.g., directions in hidden activation space) that mediate concepts(e.g., the birth year…
AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders
Zhengxuan Wu, Aryaman Arora, Atticus Geiger +5
Fine-grained steering of language model outputs is essential for safety and reliability. Prompting and finetuning are widely used to achieve these goals, but interpretability resea…
Bayesian scaling laws for in-context learning
Aryaman Arora, Dan Jurafsky, Christopher Potts +1
In-context learning (ICL) is a powerful technique for getting language models to perform complex tasks with no training updates. Prior work has established strong correlations betw…
Recurrent Neural Networks Learn to Store and Generate Sequences using Non-Linear Representations
Róbert Csordás, Christopher Potts, Christopher D. Manning +1
The Linear Representation Hypothesis (LRH) states that neural networks learn to encode concepts as directions in activation space, and a strong version of the LRH states that model…
Demystifying Verbatim Memorization in Large Language Models
Jing Huang, Diyi Yang, Christopher Potts
Large Language Models (LLMs) frequently memorize long sequences verbatim, often with serious legal and privacy implications. Much prior work has studied such verbatim memorization…
ReFT: Representation Finetuning for Language Models
Zhengxuan Wu, Aryaman Arora, Zheng Wang +4
Parameter-efficient finetuning (PEFT) methods seek to adapt large neural models via updates to a small number of weights. However, much prior interpretability work has shown that r…
MoEUT: Mixture-of-Experts Universal Transformers
Róbert Csordás, Kazuki Irie, Jürgen Schmidhuber +2
Previous work on Universal Transformers (UTs) has demonstrated the importance of parameter sharing across layers. By allowing recurrence in depth, UTs have advantages over standard…
Mapping the Increasing Use of LLMs in Scientific Papers
Weixin Liang, Yaohui Zhang, Zhengxuan Wu +11
Scientific publishing lays the foundation of science by disseminating research findings, fostering collaboration, encouraging reproducibility, and ensuring that scientific knowledg…
pyvene: A Library for Understanding and Improving PyTorch Models via Interventions
Zhengxuan Wu, Atticus Geiger, Aryaman Arora +5
Interventions on model-internal states are fundamental operations in many areas of AI, including model editing, steering, robustness, and interpretability. To facilitate such resea…
CausalGym: Benchmarking causal interpretability methods on linguistic tasks
Aryaman Arora, Dan Jurafsky, Christopher Potts
Language models (LMs) have proven to be powerful tools for psycholinguistic research, but most prior work has focused on purely behavioural measures (e.g., surprisal comparisons).…
RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representations
Jing Huang, Zhengxuan Wu, Christopher Potts +2
Individual neurons participate in the representation of multiple high-level concepts. To what extent can different interpretability methods successfully disentangle these roles? To…
A Reply to Makelov et al. (2023)'s "Interpretability Illusion" Arguments
Zhengxuan Wu, Atticus Geiger, Jing Huang +4
We respond to the recent paper by Makelov et al. (2023), which reviews subspace interchange intervention methods like distributed alignment search (DAS; Geiger et al. 2023) and cla…
Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretability
Atticus Geiger, Duligur Ibeling, Amir Zur +8
Causal abstraction provides a theoretical foundation for mechanistic interpretability, the field concerned with providing intelligible algorithms that are faithful simplifications…