Zhengxuan Wu

Zhengxuan Wu, Christopher D. Manning, Christopher Potts · stanford

Compositional generalization benchmarks for semantic parsing seek to assess whether models can accurately compute meanings for novel sentences, but operationalize this in terms of logical form (LF) prediction. This raises the concern that semantically irrelevant details of the chosen LFs could shape model performance. We argue that this concern is realized for the COGS benchmark. COGS poses generalization splits that appear impossible for present-day models, which could be taken as an indictment of those models. However, we show that the negative results trace to incidental features of COGS LFs. Converting these LFs to semantically equivalent ones and factoring out capabilities unrelated to semantic interpretation, we find that even baseline models get traction. A recent variable-free translation of COGS LFs suggests similar conclusions, but we observe this format is not semantically equivalent; it is incapable of accurately representing some COGS meanings. These findings inform our proposal for ReCOGS, a modified version of COGS that comes closer to assessing the target semantic capabilities while remaining very challenging. Overall, our results reaffirm the importance of compositional generalization and careful benchmark task design.

Tailin Wu, Megan Tjandrasuwita, Zhengxuan Wu et al. · mit

Humans have the remarkable ability to recognize and acquire novel visual concepts in a zero-shot manner. Given a high-level, symbolic description of a novel concept in terms of previously learned visual concepts and their relations, humans can recognize novel concepts without seeing any examples. Moreover, they can acquire new concepts by parsing and communicating symbolic structures using learned visual concepts and relations. Endowing these capabilities in machines is pivotal in improving their generalization capability at inference time. In this work, we introduce Zero-shot Concept Recognition and Acquisition (ZeroC), a neuro-symbolic architecture that can recognize and acquire novel concepts in a zero-shot way. ZeroC represents concepts as graphs of constituent concept models (as nodes) and their relations (as edges). To allow inference time composition, we employ energy-based models (EBMs) to model concepts and relations. We design ZeroC architecture so that it allows a one-to-one mapping between a symbolic graph structure of a concept and its corresponding EBM, which for the first time, allows acquiring new concepts, communicating its graph structure, and applying it to classification and detection tasks (even across domains) at inference time. We introduce algorithms for learning and inference with ZeroC. We evaluate ZeroC on a challenging grid-world dataset which is designed to probe zero-shot concept recognition and acquisition, and demonstrate its capability.

Jing Huang, Atticus Geiger, Karel D'Oosterlinck et al. · stanford

Natural language is an appealing medium for explaining how large language models process and store information, but evaluating the faithfulness of such explanations is challenging. To help address this, we develop two modes of evaluation for natural language explanations that claim individual neurons represent a concept in a text input. In the observational mode, we evaluate claims that a neuron $a$ activates on all and only input strings that refer to a concept picked out by the proposed explanation $E$. In the intervention mode, we construe $E$ as a claim that the neuron $a$ is a causal mediator of the concept denoted by $E$. We apply our framework to the GPT-4-generated explanations of GPT-2 XL neurons of Bills et al. (2023) and show that even the most confident explanations have high error rates and little to no causal efficacy. We close the paper by critically assessing whether natural language is a good choice for explanations and whether neurons are the best level of analysis.

Zhengxuan Wu, Karel D'Oosterlinck, Atticus Geiger et al.

Explainability methods for NLP systems encounter a version of the fundamental problem of causal inference: for a given ground-truth input text, we never truly observe the counterfactual texts necessary for isolating the causal effects of model representations on outputs. In response, many explainability methods make no use of counterfactual texts, assuming they will be unavailable. In this paper, we show that robust causal explainability methods can be created using approximate counterfactuals, which can be written by humans to approximate a specific counterfactual or simply sampled using metadata-guided heuristics. The core of our proposal is the Causal Proxy Model (CPM). A CPM explains a black-box model $\mathcal{N}$ because it is trained to have the same actual input/output behavior as $\mathcal{N}$ while creating neural representations that can be intervened upon to simulate the counterfactual input/output behavior of $\mathcal{N}$. Furthermore, we show that the best CPM for $\mathcal{N}$ performs comparably to $\mathcal{N}$ in making factual predictions, which means that the CPM can simply replace $\mathcal{N}$, leading to more explainable deployed models. Our code is available at https://github.com/frankaging/Causal-Proxy-Model.

Jing Huang, Zhengxuan Wu, Kyle Mahowald et al. · stanford

Language tasks involving character-level manipulations (e.g., spelling corrections, arithmetic operations, word games) are challenging for models operating on subword units. To address this, we develop a causal intervention framework to learn robust and interpretable character representations inside subword-based language models. Our method treats each character as a typed variable in a causal model and learns such causal structures by adapting the interchange intervention training method of Geiger et al. (2021). We additionally introduce a suite of character-level tasks that systematically vary in their dependence on meaning and sequence-level context. While character-level models still perform best on purely form-based tasks like string reversal, our method outperforms character-level models on more complex tasks that blend form, meaning, and context, such as spelling correction in context and word search games. Compared with standard subword-based models, our approach also significantly improves robustness on unseen token sequences and leads to human-interpretable internal representations of characters.

Atticus Geiger, Duligur Ibeling, Amir Zur et al. · stanford

Causal abstraction provides a theoretical foundation for mechanistic interpretability, the field concerned with providing intelligible algorithms that are faithful simplifications of the known, but opaque low-level details of black box AI models. Our contributions are (1) generalizing the theory of causal abstraction from mechanism replacement (i.e., hard and soft interventions) to arbitrary mechanism transformation (i.e., functionals from old mechanisms to new mechanisms), (2) providing a flexible, yet precise formalization for the core concepts of polysemantic neurons, the linear representation hypothesis, modular features, and graded faithfulness, and (3) unifying a variety of mechanistic interpretability methods in the common language of causal abstraction, namely, activation and path patching, causal mediation analysis, causal scrubbing, causal tracing, circuit analysis, concept erasure, sparse autoencoders, differential binary masking, distributed alignment search, and steering.

Atticus Geiger, Zhengxuan Wu, Christopher Potts et al.

Causal abstraction is a promising theoretical framework for explainable artificial intelligence that defines when an interpretable high-level causal model is a faithful simplification of a low-level deep learning system. However, existing causal abstraction methods have two major limitations: they require a brute-force search over alignments between the high-level model and the low-level one, and they presuppose that variables in the high-level model will align with disjoint sets of neurons in the low-level one. In this paper, we present distributed alignment search (DAS), which overcomes these limitations. In DAS, we find the alignment between high-level and low-level models using gradient descent rather than conducting a brute-force search, and we allow individual neurons to play multiple distinct roles by analyzing representations in non-standard bases-distributed representations. Our experiments show that DAS can discover internal structure that prior approaches miss. Overall, DAS removes previous obstacles to conducting causal abstraction analyses and allows us to find conceptual structure in trained neural nets.

Eldar David Abraham, Karel D'Oosterlinck, Amir Feder et al.

The increasing size and complexity of modern ML systems has improved their predictive capabilities but made their behavior harder to explain. Many techniques for model explanation have been developed in response, but we lack clear criteria for assessing these techniques. In this paper, we cast model explanation as the causal inference problem of estimating causal effects of real-world concepts on the output behavior of ML models given actual input data. We introduce CEBaB, a new benchmark dataset for assessing concept-based explanation methods in Natural Language Processing (NLP). CEBaB consists of short restaurant reviews with human-generated counterfactual reviews in which an aspect (food, noise, ambiance, service) of the dining experience was modified. Original and counterfactual reviews are annotated with multiply-validated sentiment ratings at the aspect-level and review-level. The rich structure of CEBaB allows us to go beyond input features to study the effects of abstract, real-world concepts on model behavior. We use CEBaB to compare the quality of a range of concept-based explanation methods covering different assumptions and conceptions of the problem, and we seek to establish natural metrics for comparative assessments of these methods.

Aryaman Arora, Zhengxuan Wu, Jacob Steinhardt et al. · stanford

The high-level concepts that a neural network uses to perform computation need not be aligned to individual neurons (Smolensky, 1986). Language model interpretability research has thus turned to techniques such as \textit{sparse autoencoders} (SAEs) to decompose the neuron basis into more interpretable units of model computation, for tasks such as \textit{circuit tracing}. However, not all neuron-based representations are uninterpretable. For the first time, we empirically show that \textbf{MLP neurons are as sparse a feature basis as SAEs}. We use this finding to develop an end-to-end pipeline for circuit tracing on the MLP neuron basis, which locates causal circuitry on a variety of tasks using gradient-based attribution. On a standard subject-verb agreement benchmark (Marks et al., 2025), a circuit of $\approx 10^2$ MLP neurons is enough to control model behaviour. On the multi-hop city $\to$ state $\to$ capital task from Lindsey et al., 2025, we find a circuit in which small sets of neurons encode specific latent reasoning steps (e.g.~`map city to its state'), and can be steered to change the model's output. This work thus advances automated interpretability of language models without additional training costs.

Zhengxuan Wu, Yuhao Zhang, Peng Qi et al. · stanford

Modern language models (LMs) need to follow human instructions while being faithful; yet, they often fail to achieve both. Here, we provide concrete evidence of a trade-off between instruction following (i.e., follow open-ended instructions) and faithfulness (i.e., ground responses in given context) when training LMs with these objectives. For instance, fine-tuning LLaMA-7B on instruction following datasets renders it less faithful. Conversely, instruction-tuned Vicuna-7B shows degraded performance at following instructions when further optimized on tasks that require contextual grounding. One common remedy is multi-task learning (MTL) with data mixing, yet it remains far from achieving a synergic outcome. We propose a simple yet effective method that relies on Rejection Sampling for Continued Self-instruction Tuning (ReSet), which significantly outperforms vanilla MTL. Surprisingly, we find that less is more, as training ReSet with high-quality, yet substantially smaller data (three-fold less) yields superior results. Our findings offer a better understanding of objective discrepancies in alignment training of LMs.

Zhengxuan Wu, Aryaman Arora, Zheng Wang et al. · stanford

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 representations encode rich semantic information, suggesting that editing representations might be a more powerful alternative. We pursue this hypothesis by developing a family of Representation Finetuning (ReFT) methods. ReFT methods operate on a frozen base model and learn task-specific interventions on hidden representations. We define a strong instance of the ReFT family, Low-rank Linear Subspace ReFT (LoReFT), and we identify an ablation of this method that trades some performance for increased efficiency. Both are drop-in replacements for existing PEFTs and learn interventions that are 15x--65x more parameter-efficient than LoRA. We showcase LoReFT on eight commonsense reasoning tasks, four arithmetic reasoning tasks, instruction-tuning, and GLUE. In all these evaluations, our ReFTs deliver the best balance of efficiency and performance, and almost always outperform state-of-the-art PEFTs. We release a generic ReFT training library publicly at https://github.com/stanfordnlp/pyreft.

Jing Huang, Zhengxuan Wu, Christopher Potts et al. · deepmind, stanford

Individual neurons participate in the representation of multiple high-level concepts. To what extent can different interpretability methods successfully disentangle these roles? To help address this question, we introduce RAVEL (Resolving Attribute-Value Entanglements in Language Models), a dataset that enables tightly controlled, quantitative comparisons between a variety of existing interpretability methods. We use the resulting conceptual framework to define the new method of Multi-task Distributed Alignment Search (MDAS), which allows us to find distributed representations satisfying multiple causal criteria. With Llama2-7B as the target language model, MDAS achieves state-of-the-art results on RAVEL, demonstrating the importance of going beyond neuron-level analyses to identify features distributed across activations. We release our benchmark at https://github.com/explanare/ravel.

Zhengxuan Wu, Atticus Geiger, Aryaman Arora et al. · stanford

Interventions on model-internal states are fundamental operations in many areas of AI, including model editing, steering, robustness, and interpretability. To facilitate such research, we introduce $\textbf{pyvene}$, an open-source Python library that supports customizable interventions on a range of different PyTorch modules. $\textbf{pyvene}$ supports complex intervention schemes with an intuitive configuration format, and its interventions can be static or include trainable parameters. We show how $\textbf{pyvene}$ provides a unified and extensible framework for performing interventions on neural models and sharing the intervened upon models with others. We illustrate the power of the library via interpretability analyses using causal abstraction and knowledge localization. We publish our library through Python Package Index (PyPI) and provide code, documentation, and tutorials at https://github.com/stanfordnlp/pyvene.

Andrew Lanpouthakoun, Aryaman Arora, Zhengxuan Wu et al.

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 specialised kernels and memory management techniques. This is because, theoretically and empirically, a mismatch exists between prefill (processing a large number of tokens at once) and decode (generating a single token autoregressively): the latter has far lower throughput when serving multiple adapters. Rather than optimising performance relative to parameter count, for efficient multi-adapter serving, we instead ought to optimise performance relative to serving throughput. We therefore propose PreFT (Prefill-only Finetuning), wherein we only apply the adapter to prefill tokens and discard it afterwards. PreFT significantly increases throughput with minimal effect on performance. We develop and release an efficient implementation of two prefill-only PEFTs, LoRA and ReFT, on the vLLM inference engine. We first show that serving multi-user PreFTs is more efficient than traditional PEFTs ($1.9\times$ the throughput when serving $512$ adapters on Llama 3.1 70B). Then, we compare the performance of prefill-only vs. all-token adapters on a variety of supervised finetuning and reinforcement learning tasks with LMs at varying scales. On SFT, we observe that the evaluation loss of PreFTs is higher than PEFTs, but can be compensated by increasing rank with nearly no reduction in throughput. On RL, we consistently find that PreFTs approach parity with standard PEFTs. Together, this work validates prefill-only adaptation of LLMs as a more favourable accuracy-throughput tradeoff than existing PEFTs for personalised serving.

Weixin Liang, Yaohui Zhang, Zhengxuan Wu et al. · berkeley, stanford

Scientific publishing lays the foundation of science by disseminating research findings, fostering collaboration, encouraging reproducibility, and ensuring that scientific knowledge is accessible, verifiable, and built upon over time. Recently, there has been immense speculation about how many people are using large language models (LLMs) like ChatGPT in their academic writing, and to what extent this tool might have an effect on global scientific practices. However, we lack a precise measure of the proportion of academic writing substantially modified or produced by LLMs. To address this gap, we conduct the first systematic, large-scale analysis across 950,965 papers published between January 2020 and February 2024 on the arXiv, bioRxiv, and Nature portfolio journals, using a population-level statistical framework to measure the prevalence of LLM-modified content over time. Our statistical estimation operates on the corpus level and is more robust than inference on individual instances. Our findings reveal a steady increase in LLM usage, with the largest and fastest growth observed in Computer Science papers (up to 17.5%). In comparison, Mathematics papers and the Nature portfolio showed the least LLM modification (up to 6.3%). Moreover, at an aggregate level, our analysis reveals that higher levels of LLM-modification are associated with papers whose first authors post preprints more frequently, papers in more crowded research areas, and papers of shorter lengths. Our findings suggests that LLMs are being broadly used in scientific writings.

Joakim Edin, Róbert Csordás, Tuukka Ruotsalo et al.

Ensuring faithful interpretability in large language models is imperative for trustworthy and reliable AI. A key obstacle is self-repair, a phenomenon where networks compensate for reduced signal in one component by amplifying others, masking the true importance of the ablated component. While prior work attributes self-repair to layer normalization and back-up components that compensate for ablated components, we identify a novel form occurring within the attention mechanism, where softmax redistribution conceals the influence of important attention scores. This leads traditional ablation and gradient-based methods to underestimate the significance of all components contributing to these attention scores. We introduce Gradient Interaction Modifications (GIM), a technique that accounts for self-repair during backpropagation. Extensive experiments across multiple large language models (Gemma 2B/9B, LLAMA 1B/3B/8B, Qwen 1.5B/3B) and diverse tasks demonstrate that GIM significantly improves faithfulness over existing circuit identification and feature attribution methods. Our work is a significant step toward better understanding the inner mechanisms of LLMs, which is crucial for improving them and ensuring their safety. Our code is available at https://github.com/JoakimEdin/gim.

Zhengxuan Wu, Atticus Geiger, Thomas Icard et al.

Obtaining human-interpretable explanations of large, general-purpose language models is an urgent goal for AI safety. However, it is just as important that our interpretability methods are faithful to the causal dynamics underlying model behavior and able to robustly generalize to unseen inputs. Distributed Alignment Search (DAS) is a powerful gradient descent method grounded in a theory of causal abstraction that has uncovered perfect alignments between interpretable symbolic algorithms and small deep learning models fine-tuned for specific tasks. In the present paper, we scale DAS significantly by replacing the remaining brute-force search steps with learned parameters -- an approach we call Boundless DAS. This enables us to efficiently search for interpretable causal structure in large language models while they follow instructions. We apply Boundless DAS to the Alpaca model (7B parameters), which, off the shelf, solves a simple numerical reasoning problem. With Boundless DAS, we discover that Alpaca does this by implementing a causal model with two interpretable boolean variables. Furthermore, we find that the alignment of neural representations with these variables is robust to changes in inputs and instructions. These findings mark a first step toward faithfully understanding the inner-workings of our ever-growing and most widely deployed language models. Our tool is extensible to larger LLMs and is released publicly at `https://github.com/stanfordnlp/pyvene`.

Douwe Kiela, Max Bartolo, Yixin Nie et al.

We introduce Dynabench, an open-source platform for dynamic dataset creation and model benchmarking. Dynabench runs in a web browser and supports human-and-model-in-the-loop dataset creation: annotators seek to create examples that a target model will misclassify, but that another person will not. In this paper, we argue that Dynabench addresses a critical need in our community: contemporary models quickly achieve outstanding performance on benchmark tasks but nonetheless fail on simple challenge examples and falter in real-world scenarios. With Dynabench, dataset creation, model development, and model assessment can directly inform each other, leading to more robust and informative benchmarks. We report on four initial NLP tasks, illustrating these concepts and highlighting the promise of the platform, and address potential objections to dynamic benchmarking as a new standard for the field.

Christopher Potts, Zhengxuan Wu, Atticus Geiger et al.

We introduce DynaSent ('Dynamic Sentiment'), a new English-language benchmark task for ternary (positive/negative/neutral) sentiment analysis. DynaSent combines naturally occurring sentences with sentences created using the open-source Dynabench Platform, which facilities human-and-model-in-the-loop dataset creation. DynaSent has a total of 121,634 sentences, each validated by five crowdworkers, and its development and test splits are designed to produce chance performance for even the best models we have been able to develop; when future models solve this task, we will use them to create DynaSent version 2, continuing the dynamic evolution of this benchmark. Here, we report on the dataset creation effort, focusing on the steps we took to increase quality and reduce artifacts. We also present evidence that DynaSent's Neutral category is more coherent than the comparable category in other benchmarks, and we motivate training models from scratch for each round over successive fine-tuning.

Zhengxuan Wu, Aryaman Arora, Atticus Geiger et al. · stanford

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 researchers have proposed a variety of representation-based techniques as well, including sparse autoencoders (SAEs), linear artificial tomography, supervised steering vectors, linear probes, and representation finetuning. At present, there is no benchmark for making direct comparisons between these proposals. Therefore, we introduce AxBench, a large-scale benchmark for steering and concept detection, and report experiments on Gemma-2-2B and 9B. For steering, we find that prompting outperforms all existing methods, followed by finetuning. For concept detection, representation-based methods such as difference-in-means, perform the best. On both evaluations, SAEs are not competitive. We introduce a novel weakly-supervised representational method (Rank-1 Representation Finetuning; ReFT-r1), which is competitive on both tasks while providing the interpretability advantages that prompting lacks. Along with AxBench, we train and publicly release SAE-scale feature dictionaries for ReFT-r1 and DiffMean.

Shiqi Chen, Miao Xiong, Junteng Liu et al.

Large language models (LLMs) frequently hallucinate and produce factual errors, yet our understanding of why they make these errors remains limited. In this study, we delve into the underlying mechanisms of LLM hallucinations from the perspective of inner representations, and discover a salient pattern associated with hallucinations: correct generations tend to have sharper context activations in the hidden states of the in-context tokens, compared to the incorrect ones. Leveraging this insight, we propose an entropy-based metric to quantify the ``sharpness'' among the in-context hidden states and incorporate it into the decoding process to formulate a constrained decoding approach. Experiments on various knowledge-seeking and hallucination benchmarks demonstrate our approach's consistent effectiveness, for example, achieving up to an 8.6 point improvement on TruthfulQA. We believe this study can improve our understanding of hallucinations and serve as a practical solution for hallucination mitigation.

Zhengxuan Wu, Atticus Geiger, Jing Huang et al. · stanford

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 claims that these methods potentially cause "interpretability illusions". We first review Makelov et al. (2023)'s technical notion of what an "interpretability illusion" is, and then we show that even intuitive and desirable explanations can qualify as illusions in this sense. As a result, their method of discovering "illusions" can reject explanations they consider "non-illusory". We then argue that the illusions Makelov et al. (2023) see in practice are artifacts of their training and evaluation paradigms. We close by emphasizing that, though we disagree with their core characterization, Makelov et al. (2023)'s examples and discussion have undoubtedly pushed the field of interpretability forward.

Jiachen Zhao, Jing Huang, Zhengxuan Wu et al. · stanford

LLMs are trained to refuse harmful instructions, but do they truly understand harmfulness beyond just refusing? Prior work has shown that LLMs' refusal behaviors can be mediated by a one-dimensional subspace, i.e., a refusal direction. In this work, we identify a new dimension to analyze safety mechanisms in LLMs, i.e., harmfulness, which is encoded internally as a separate concept from refusal. There exists a harmfulness direction that is distinct from the refusal direction. As causal evidence, steering along the harmfulness direction can lead LLMs to interpret harmless instructions as harmful, but steering along the refusal direction tends to elicit refusal responses directly without reversing the model's judgment on harmfulness. Furthermore, using our identified harmfulness concept, we find that certain jailbreak methods work by reducing the refusal signals without reversing the model's internal belief of harmfulness. We also find that adversarially finetuning models to accept harmful instructions has minimal impact on the model's internal belief of harmfulness. These insights lead to a practical safety application: The model's latent harmfulness representation can serve as an intrinsic safeguard (Latent Guard) for detecting unsafe inputs and reducing over-refusals that is robust to finetuning attacks. For instance, our Latent Guard achieves performance comparable to or better than Llama Guard 3 8B, a dedicated finetuned safeguard model, across different jailbreak methods. Our findings suggest that LLMs' internal understanding of harmfulness is more robust than their refusal decision to diverse input instructions, offering a new perspective to study AI safety.

Aryaman Arora, Zhengxuan Wu, Jacob Steinhardt et al.

In language model interpretability research, \textbf{circuit tracing} aims to identify which internal features causally contributed to a particular output and how they affected each other, with the goal of explaining the computations underlying some behaviour. However, all prior circuit tracing work has relied on ad-hoc human interpretation of the role that each feature in the circuit plays, via manual inspection of data artifacts such as the dataset examples the component activates on. We introduce \textbf{ADAG}, an end-to-end pipeline for describing these attribution graphs which is fully automated. To achieve this, we introduce \textit{attribution profiles} which quantify the functional role of a feature via its input and output gradient effects. We then introduce a novel clustering algorithm for grouping features, and an LLM explainer--simulator setup which generates and scores natural-language explanations of the functional role of these feature groups. We run our system on known human-analysed circuit-tracing tasks and recover interpretable circuits, and further show ADAG can find steerable clusters which are responsible for a harmful advice jailbreak in Llama 3.1 8B Instruct.

Jiuding Sun, Sidharth Baskaran, Zhengxuan Wu et al.

Steering language models (LMs) by modifying internal activations is a popular approach for controlling text generation. Unsupervised dictionary learning methods, e.g., sparse autoencoders, can be scaled to produce many steering vectors, but lack guarantees on the individual efficacy of each vector and control over the coverage of relevant steering tasks. In contrast, supervised methods for constructing steering vectors are targeted and effective, but require more data collection and training for each additional steering vector produced. In this work, we introduce HyperSteer, a family of hypernetwork-based architectures which are trained end-to-end to generate steering vectors conditioned on the natural language steering prompts and the internals of the steered LM. In our evaluations, we show that scaling HyperSteer with thousands of steering prompts exceeds the performance of state-of-the-art activation steering methods, even on steering prompts never seen during training. Moreover, HyperSteer performs on par with steering-via-prompting.

Zhengxuan Wu, Qinan Yu, Aryaman Arora et al. · stanford

Steering methods for language models (LMs) seek to provide fine-grained and interpretable control over model generations by variously changing model inputs, weights, or representations to adjust behavior. Recent work has shown that adjusting weights or representations is often less effective than steering by prompting, for instance when wanting to introduce or suppress a particular concept. We demonstrate how to improve representation steering via our new Reference-free Preference Steering (RePS), a bidirectional preference-optimization objective that jointly does concept steering and suppression. We train three parameterizations of RePS and evaluate them on AxBench, a large-scale model steering benchmark. On Gemma models with sizes ranging from 2B to 27B, RePS outperforms all existing steering methods trained with a language modeling objective and substantially narrows the gap with prompting -- while promoting interpretability and minimizing parameter count. In suppression, RePS matches the language-modeling objective on Gemma-2 and outperforms it on the larger Gemma-3 variants while remaining resilient to prompt-based jailbreaking attacks that defeat prompting. Overall, our results suggest that RePS provides an interpretable and robust alternative to prompting for both steering and suppression.

Zexuan Zhong, Zhengxuan Wu, Christopher D. Manning et al.

The information stored in large language models (LLMs) falls out of date quickly, and retraining from scratch is often not an option. This has recently given rise to a range of techniques for injecting new facts through updating model weights. Current evaluation paradigms are extremely limited, mainly validating the recall of edited facts, but changing one fact should cause rippling changes to the model's related beliefs. If we edit the UK Prime Minister to now be Rishi Sunak, then we should get a different answer to Who is married to the British Prime Minister? In this work, we present a benchmark, MQuAKE (Multi-hop Question Answering for Knowledge Editing), comprising multi-hop questions that assess whether edited models correctly answer questions where the answer should change as an entailed consequence of edited facts. While we find that current knowledge-editing approaches can recall edited facts accurately, they fail catastrophically on the constructed multi-hop questions. We thus propose a simple memory-based approach, MeLLo, which stores all edited facts externally while prompting the language model iteratively to generate answers that are consistent with the edited facts. While MQuAKE remains challenging, we show that MeLLo scales well with LLMs (e.g., OpenAI GPT-3.5-turbo) and outperforms previous model editors by a large margin.

Zhengxuan Wu, Alex Tamkin, Isabel Papadimitriou

When we transfer a pretrained language model to a new language, there are many axes of variation that change at once. To disentangle the impact of different factors like syntactic similarity and vocabulary similarity, we propose a set of controlled transfer studies: we systematically transform the language of the GLUE benchmark, altering one axis of crosslingual variation at a time, and then measure the resulting drops in a pretrained model's downstream performance. We find that models can largely recover from syntactic-style shifts, but cannot recover from vocabulary misalignment and embedding matrix re-initialization, even with continued pretraining on 15 million tokens. %On the other hand, transferring to a dataset with an unaligned vocabulary is extremely hard to recover from in the low-data regime. Moreover, good-quality tokenizers in the transfer language do not make vocabulary alignment easier. Our experiments provide insights into the factors of cross-lingual transfer that researchers should most focus on when designing language transfer scenarios.

Zhengxuan Wu, Atticus Geiger, Josh Rozner et al.

Distillation efforts have led to language models that are more compact and efficient without serious drops in performance. The standard approach to distillation trains a student model against two objectives: a task-specific objective (e.g., language modeling) and an imitation objective that encourages the hidden states of the student model to be similar to those of the larger teacher model. In this paper, we show that it is beneficial to augment distillation with a third objective that encourages the student to imitate the causal computation process of the teacher through interchange intervention training(IIT). IIT pushes the student model to become a causal abstraction of the teacher model - a simpler model with the same causal structure. IIT is fully differentiable, easily implemented, and combines flexibly with other objectives. Compared with standard distillation of BERT, distillation via IIT results in lower perplexity on Wikipedia (masked language modeling) and marked improvements on the GLUE benchmark (natural language understanding), SQuAD (question answering), and CoNLL-2003 (named entity recognition).

Atticus Geiger, Zhengxuan Wu, Hanson Lu et al.

In many areas, we have well-founded insights about causal structure that would be useful to bring into our trained models while still allowing them to learn in a data-driven fashion. To achieve this, we present the new method of interchange intervention training (IIT). In IIT, we (1) align variables in a causal model (e.g., a deterministic program or Bayesian network) with representations in a neural model and (2) train the neural model to match the counterfactual behavior of the causal model on a base input when aligned representations in both models are set to be the value they would be for a source input. IIT is fully differentiable, flexibly combines with other objectives, and guarantees that the target causal model is a causal abstraction of the neural model when its loss is zero. We evaluate IIT on a structural vision task (MNIST-PVR), a navigational language task (ReaSCAN), and a natural language inference task (MQNLI). We compare IIT against multi-task training objectives and data augmentation. In all our experiments, IIT achieves the best results and produces neural models that are more interpretable in the sense that they more successfully realize the target causal model.

Zhengxuan Wu, Elisa Kreiss, Desmond C. Ong et al.

The ability to compositionally map language to referents, relations, and actions is an essential component of language understanding. The recent gSCAN dataset (Ruis et al. 2020, NeurIPS) is an inspiring attempt to assess the capacity of models to learn this kind of grounding in scenarios involving navigational instructions. However, we show that gSCAN's highly constrained design means that it does not require compositional interpretation and that many details of its instructions and scenarios are not required for task success. To address these limitations, we propose ReaSCAN, a benchmark dataset that builds off gSCAN but requires compositional language interpretation and reasoning about entities and relations. We assess two models on ReaSCAN: a multi-modal baseline and a state-of-the-art graph convolutional neural model. These experiments show that ReaSCAN is substantially harder than gSCAN for both neural architectures. This suggests that ReaSCAN can serve as a valuable benchmark for advancing our understanding of models' compositional generalization and reasoning capabilities.

Zhengxuan Wu, Nelson F. Liu, Christopher Potts

There is growing evidence that pretrained language models improve task-specific fine-tuning not just for the languages seen in pretraining, but also for new languages and even non-linguistic data. What is the nature of this surprising cross-domain transfer? We offer a partial answer via a systematic exploration of how much transfer occurs when models are denied any information about word identity via random scrambling. In four classification tasks and two sequence labeling tasks, we evaluate baseline models, LSTMs using GloVe embeddings, and BERT. We find that only BERT shows high rates of transfer into our scrambled domains, and for classification but not sequence labeling tasks. Our analyses seek to explain why transfer succeeds for some tasks but not others, to isolate the separate contributions of pretraining versus fine-tuning, and to quantify the role of word frequency. These findings help explain where and why cross-domain transfer occurs, which can guide future studies and practical fine-tuning efforts.

Geza Kovacs, Zhengxuan Wu, Michael S. Bernstein

How effectively do we adhere to nudges and interventions that help us control our online browsing habits? If we have a temporary lapse and disable the behavior change system, do we later resume our adherence, or has the dam broken? In this paper, we investigate these questions through log analyses of 8,000+ users on HabitLab, a behavior change platform that helps users reduce their time online. We find that, while users typically begin with high-challenge interventions, over time they allow themselves to slip into easier and easier interventions. Despite this, many still expect to return to the harder interventions imminently: they repeatedly choose to be asked to change difficulty again on the next visit, declining to have the system save their preference for easy interventions.

Zhengxuan Wu, Desmond C. Ong

BERT, as one of the pretrianed language models, attracts the most attention in recent years for creating new benchmarks across GLUE tasks via fine-tuning. One pressing issue is to open up the blackbox and explain the decision makings of BERT. A number of attribution techniques have been proposed to explain BERT models, but are often limited to sequence to sequence tasks. In this paper, we adapt existing attribution methods on explaining decision makings of BERT in sequence classification tasks. We conduct extensive analyses of four existing attribution methods by applying them to four different datasets in sentiment analysis. We compare the reliability and robustness of each method via various ablation studies. Furthermore, we test whether attribution methods explain generalized semantics across semantically similar tasks. Our work provides solid guidance for using attribution methods to explain decision makings of BERT for downstream classification tasks.

Zhengxuan Wu, Desmond C. Ong

Aspect-based sentiment analysis (ABSA) and Targeted ASBA (TABSA) allow finer-grained inferences about sentiment to be drawn from the same text, depending on context. For example, a given text can have different targets (e.g., neighborhoods) and different aspects (e.g., price or safety), with different sentiment associated with each target-aspect pair. In this paper, we investigate whether adding context to self-attention models improves performance on (T)ABSA. We propose two variants of Context-Guided BERT (CG-BERT) that learn to distribute attention under different contexts. We first adapt a context-aware Transformer to produce a CG-BERT that uses context-guided softmax-attention. Next, we propose an improved Quasi-Attention CG-BERT model that learns a compositional attention that supports subtractive attention. We train both models with pretrained BERT on two (T)ABSA datasets: SentiHood and SemEval-2014 (Task 4). Both models achieve new state-of-the-art results with our QACG-BERT model having the best performance. Furthermore, we provide analyses of the impact of context in the our proposed models. Our work provides more evidence for the utility of adding context-dependencies to pretrained self-attention-based language models for context-based natural language tasks.

Zhengxuan Wu, Thanh-Son Nguyen, Desmond C. Ong

Neural attention, especially the self-attention made popular by the Transformer, has become the workhorse of state-of-the-art natural language processing (NLP) models. Very recent work suggests that the self-attention in the Transformer encodes syntactic information; Here, we show that self-attention scores encode semantics by considering sentiment analysis tasks. In contrast to gradient-based feature attribution methods, we propose a simple and effective Layer-wise Attention Tracing (LAT) method to analyze structured attention weights. We apply our method to Transformer models trained on two tasks that have surface dissimilarities, but share common semantics---sentiment analysis of movie reviews and time-series valence prediction in life story narratives. Across both tasks, words with high aggregated attention weights were rich in emotional semantics, as quantitatively validated by an emotion lexicon labeled by human annotators. Our results show that structured attention weights encode rich semantics in sentiment analysis, and match human interpretations of semantics.

Zhengxuan Wu, Desmond C. Ong

Grounding language in contextual information is crucial for fine-grained natural language understanding. One important task that involves grounding contextual modifiers is color generation. Given a reference color "green", and a modifier "bluey", how does one generate a color that could represent "bluey green"? We propose a computational pragmatics model that formulates this color generation task as a recursive game between speakers and listeners. In our model, a pragmatic speaker reasons about the inferences that a listener would make, and thus generates a modified color that is maximally informative to help the listener recover the original referents. In this paper, we show that incorporating pragmatic information provides significant improvements in performance compared with other state-of-the-art deep learning models where pragmatic inference and flexibility in representing colors from a large continuous space are lacking. Our model has an absolute 98% increase in performance for the test cases where the reference colors are unseen during training, and an absolute 40% increase in performance for the test cases where both the reference colors and the modifiers are unseen during training.

Desmond C. Ong, Zhengxuan Wu, Tan Zhi-Xuan et al.

Human emotions unfold over time, and more affective computing research has to prioritize capturing this crucial component of real-world affect. Modeling dynamic emotional stimuli requires solving the twin challenges of time-series modeling and of collecting high-quality time-series datasets. We begin by assessing the state-of-the-art in time-series emotion recognition, and we review contemporary time-series approaches in affective computing, including discriminative and generative models. We then introduce the first version of the Stanford Emotional Narratives Dataset (SENDv1): a set of rich, multimodal videos of self-paced, unscripted emotional narratives, annotated for emotional valence over time. The complex narratives and naturalistic expressions in this dataset provide a challenging test for contemporary time-series emotion recognition models. We demonstrate several baseline and state-of-the-art modeling approaches on the SEND, including a Long Short-Term Memory model and a multimodal Variational Recurrent Neural Network, which perform comparably to the human-benchmark. We end by discussing the implications for future research in time-series affective computing.

Zhengxuan Wu, Yueyi Jiang

Word embedding models such as GloVe are widely used in natural language processing (NLP) research to convert words into vectors. Here, we provide a preliminary guide to probe latent emotions in text through GloVe word vectors. First, we trained a neural network model to predict continuous emotion valence ratings by taking linguistic inputs from Stanford Emotional Narratives Dataset (SEND). After interpreting the weights in the model, we found that only a few dimensions of the word vectors contributed to expressing emotions in text, and words were clustered on the basis of their emotional polarities. Furthermore, we performed a linear transformation that projected high dimensional embedded vectors into an emotion space. Based on NRC Emotion Lexicon (EmoLex), we visualized the entanglement of emotions in the lexicon by using both projected and raw GloVe word vectors. We showed that, in the proposed emotion space, we were able to better disentangle emotions than using raw GloVe vectors alone. In addition, we found that the sum vectors of different pairs of emotion words successfully captured expressed human feelings in the EmoLex. For example, the sum of two embedded word vectors expressing Joy and Trust which express Love shared high similarity (similarity score .62) with the embedded vector expressing Optimism. On the contrary, this sum vector was dissimilar (similarity score -.19) with the the embedded vector expressing Remorse. In this paper, we argue that through the proposed emotion space, arithmetic of emotions is preserved in the word vectors. The affective representation uncovered in emotion vector space could shed some light on how to help machines to disentangle emotion expressed in word embeddings.

Zhengxuan Wu, Xiyu Zhang, Tan Zhi-Xuan et al.

Attention mechanisms in deep neural networks have achieved excellent performance on sequence-prediction tasks. Here, we show that these recently-proposed attention-based mechanisms---in particular, the Transformer with its parallelizable self-attention layers, and the Memory Fusion Network with attention across modalities and time---also generalize well to multimodal time-series emotion recognition. Using a recently-introduced dataset of emotional autobiographical narratives, we adapt and apply these two attention mechanisms to predict emotional valence over time. Our models perform extremely well, in some cases reaching a performance comparable with human raters. We end with a discussion of the implications of attention mechanisms to affective computing.