We study the feasibility of conducting alignment audits: investigations into whether models have undesired objectives. As a testbed, we train a language model with a hidden objective. Our training pipeline first teaches the model about exploitable errors in RLHF reward models (RMs), then trains the model to exploit some of these errors. We verify via out-of-distribution evaluations that the model generalizes to exhibit whatever behaviors it believes RMs rate highly, including ones not reinforced during training. We leverage this model to study alignment audits in two ways. First, we conduct a blind auditing game where four teams, unaware of the model's hidden objective or training, investigate it for concerning behaviors and their causes. Three teams successfully uncovered the model's hidden objective using techniques including interpretability with sparse autoencoders (SAEs), behavioral attacks, and training data analysis. Second, we conduct an unblinded follow-up study of eight techniques for auditing the model, analyzing their strengths and limitations. Overall, our work provides a concrete example of using alignment audits to discover a model's hidden objective and proposes a methodology for practicing and validating progress in alignment auditing.

An approach to improve neural network interpretability is via clusterability, i.e., splitting a model into disjoint clusters that can be studied independently. We define a measure for clusterability and show that pre-trained models form highly enmeshed clusters via spectral graph clustering. We thus train models to be more modular using a "clusterability loss" function that encourages the formation of non-interacting clusters. Using automated interpretability techniques, we show that our method can help train models that are more modular and learn different, disjoint, and smaller circuits. We investigate CNNs trained on MNIST and CIFAR, small transformers trained on modular addition, and language models. Our approach provides a promising direction for training neural networks that learn simpler functions and are easier to interpret.

An approach to improve network interpretability is via clusterability, i.e., splitting a model into disjoint clusters that can be studied independently. We find pretrained models to be highly unclusterable and thus train models to be more modular using an ``enmeshment loss'' function that encourages the formation of non-interacting clusters. Using automated interpretability measures, we show that our method finds clusters that learn different, disjoint, and smaller circuits for CIFAR-10 labels. Our approach provides a promising direction for making neural networks easier to interpret.

Grokking, a phenomenon where machine learning models generalize long after overfitting, has been primarily observed and studied in algorithmic tasks. This paper explores grokking in real-world datasets using deep neural networks for classification under the cross-entropy loss. We challenge the prevalent hypothesis that the $L_2$ norm of weights is the primary cause of grokking by demonstrating that grokking can occur outside the expected range of weight norms. To better understand grokking, we introduce three new progress measures: activation sparsity, absolute weight entropy, and approximate local circuit complexity. These measures are conceptually related to generalization and demonstrate a stronger correlation with grokking in real-world datasets compared to weight norms. Our findings suggest that while weight norms might usually correlate with grokking and our progress measures, they are not causative, and our proposed measures provide a better understanding of the dynamics of grokking.

Recent works have shown that large language models (LLMs) work remarkably well on a wide range of tasks through in-context learning and optimization of in-context examples (ICE). However, most of these studies assume either a fixed or no instruction provided in the prompt, leading to the apparent consensus that the optimization of in-context examples is critical for better performance. We challenge this consensus for instruction-tuned LLMs by investigating the necessity of optimizing in-context examples when task-specific instructions are provided, and find that there are tasks for which various ways of optimizing in-context examples yield diminishing returns. We introduce a task-specific metric called \metriclong{} (\metric) that quantifies the learnability of tasks from a given instruction, and provides a heuristic that helps decide whether to optimize for instructions or ICE for any new task. On a wide range of tasks and a systematically created instruction set with gradually added details, we validate our hypothesis empirically by computing \metric with query-dependent bins of examples, comparing different instructions with ICE selection methods, and performing label perturbation experiments. We conclude that tasks can be divided into two broad classes based on the \metric metric, where the returns on ICE optimization follow predictable trends when instructions are provided in the prompt.

The healthcare landscape is evolving, with patients seeking more reliable information about their health conditions, treatment options, and potential risks. Despite the abundance of information sources, the digital age overwhelms individuals with excess, often inaccurate information. Patients primarily trust doctors and hospital staff, highlighting the need for expert-endorsed health information. However, the pressure on experts has led to reduced communication time, impacting information sharing. To address this gap, we propose CataractBot, an experts-in-the-loop chatbot powered by large language models (LLMs). Developed in collaboration with a tertiary eye hospital in India, CataractBot answers cataract surgery related questions instantly by querying a curated knowledge base, and provides expert-verified responses asynchronously. CataractBot features multimodal support and multilingual capabilities. In an in-the-wild deployment study with 49 participants, CataractBot proved valuable, providing anytime accessibility, saving time, and accommodating diverse literacy levels. Trust was established through expert verification. Broadly, our results could inform future work on designing expert-mediated LLM bots.

Mechanistic interpretability (MI) aims to understand AI models by reverse-engineering the exact algorithms neural networks learn. Most works in MI so far have studied behaviors and capabilities that are trivial and token-aligned. However, most capabilities are not that trivial, which advocates for the study of hidden representations inside these networks as the unit of analysis. We do a literature review, formalize representations for features and behaviors, highlight their importance and evaluation, and perform some basic exploration in the mechanistic interpretability of representations. With discussion and exploratory results, we justify our position that studying representations is an important and under-studied field, and that currently established methods in MI are not sufficient to understand representations, thus pushing for the research community to work toward new frameworks for studying representations.