A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models

Large Language Models (LLMs) have revolutionized natural language processing, yet their internal mechanisms remain largely opaque. Recently, mechanistic interpretability has attracted significant attention from the research community as a means to understand the inner workings of LLMs. Among various mechanistic interpretability approaches, Sparse Autoencoders (SAEs) have emerged as a particularly promising method due to their ability to disentangle the complex, superimposed features within LLMs into more interpretable components. This paper presents a comprehensive examination of SAEs as a promising approach to interpreting and understanding LLMs. We provide a systematic overview of SAE principles, architectures, and applications specifically tailored for LLM analysis, covering theoretical foundations, implementation strategies, and recent developments in sparsity mechanisms. We also explore how SAEs can be leveraged to explain the internal workings of LLMs, steer model behaviors in desired directions, and develop more transparent training methodologies for future models. Despite the challenges that remain around SAE implementation and scaling, they continue to provide valuable tools for understanding the internal mechanisms of large language models.

Interpreting and Steering LLMs with Mutual Information-based Explanations on Sparse Autoencoders

Large language models (LLMs) excel at handling human queries, but they can occasionally generate flawed or unexpected responses. Understanding their internal states is crucial for understanding their successes, diagnosing their failures, and refining their capabilities. Although sparse autoencoders (SAEs) have shown promise for interpreting LLM internal representations, limited research has explored how to better explain SAE features, i.e., understanding the semantic meaning of features learned by SAE. Our theoretical analysis reveals that existing explanation methods suffer from the frequency bias issue, where they emphasize linguistic patterns over semantic concepts, while the latter is more critical to steer LLM behaviors. To address this, we propose using a fixed vocabulary set for feature interpretations and designing a mutual information-based objective, aiming to better capture the semantic meaning behind these features. We further propose two runtime steering strategies that adjust the learned feature activations based on their corresponding explanations. Empirical results show that, compared to baselines, our method provides more discourse-level explanations and effectively steers LLM behaviors to defend against jailbreak attacks. These findings highlight the value of explanations for steering LLM behaviors in downstream applications. We will release our code and data once accepted.

Self-Regularization with Latent Space Explanations for Controllable LLM-based Classification

Modern text classification methods heavily rely on contextual embeddings from large language models (LLMs). Compared to human-engineered features, these embeddings provide automatic and effective representations for classification model training. However, they also introduce a challenge: we lose the ability to manually remove unintended features, such as sensitive or task-irrelevant features, to guarantee regulatory compliance or improve the generalizability of classification models. This limitation arises because LLM embeddings are opaque and difficult to interpret. In this paper, we propose a novel framework to identify and regularize unintended features in the LLM latent space. Specifically, we first pre-train a sparse autoencoder (SAE) to extract interpretable features from LLM latent spaces. To ensure the SAE can capture task-specific features, we further fine-tune it on task-specific datasets. In training the classification model, we propose a simple and effective regularizer, by minimizing the similarity between the classifier weights and the identified unintended feature, to remove the impacts of these unintended features toward classification. We evaluate the proposed framework on three real-world tasks, including toxic chat detection, reward modeling, and disease diagnosis. Results show that the proposed framework can significantly improve the classifier's generalizability by regularizing those features that are not semantically correlated to each task. This work pioneers controllable text classification on LLM latent spaces by leveraging interpreted features to address generalizability, fairness, and privacy challenges. We will release our code and data once accepted.

LMOD: A Large Multimodal Ophthalmology Dataset and Benchmark for Large Vision-Language Models

Ophthalmology relies heavily on detailed image analysis for diagnosis and treatment planning. While large vision-language models (LVLMs) have shown promise in understanding complex visual information, their performance on ophthalmology images remains underexplored. We introduce LMOD, a dataset and benchmark for evaluating LVLMs on ophthalmology images, covering anatomical understanding, diagnostic analysis, and demographic extraction. LMODincludes 21,993 images spanning optical coherence tomography, scanning laser ophthalmoscopy, eye photos, surgical scenes, and color fundus photographs. We benchmark 13 state-of-the-art LVLMs and find that they are far from perfect for comprehending ophthalmology images. Models struggle with diagnostic analysis and demographic extraction, reveal weaknesses in spatial reasoning, diagnostic analysis, handling out-of-domain queries, and safeguards for handling biomarkers of ophthalmology images.

Retrieval-Enhanced Knowledge Editing for Multi-Hop Question Answering in Language Models

Large Language Models (LLMs) have shown proficiency in question-answering tasks but often struggle to integrate real-time knowledge updates, leading to potentially outdated or inaccurate responses. This problem becomes even more challenging when dealing with multi-hop questions since they require LLMs to update and integrate multiple knowledge pieces relevant to the questions. To tackle the problem, we propose the Retrieval-Augmented model Editing (RAE) framework tailored for multi-hop question answering. RAE first retrieves edited facts and then refines the language model through in-context learning. Specifically, our retrieval approach, based on mutual information maximization, leverages the reasoning abilities of LLMs to identify chain facts that na\"ive similarity-based searches might miss. Additionally, our framework incorporates a pruning strategy to eliminate redundant information from the retrieved facts, which enhances the editing accuracy and mitigates the hallucination problem. Our framework is supported by theoretical justification for its fact retrieval efficacy. Finally, comprehensive evaluation across various LLMs validates RAE's ability in providing accurate answers with updated knowledge.

Usable XAI: 10 Strategies Towards Exploiting Explainability in the LLM Era

Explainable AI (XAI) refers to techniques that provide human-understandable insights into the workings of AI models. Recently, the focus of XAI is being extended towards Large Language Models (LLMs) which are often criticized for their lack of transparency. This extension calls for a significant transformation in XAI methodologies because of two reasons. First, many existing XAI methods cannot be directly applied to LLMs due to their complexity advanced capabilities. Second, as LLMs are increasingly deployed across diverse industry applications, the role of XAI shifts from merely opening the "black box" to actively enhancing the productivity and applicability of LLMs in real-world settings. Meanwhile, unlike traditional machine learning models that are passive recipients of XAI insights, the distinct abilities of LLMs can reciprocally enhance XAI. Therefore, in this paper, we introduce Usable XAI in the context of LLMs by analyzing (1) how XAI can benefit LLMs and AI systems, and (2) how LLMs can contribute to the advancement of XAI. We introduce 10 strategies, introducing the key techniques for each and discussing their associated challenges. We also provide case studies to demonstrate how to obtain and leverage explanations. The code used in this paper can be found at: https://github.com/JacksonWuxs/UsableXAI_LLM.

InFoBench: Evaluating Instruction Following Ability in Large Language Models

This paper introduces the Decomposed Requirements Following Ratio (DRFR), a new metric for evaluating Large Language Models' (LLMs) ability to follow instructions. Addressing a gap in current methodologies, DRFR breaks down complex instructions into simpler criteria, facilitating a detailed analysis of LLMs' compliance with various aspects of tasks. Alongside this metric, we present InFoBench, a benchmark comprising 500 diverse instructions and 2,250 decomposed questions across multiple constraint categories. Our experiments compare DRFR with traditional scoring methods and explore annotation sources, including human experts, crowd-sourced workers, and GPT-4. The findings demonstrate DRFR's higher reliability and the effectiveness of using GPT-4 as a cost-efficient annotator. The evaluation of several advanced LLMs using this framework reveals their strengths and areas needing improvement, particularly in complex instruction-following. This study contributes a novel metric and benchmark, offering insights for future LLM development and evaluation.

Applying Large Language Models and Chain-of-Thought for Automatic Scoring

This study investigates the application of large language models (LLMs), specifically GPT-3.5 and GPT-4, with Chain-of-Though (CoT)in the automatic scoring of student-written responses to science assessments. We focused on overcoming the challenges of accessibility, technical complexity, and lack of explainability that have previously limited the use of automatic assessment tools among researchers and educators. We used a testing dataset comprising six assessment tasks (three binomial and three trinomial) with 1,650 student responses. We employed six prompt engineering strategies, combining zero-shot or few-shot learning with CoT, either alone or alongside item stem and scoring rubrics. Results indicated that few-shot (acc = .67) outperformed zero-shot learning (acc = .60), with 12.6\% increase. CoT, when used without item stem and scoring rubrics, did not significantly affect scoring accuracy (acc = .60). However, CoT prompting paired with contextual item stems and rubrics proved to be a significant contributor to scoring accuracy (13.44\% increase for zero-shot; 3.7\% increase for few-shot). Using a novel approach PPEAS, we found a more balanced accuracy across different proficiency categories, highlighting the importance of domain-specific reasoning in enhancing the effectiveness of LLMs in scoring tasks. Additionally, we also found that GPT-4 demonstrated superior performance over GPT-3.5 in various scoring tasks, showing 8.64\% difference. The study revealed that the single-call strategy with GPT-4, particularly using greedy sampling, outperformed other approaches, including ensemble voting strategies. This study demonstrates the potential of LLMs in facilitating automatic scoring, emphasizing that CoT enhances accuracy, particularly when used with item stem and scoring rubrics.

From Language Modeling to Instruction Following: Understanding the Behavior Shift in LLMs after Instruction Tuning

Large Language Models (LLMs) have achieved remarkable success, demonstrating powerful instruction-following capabilities across diverse tasks. Instruction fine-tuning is critical in enabling LLMs to align with user intentions and effectively follow instructions. In this work, we investigate how instruction fine-tuning modifies pre-trained models, focusing on two perspectives: instruction recognition and knowledge evolution. To study the behavior shift of LLMs, we employ a suite of local and global explanation methods, including a gradient-based approach for input-output attribution and techniques for interpreting patterns and concepts in self-attention and feed-forward layers. Our findings reveal three significant impacts of instruction fine-tuning: 1) It empowers LLMs to better recognize the instruction parts from user prompts, thereby facilitating high-quality response generation and addressing the ``lost-in-the-middle'' issue observed in pre-trained models; 2) It aligns the knowledge stored in feed-forward layers with user-oriented tasks, exhibiting minimal shifts across linguistic levels. 3) It facilitates the learning of word-word relations with instruction verbs through the self-attention mechanism, particularly in the lower and middle layers, indicating enhanced recognition of instruction words. These insights contribute to a deeper understanding of the behavior shifts in LLMs after instruction fine-tuning and lay the groundwork for future research aimed at interpreting and optimizing LLMs for various applications. We will release our code and data soon.

Towards Personalized Cold-Start Recommendation with Prompts

Recommender systems play a crucial role in helping users discover information that aligns with their interests based on their past behaviors. However, developing personalized recommendation systems becomes challenging when historical records of user-item interactions are unavailable, leading to what is known as the system cold-start recommendation problem. This issue is particularly prominent in start-up businesses or platforms with insufficient user engagement history. Previous studies focus on user or item cold-start scenarios, where systems could make recommendations for new users or items but are still trained with historical user-item interactions in the same domain, which cannot solve our problem. To bridge the gap, our research introduces an innovative and effective approach, capitalizing on the capabilities of pre-trained language models. We transform the recommendation process into sentiment analysis of natural languages containing information of user profiles and item attributes, where the sentiment polarity is predicted with prompt learning. By harnessing the extensive knowledge housed within language models, the prediction can be made without historical user-item interaction records. A benchmark is also introduced to evaluate the proposed method under the cold-start setting, and the results demonstrate the effectiveness of our method. To the best of our knowledge, this is the first study to tackle the system cold-start recommendation problem. The benchmark and implementation of the method are available at https://github.com/JacksonWuxs/PromptRec.