What Do the Circuits Mean? A Knowledge Edit View

In the field of language model interpretability, circuit discovery is gaining popularity. Despite this, the true meaning of these circuits remain largely unanswered. We introduce a novel method to learn their meanings as a holistic object through the lens of knowledge editing. We extract circuits in the GPT2-XL model using diverse text classification datasets, and use hierarchical relations datasets to explore knowledge editing in the circuits. Our findings indicate that these circuits contain entity knowledge but resist new knowledge more than complementary circuits during knowledge editing. Additionally, we examine the impact of circuit size, discovering that an ideal "theoretical circuit" where essential knowledge is concentrated likely incorporates more than 5% but less than 50% of the model's parameters. We also assess the overlap between circuits from different datasets, finding moderate similarities. What constitutes these circuits, then? We find that up to 60% of the circuits consist of layer normalization modules rather than attention or MLP modules, adding evidence to the ongoing debates regarding knowledge localization. In summary, our findings offer new insights into the functions of the circuits, and introduce research directions for further interpretability and safety research of language models.

Self-Supervised Embeddings for Detecting Individual Symptoms of Depression

Depression, a prevalent mental health disorder impacting millions globally, demands reliable assessment systems. Unlike previous studies that focus solely on either detecting depression or predicting its severity, our work identifies individual symptoms of depression while also predicting its severity using speech input. We leverage self-supervised learning (SSL)-based speech models to better utilize the small-sized datasets that are frequently encountered in this task. Our study demonstrates notable performance improvements by utilizing SSL embeddings compared to conventional speech features. We compare various types of SSL pretrained models to elucidate the type of speech information (semantic, speaker, or prosodic) that contributes the most in identifying different symptoms. Additionally, we evaluate the impact of combining multiple SSL embeddings on performance. Furthermore, we show the significance of multi-task learning for identifying depressive symptoms effectively.

How Well Can Knowledge Edit Methods Edit Perplexing Knowledge?

As large language models (LLMs) are widely deployed, targeted editing of their knowledge has become a critical challenge. Recently, advancements in model editing techniques, such as Rank-One Model Editing (ROME), have paved the way for updating LLMs with new knowledge. However, the efficacy of these methods varies across different types of knowledge. This study investigates the capability of knowledge editing methods to incorporate new knowledge with varying degrees of "perplexingness", a term we use to describe the initial difficulty LLMs have in understanding new concepts. We begin by quantifying the "perplexingness" of target knowledge using pre-edit conditional probabilities, and assess the efficacy of edits through post-edit conditional probabilities. Utilizing the widely-used CounterFact dataset, we find significant negative correlations between the "perplexingness" of the new knowledge and the edit efficacy across all 12 scenarios. To dive deeper into this phenomenon, we introduce a novel dataset, HierarchyData, consisting of 99 hyponym-hypernym pairs across diverse categories. Our analysis reveal that more abstract concepts (hypernyms) tend to be more perplexing than their specific counterparts (hyponyms). Further exploration into the influence of knowledge hierarchy on editing outcomes indicates that knowledge positioned at higher hierarchical levels is more challenging to modify in some scenarios. Our research highlights a previously overlooked aspect of LLM editing: the variable efficacy of editing methods in handling perplexing knowledge. By revealing how hierarchical relationships can influence editing outcomes, our findings offer new insights into the challenges of updating LLMs and pave the way for more nuanced approaches to model editing in the future.

The GPT-WritingPrompts Dataset: A Comparative Analysis of Character Portrayal in Short Stories

The improved generative capabilities of large language models have made them a powerful tool for creative writing and storytelling. It is therefore important to quantitatively understand the nature of generated stories, and how they differ from human storytelling. We augment the Reddit WritingPrompts dataset with short stories generated by GPT-3.5, given the same prompts. We quantify and compare the emotional and descriptive features of storytelling from both generative processes, human and machine, along a set of six dimensions. We find that generated stories differ significantly from human stories along all six dimensions, and that human and machine generations display similar biases when grouped according to the narrative point-of-view and gender of the main protagonist. We release our dataset and code at https://github.com/KristinHuangg/gpt-writing-prompts.

Scenarios and Approaches for Situated Natural Language Explanations

Large language models (LLMs) can be used to generate natural language explanations (NLE) that are adapted to different users' situations. However, there is yet to be a quantitative evaluation of the extent of such adaptation. To bridge this gap, we collect a benchmarking dataset, Situation-Based Explanation. This dataset contains 100 explanandums. Each explanandum is paired with explanations targeted at three distinct audience types-such as educators, students, and professionals-enabling us to assess how well the explanations meet the specific informational needs and contexts of these diverse groups e.g. students, teachers, and parents. For each "explanandum paired with an audience" situation, we include a human-written explanation. These allow us to compute scores that quantify how the LLMs adapt the explanations to the situations. On an array of pretrained language models with varying sizes, we examine three categories of prompting methods: rule-based prompting, meta-prompting, and in-context learning prompting. We find that 1) language models can generate prompts that result in explanations more precisely aligned with the target situations, 2) explicitly modeling an "assistant" persona by prompting "You are a helpful assistant..." is not a necessary prompt technique for situated NLE tasks, and 3) the in-context learning prompts only can help LLMs learn the demonstration template but can't improve their inference performance. SBE and our analysis facilitate future research towards generating situated natural language explanations.

Filtered not Mixed: Stochastic Filtering-Based Online Gating for Mixture of Large Language Models

We propose MoE-F -- a formalised mechanism for combining $N$ pre-trained expert Large Language Models (LLMs) in online time-series prediction tasks by adaptively forecasting the best weighting of LLM predictions at every time step. Our mechanism leverages the conditional information in each expert's running performance to forecast the best combination of LLMs for predicting the time series in its next step. Diverging from static (learned) Mixture of Experts (MoE) methods, MoE-F employs time-adaptive stochastic filtering techniques to combine experts. By framing the expert selection problem as a finite state-space, continuous-time Hidden Markov model (HMM), we can leverage the Wohman-Shiryaev filter. Our approach first constructs $N$ parallel filters corresponding to each of the $N$ individual LLMs. Each filter proposes its best combination of LLMs, given the information that they have access to. Subsequently, the $N$ filter outputs are aggregated to optimize a lower bound for the loss of the aggregated LLMs, which can be optimized in closed-form, thus generating our ensemble predictor. Our contributions here are: (I) the MoE-F algorithm -- deployable as a plug-and-play filtering harness, (II) theoretical optimality guarantees of the proposed filtering-based gating algorithm, and (III) empirical evaluation and ablative results using state of the art foundational and MoE LLMs on a real-world Financial Market Movement task where MoE-F attains a remarkable 17% absolute and 48.5% relative F1 measure improvement over the next best performing individual LLM expert.

$\texttt{ACCORD}$: Closing the Commonsense Measurability Gap

We present $\texttt{ACCORD}$, a framework and benchmark suite for disentangling the commonsense grounding and reasoning abilities of large language models (LLMs) through controlled, multi-hop counterfactuals. $\texttt{ACCORD}$ introduces formal elements to commonsense reasoning to explicitly control and quantify reasoning complexity beyond the typical 1 or 2 hops. Uniquely, $\texttt{ACCORD}$ can automatically generate benchmarks of arbitrary reasoning complexity, and so it scales with future LLM improvements. Benchmarking state-of-the-art LLMs -- including GPT-4o (2024-05-13), Llama-3-70B-Instruct, and Mixtral-8x22B-Instruct-v0.1 -- shows performance degrading to random chance with only moderate scaling, leaving substantial headroom for improvement. We release a leaderboard of the benchmark suite tested in this work, as well as code for automatically generating more complex benchmarks.

Auxiliary Knowledge-Induced Learning for Automatic Multi-Label Medical Document Classification

The International Classification of Diseases (ICD) is an authoritative medical classification system of different diseases and conditions for clinical and management purposes. ICD indexing assigns a subset of ICD codes to a medical record. Since human coding is labour-intensive and error-prone, many studies employ machine learning to automate the coding process. ICD coding is a challenging task, as it needs to assign multiple codes to each medical document from an extremely large hierarchically organized collection. In this paper, we propose a novel approach for ICD indexing that adopts three ideas: (1) we use a multi-level deep dilated residual convolution encoder to aggregate the information from the clinical notes and learn document representations across different lengths of the texts; (2) we formalize the task of ICD classification with auxiliary knowledge of the medical records, which incorporates not only the clinical texts but also different clinical code terminologies and drug prescriptions for better inferring the ICD codes; and (3) we introduce a graph convolutional network to leverage the co-occurrence patterns among ICD codes, aiming to enhance the quality of label representations. Experimental results show the proposed method achieves state-of-the-art performance on a number of measures.

Multi-stage Retrieve and Re-rank Model for Automatic Medical Coding Recommendation

The International Classification of Diseases (ICD) serves as a definitive medical classification system encompassing a wide range of diseases and conditions. The primary objective of ICD indexing is to allocate a subset of ICD codes to a medical record, which facilitates standardized documentation and management of various health conditions. Most existing approaches have suffered from selecting the proper label subsets from an extremely large ICD collection with a heavy long-tailed label distribution. In this paper, we leverage a multi-stage ``retrieve and re-rank'' framework as a novel solution to ICD indexing, via a hybrid discrete retrieval method, and re-rank retrieved candidates with contrastive learning that allows the model to make more accurate predictions from a simplified label space. The retrieval model is a hybrid of auxiliary knowledge of the electronic health records (EHR) and a discrete retrieval method (BM25), which efficiently collects high-quality candidates. In the last stage, we propose a label co-occurrence guided contrastive re-ranking model, which re-ranks the candidate labels by pulling together the clinical notes with positive ICD codes. Experimental results show the proposed method achieves state-of-the-art performance on a number of measures on the MIMIC-III benchmark.

Representation noising effectively prevents harmful fine-tuning on LLMs

Releasing open-source large language models (LLMs) presents a dual-use risk since bad actors can easily fine-tune these models for harmful purposes. Even without the open release of weights, weight stealing and fine-tuning APIs make closed models vulnerable to harmful fine-tuning attacks (HFAs). While safety measures like preventing jailbreaks and improving safety guardrails are important, such measures can easily be reversed through fine-tuning. In this work, we propose Representation Noising (RepNoise), a defence mechanism that is effective even when attackers have access to the weights and the defender no longer has any control. RepNoise works by removing information about harmful representations such that it is difficult to recover them during fine-tuning. Importantly, our defence is also able to generalize across different subsets of harm that have not been seen during the defence process. Our method does not degrade the general capability of LLMs and retains the ability to train the model on harmless tasks. We provide empirical evidence that the effectiveness of our defence lies in its "depth": the degree to which information about harmful representations is removed across all layers of the LLM.