FaithUn: Toward Faithful Forgetting in Language Models by Investigating the Interconnectedness of Knowledge

Various studies have attempted to remove sensitive or private knowledge from a language model to prevent its unauthorized exposure. However, prior studies have overlooked the complex and interconnected nature of knowledge, where related knowledge must be carefully examined. Specifically, they have failed to evaluate whether an unlearning method faithfully erases interconnected knowledge that should be removed, retaining knowledge that appears relevant but exists in a completely different context. To resolve this problem, we first define a new concept called superficial unlearning, which refers to the phenomenon where an unlearning method either fails to erase the interconnected knowledge it should remove or unintentionally erases irrelevant knowledge. Based on the definition, we introduce a new benchmark, FaithUn, to analyze and evaluate the faithfulness of unlearning in real-world knowledge QA settings. Furthermore, we propose a novel unlearning method, KLUE, which updates only knowledge-related neurons to achieve faithful unlearning. KLUE identifies knowledge neurons using an explainability method and updates only those neurons using selected unforgotten samples. Experimental results demonstrate that widely-used unlearning methods fail to ensure faithful unlearning, while our method shows significant effectiveness in real-world QA unlearning.

Drift: Decoding-time Personalized Alignments with Implicit User Preferences

Personalized alignments for individual users have been a long-standing goal in large language models (LLMs). We introduce Drift, a novel framework that personalizes LLMs at decoding time with implicit user preferences. Traditional Reinforcement Learning from Human Feedback (RLHF) requires thousands of annotated examples and expensive gradient updates. In contrast, Drift personalizes LLMs in a training-free manner, using only a few dozen examples to steer a frozen model through efficient preference modeling. Our approach models user preferences as a composition of predefined, interpretable attributes and aligns them at decoding time to enable personalized generation. Experiments on both a synthetic persona dataset (Perspective) and a real human-annotated dataset (PRISM) demonstrate that Drift significantly outperforms RLHF baselines while using only 50-100 examples. Our results and analysis show that Drift is both computationally efficient and interpretable.

LLMs can be easily Confused by Instructional Distractions

Despite the fact that large language models (LLMs) show exceptional skill in instruction following tasks, this strength can turn into a vulnerability when the models are required to disregard certain instructions. Instruction-following tasks typically involve a clear task description and input text containing the target data to be processed. However, when the input itself resembles an instruction, confusion may arise, even if there is explicit prompting to distinguish between the task instruction and the input. We refer to this phenomenon as instructional distraction. In this paper, we introduce a novel benchmark, named DIM-Bench, specifically designed to assess LLMs' performance under instructional distraction. The benchmark categorizes real-world instances of instructional distraction and evaluates LLMs across four instruction tasks: rewriting, proofreading, translation, and style transfer -- alongside five input tasks: reasoning, code generation, mathematical reasoning, bias detection, and question answering. Our experimental results reveal that even the most advanced LLMs are susceptible to instructional distraction, often failing to accurately follow user intent in such cases.

Generating Diverse Hypotheses for Inductive Reasoning

Inductive reasoning - the process of inferring general rules from a small number of observations - is a fundamental aspect of human intelligence. Recent works suggest that large language models (LLMs) can engage in inductive reasoning by sampling multiple hypotheses about the rules and selecting the one that best explains the observations. However, due to the IID sampling, semantically redundant hypotheses are frequently generated, leading to significant wastage of compute. In this paper, we 1) demonstrate that increasing the temperature to enhance the diversity is limited due to text degeneration issue, and 2) propose a novel method to improve the diversity while maintaining text quality. We first analyze the effect of increasing the temperature parameter, which is regarded as the LLM's diversity control, on IID hypotheses. Our analysis shows that as temperature rises, diversity and accuracy of hypotheses increase up to a certain point, but this trend saturates due to text degeneration. To generate hypotheses that are more semantically diverse and of higher quality, we propose a novel approach inspired by human inductive reasoning, which we call Mixture of Concepts (MoC). When applied to several inductive reasoning benchmarks, MoC demonstrated significant performance improvements compared to standard IID sampling and other approaches.

PLM-Based Discrete Diffusion Language Models with Entropy-Adaptive Gibbs Sampling

Recently, discrete diffusion language models have demonstrated promising results in NLP. However, there has been limited research on integrating Pretrained Language Models (PLMs) into discrete diffusion models, resulting in underwhelming performance in downstream NLP generation tasks. This integration is particularly challenging because of the discrepancy between step-wise denoising strategy of diffusion models and single-step mask prediction approach of MLM-based PLMs. In this paper, we introduce Diffusion-EAGS, a novel approach that effectively integrates PLMs with the diffusion models. Furthermore, as it is challenging for PLMs to determine where to apply denoising during the diffusion process, we integrate an entropy tracking module to assist them. Finally, we propose entropy-based noise scheduling in the forward process to improve the effectiveness of entropy-adaptive sampling throughout the generation phase. Experimental results show that Diffusion-EAGS outperforms existing diffusion baselines in downstream generation tasks, achieving high text quality and diversity with precise token-level control. We also show that our model is capable of adapting to bilingual and low-resource settings, which are common in real-world applications.

Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation

In line with the principle of honesty, there has been a growing effort to train large language models (LLMs) to generate outputs containing epistemic markers. However, evaluation in the presence of epistemic markers has been largely overlooked, raising a critical question: Could the use of epistemic markers in LLM-generated outputs lead to unintended negative consequences? To address this, we present EMBER, a benchmark designed to assess the robustness of LLM-judges to epistemic markers in both single and pairwise evaluation settings. Our findings, based on evaluations using EMBER, reveal that all tested LLM-judges, including GPT-4o, show a notable lack of robustness in the presence of epistemic markers. Specifically, we observe a negative bias toward epistemic markers, with a stronger bias against markers expressing uncertainty. This suggests that LLM-judges are influenced by the presence of these markers and do not focus solely on the correctness of the content.

SWITCH: Studying with Teacher for Knowledge Distillation of Large Language Models

Despite the success of Large Language Models (LLMs), they still face challenges related to high inference costs and memory requirements. To address these issues, Knowledge Distillation (KD) has emerged as a popular method for model compression, with student-generated outputs (SGOs) being particularly notable for reducing the mismatch between training and inference. However, SGOs often produce noisy and biased sequences, which can lead to misguidance from the teacher model, especially in long sequences. To mitigate these challenges, we propose SWITCH (Studying WIth TeaCHer for Knowledge Distillation), a novel approach that strategically incorporates the teacher model during the student's sequence generation. SWITCH identifies discrepancies between the token probabilities of the teacher and student models, allowing the teacher to intervene selectively, particularly in long sequences that are more prone to teacher misguidance. Extensive experimental results across three model families and five instruction-following datasets show that SWITCH surpasses traditional KD methods, particularly excelling in the generation of long sequential data.

Guaranteed Generation from Large Language Models

As large language models (LLMs) are increasingly used across various applications, there is a growing need to control text generation to satisfy specific constraints or requirements. This raises a crucial question: Is it possible to guarantee strict constraint satisfaction in generated outputs while preserving the distribution of the original model as much as possible? We first define the ideal distribution - the one closest to the original model, which also always satisfies the expressed constraint - as the ultimate goal of guaranteed generation. We then state a fundamental limitation, namely that it is impossible to reach that goal through autoregressive training alone. This motivates the necessity of combining training-time and inference-time methods to enforce such guarantees. Based on this insight, we propose GUARD, a simple yet effective approach that combines an autoregressive proposal distribution with rejection sampling. Through GUARD's theoretical properties, we show how controlling the KL divergence between a specific proposal and the target ideal distribution simultaneously optimizes inference speed and distributional closeness. To validate these theoretical concepts, we conduct extensive experiments on two text generation settings with hard-to-satisfy constraints: a lexical constraint scenario and a sentiment reversal scenario. These experiments show that GUARD achieves perfect constraint satisfaction while almost preserving the ideal distribution with highly improved inference efficiency. GUARD provides a principled approach to enforcing strict guarantees for LLMs without compromising their generative capabilities.

A Character-Centric Creative Story Generation via Imagination

Creative story generation with diverse and detailed story elements is a long-standing goal for large language models. While existing methodologies generate long and coherent stories, they fall significantly short of human capabilities in terms of diversity and character detail. To address this, we introduce a novel story generation framework called CCI (Character-centric Creative story generation via Imagination). CCI features two innovative modules for creative story generation: IG (Image-Guided Imagination) and MW (Multi-Writer model). In the IG module, we utilize DALL-E 3 to create visual representations of key story elements. The IG generates more novel and concrete characters, backgrounds, and main plots than text-only methods. The MW module uses these story elements created by IG to generate multiple description candidates for the protagonist and select the best one. This method incorporates vivid and rich character descriptions into the story. We compared the stories generated by CCI and baseline models through human evaluation and statistical analysis. The results showed significant improvements in the creativity. Furthermore, by enabling interactive multi-modal story generation with users, we have opened up possibilities for human-LLM integration in cultural development.

Persona is a Double-edged Sword: Enhancing the Zero-shot Reasoning by Ensembling the Role-playing and Neutral Prompts

Recent studies demonstrate that prompting an appropriate role-playing persona to an LLM improves its reasoning capability. However, assigning a proper persona is difficult since an LLM's performance is extremely sensitive to assigned prompts; therefore, personas sometimes hinder LLMs and degrade their reasoning capabilities. In this paper, we propose a novel framework, Jekyll \& Hyde, which ensembles the results of role-playing and neutral prompts to eradicate performance degradation via unilateral use of role-playing prompted LLM and enhance the robustness of an LLM's reasoning ability. Specifically, Jekyll \& Hyde collects two potential solutions from both role-playing and neutral prompts and selects a better solution after cross-checking via an LLM evaluator. However, LLM-based evaluators tend to be affected by the order of those potential solutions within the prompt when selecting the proper solution; thus, we also propose a robust LLM evaluator to mitigate the position bias. The experimental analysis demonstrates that role-playing prompts distract LLMs and degrade their reasoning abilities in 4 out of 12 datasets, even when using GPT-4. In addition, we reveal that Jekyll \& Hyde improves reasoning capabilities by selecting better choices among the potential solutions on twelve widely-used reasoning datasets. We further show that our proposed LLM evaluator outperforms other baselines, proving the LLMs' position bias is successfully mitigated.