ReasonEdit: Editing Vision-Language Models using Human Reasoning

Model editing aims to correct errors in large, pretrained models without altering unrelated behaviors. While some recent works have edited vision-language models (VLMs), no existing editors tackle reasoning-heavy tasks, which typically require humans and models to reason about images. We therefore propose ReasonEdit, the first VLM editor to let users explain their reasoning during editing, introducing a new, practical model editing setup. ReasonEdit continuously stores human reasoning in a codebook, and retrieves only relevant facts during inference using a novel topology-balanced multimodal embedding method inspired by network science. Across four VLMs on multiple rationale-based visual question answering datasets, ReasonEdit achieves state-of-the-art editing performance, ultimately showing that using human reasoning during editing greatly improves edit generalization.

Aligning Language Model Benchmarks with Pairwise Preferences

Language model benchmarks are pervasive and computationally-efficient proxies for real-world performance. However, many recent works find that benchmarks often fail to predict real utility. Towards bridging this gap, we introduce benchmark alignment, where we use limited amounts of information about model performance to automatically update offline benchmarks, aiming to produce new static benchmarks that predict model pairwise preferences in given test settings. We then propose BenchAlign, the first solution to this problem, which learns preference-aligned weight- ings for benchmark questions using the question-level performance of language models alongside ranked pairs of models that could be collected during deployment, producing new benchmarks that rank previously unseen models according to these preferences. Our experiments show that our aligned benchmarks can accurately rank unseen models according to models of human preferences, even across different sizes, while remaining interpretable. Overall, our work provides insights into the limits of aligning benchmarks with practical human preferences, which stands to accelerate model development towards real utility.

Improving and Accelerating Offline RL in Large Discrete Action Spaces with Structured Policy Initialization

Reinforcement learning in discrete combinatorial action spaces requires searching over exponentially many joint actions to simultaneously select multiple sub-actions that form coherent combinations. Existing approaches either simplify policy learning by assuming independence across sub-actions, which often yields incoherent or invalid actions, or attempt to learn action structure and control jointly, which is slow and unstable. We introduce Structured Policy Initialization (SPIN), a two-stage framework that first pre-trains an Action Structure Model (ASM) to capture the manifold of valid actions, then freezes this representation and trains lightweight policy heads for control. On challenging discrete DM Control benchmarks, SPIN improves average return by up to 39% over the state of the art while reducing time to convergence by up to 12.8$\times$.

EDUMATH: Generating Standards-aligned Educational Math Word Problems

Math word problems (MWPs) are critical K-12 educational tools, and customizing them to students' interests and ability levels can increase learning outcomes. However, teachers struggle to find time to customize MWPs for each student given large class sizes and increasing burnout. We propose that LLMs can support math education by generating MWPs customized to student interests and math education standards. To this end, we use a joint human expert-LLM judge approach to evaluate over 11,000 MWPs generated by open and closed LLMs and develop the first teacher-annotated dataset for standards-aligned educational MWP generation. We show the value of our data by using it to train a 12B open model that matches the performance of larger and more capable open models. We also use our teacher-annotated data to train a text classifier that enables a 30B open LLM to outperform existing closed baselines without any training. Next, we show our models' MWPs are more similar to human-written MWPs than those from existing models. We conclude by conducting the first study of customized LLM-generated MWPs with grade school students, finding they perform similarly on our models' MWPs relative to human-written MWPs but consistently prefer our customized MWPs.

Estimating Semantic Alphabet Size for LLM Uncertainty Quantification

Many black-box techniques for quantifying the uncertainty of large language models (LLMs) rely on repeated LLM sampling, which can be computationally expensive. Therefore, practical applicability demands reliable estimation from few samples. Semantic entropy (SE) is a popular sample-based uncertainty estimator with a discrete formulation attractive for the black-box setting. Recent extensions of semantic entropy exhibit improved LLM hallucination detection, but do so with less interpretable methods that admit additional hyperparameters. For this reason, we revisit the canonical discrete semantic entropy estimator, finding that it underestimates the "true" semantic entropy, as expected from theory. We propose a modified semantic alphabet size estimator, and illustrate that using it to adjust discrete semantic entropy for sample coverage results in more accurate semantic entropy estimation in our setting of interest. Furthermore, our proposed alphabet size estimator flags incorrect LLM responses as well or better than recent top-performing approaches, with the added benefit of remaining highly interpretable.

KScope: A Framework for Characterizing the Knowledge Status of Language Models

Characterizing a large language model's (LLM's) knowledge of a given question is challenging. As a result, prior work has primarily examined LLM behavior under knowledge conflicts, where the model's internal parametric memory contradicts information in the external context. However, this does not fully reflect how well the model knows the answer to the question. In this paper, we first introduce a taxonomy of five knowledge statuses based on the consistency and correctness of LLM knowledge modes. We then propose KScope, a hierarchical framework of statistical tests that progressively refines hypotheses about knowledge modes and characterizes LLM knowledge into one of these five statuses. We apply KScope to nine LLMs across four datasets and systematically establish: (1) Supporting context narrows knowledge gaps across models. (2) Context features related to difficulty, relevance, and familiarity drive successful knowledge updates. (3) LLMs exhibit similar feature preferences when partially correct or conflicted, but diverge sharply when consistently wrong. (4) Context summarization constrained by our feature analysis, together with enhanced credibility, further improves update effectiveness and generalizes across LLMs.

Efficient Knowledge Editing via Minimal Precomputation

Knowledge editing methods like MEMIT are able to make data and compute efficient updates of factual knowledge by using a single sentence to update facts and their consequences. However, what is often overlooked is a "precomputation step", which requires a one-time but significant computational cost. The authors of MEMIT originally precompute approximately 44 million hidden vectors per edited layer, which requires a forward pass over 44 million tokens. For GPT-J (6B), this precomputation step takes 36 hours on a single GPU, while it takes approximately 40 hours for Llama2-7B. Additionally, this precomputation time grows with model size. In this paper, we show that this excessive computational cost is unnecessary. Knowledge editing using MEMIT and related methods, such as ROME and EMMET, can be performed by pre-computing a very small portion of the 44 million hidden vectors. We first present the theoretical minimum number of hidden vector precomputation required for solutions of these editing methods to exist. We then empirically show that knowledge editing using these methods can be done by pre-computing significantly fewer hidden vectors. Specifically, we show that the precomputation step can be done with less than 0.3% of the originally stipulated number of hidden vectors. This saves a significant amount of precomputation time and allows users to begin editing new models within a few minutes.

Model Editing with Graph-Based External Memory

Large language models (LLMs) have revolutionized natural language processing, yet their practical utility is often limited by persistent issues of hallucinations and outdated parametric knowledge. Although post-training model editing offers a pathway for dynamic updates, existing methods frequently suffer from overfitting and catastrophic forgetting. To tackle these challenges, we propose a novel framework that leverages hyperbolic geometry and graph neural networks for precise and stable model edits. We introduce HYPE (HYperbolic Parameter Editing), which comprises three key components: (i) Hyperbolic Graph Construction, which uses Poincar\'e embeddings to represent knowledge triples in hyperbolic space, preserving hierarchical relationships and preventing unintended side effects by ensuring that edits to parent concepts do not inadvertently affect child concepts; (ii) M\"obius-Transformed Updates, which apply hyperbolic addition to propagate edits while maintaining structural consistency within the hyperbolic manifold, unlike conventional Euclidean updates that distort relational distances; and (iii) Dual Stabilization, which combines gradient masking and periodic GNN parameter resetting to prevent catastrophic forgetting by focusing updates on critical parameters and preserving long-term knowledge. Experiments on CounterFact, CounterFact+, and MQuAKE with GPT-J and GPT2-XL demonstrate that HYPE significantly enhances edit stability, factual accuracy, and multi-hop reasoning.

Continually Self-Improving Language Models for Bariatric Surgery Question--Answering

While bariatric and metabolic surgery (MBS) is considered the gold standard treatment for severe and morbid obesity, its therapeutic efficacy hinges upon active and longitudinal engagement with multidisciplinary providers, including surgeons, dietitians/nutritionists, psychologists, and endocrinologists. This engagement spans the entire patient journey, from preoperative preparation to long-term postoperative management. However, this process is often hindered by numerous healthcare disparities, such as logistical and access barriers, which impair easy patient access to timely, evidence-based, clinician-endorsed information. To address these gaps, we introduce bRAGgen, a novel adaptive retrieval-augmented generation (RAG)-based model that autonomously integrates real-time medical evidence when response confidence dips below dynamic thresholds. This self-updating architecture ensures that responses remain current and accurate, reducing the risk of misinformation. Additionally, we present bRAGq, a curated dataset of 1,302 bariatric surgery--related questions, validated by an expert bariatric surgeon. bRAGq constitutes the first large-scale, domain-specific benchmark for comprehensive MBS care. In a two-phase evaluation, bRAGgen is benchmarked against state-of-the-art models using both large language model (LLM)--based metrics and expert surgeon review. Across all evaluation dimensions, bRAGgen demonstrates substantially superior performance in generating clinically accurate and relevant responses.

Breaking mBad! Supervised Fine-tuning for Cross-Lingual Detoxification

As large language models (LLMs) become increasingly prevalent in global applications, ensuring that they are toxicity-free across diverse linguistic contexts remains a critical challenge. We explore "Cross-lingual Detoxification", a cross-lingual paradigm that mitigates toxicity, enabling detoxification capabilities to transfer between high and low-resource languages across different script families. We analyze cross-lingual detoxification's effectiveness through 504 extensive settings to evaluate toxicity reduction in cross-distribution settings with limited data and investigate how mitigation impacts model performance on non-toxic tasks, revealing trade-offs between safety and knowledge preservation. Our code and dataset are publicly available at https://github.com/himanshubeniwal/Breaking-mBad.