Handwritten text recognition (HTR) remains a challenging task, particularly for multi-page documents where pages share common formatting and contextual features. While modern optical character recognition (OCR) engines are proficient with printed text, their performance on handwriting is limited, often requiring costly labeled data for fine-tuning. In this paper, we explore the use of multi-modal large language models (MLLMs) for transcribing multi-page handwritten documents in a zero-shot setting. We investigate various configurations of commercial OCR engines and MLLMs, utilizing the latter both as end-to-end transcribers and as post-processors, with and without image components. We propose a novel method, + FIRST PAGE, which enhances MLLM transcription by providing the OCR output of the entire document along with just the first page image . This approach leverages shared document features without incurring the high cost of processing all images. Experiments on a multi-page version of the IAM Handwriting Database demonstrate that + FIRST PAGE improves transcription accuracy, balances cost with performance, and even enhances results on out-of-sample text by extrapolating formatting and OCR error patterns from a single page. Introduction A significant proportion of all human-written text exists only in the form of physical handwritten documents.

Large Language Models (LLMs) are increasingly used for planning tasks, offering unique capabilities not found in classical planners such as generating explanations and iterative refinement. However, trust--a critical factor in the adoption of planning systems--remains underexplored in the context of LLM-based planning tasks. This study bridges this gap by comparing human trust in LLM-based planners with classical planners through a user study in a Planning Domain Definition Language (PDDL) domain. Combining subjective measures, such as trust questionnaires, with objective metrics like evaluation accuracy, our findings reveal that correctness is the primary driver of trust and performance. Explanations provided by the LLM improved evaluation accuracy but had limited impact on trust, while plan refinement showed potential for increasing trust without significantly enhancing evaluation accuracy.

Infections in Diabetic Foot Ulcers (DFUs) can cause severe complications, including tissue death and limb amputation, highlighting the need for accurate, timely diagnosis. Previous machine learning methods have focused on identifying infections by analyzing wound images alone, without utilizing additional metadata such as medical notes. In this study, we aim to improve infection detection by introducing Synthetic Caption Augmented Retrieval for Wound Infection Detection (SCARWID), a novel deep learning framework that leverages synthetic textual descriptions to augment DFU images. SCARWID consists of two components: (1) Wound-BLIP, a Vision-Language Model (VLM) fine-tuned on GPT-4o-generated descriptions to synthesize consistent captions from images; and (2) an Image-Text Fusion module that uses cross-attention to extract cross-modal embeddings from an image and its corresponding Wound-BLIP caption. Infection status is determined by retrieving the top-k similar items from a labeled support set. To enhance the diversity of training data, we utilized a latent diffusion model to generate additional wound images. As a result, SCARWID outperformed state-of-the-art models, achieving average sensitivity, specificity, and accuracy of 0.85, 0.78, and 0.81, respectively, for wound infection classification. Displaying the generated captions alongside the wound images and infection detection results enhances interpretability and trust, enabling nurses to align SCARWID outputs with their medical knowledge. This is particularly valuable when wound notes are unavailable or when assisting novice nurses who may find it difficult to identify visual attributes of wound infection.

As large language models are increasingly responsible for online content, concerns arise about the impact of repeatedly processing their own outputs. Inspired by the "broken telephone" effect in chained human communication, this study investigates whether LLMs similarly distort information through iterative generation. Through translation-based experiments, we find that distortion accumulates over time, influenced by language choice and chain complexity. While degradation is inevitable, it can be mitigated through strategic prompting techniques. These findings contribute to discussions on the long-term effects of AI-mediated information propagation, raising important questions about the reliability of LLM-generated content in iterative workflows.

Knowledge-intensive tasks, particularly open-domain question answering (ODQA), document reranking, and retrieval-augmented language modeling, require a balance between retrieval accuracy and generative flexibility. Traditional retrieval models such as BM25 and Dense Passage Retrieval (DPR), efficiently retrieve from large corpora but often lack semantic depth. Generative models like GPT-4-o provide richer contextual understanding but face challenges in maintaining factual consistency. In this work, we conduct a systematic evaluation of retrieval-based, generation-based, and hybrid models, with a primary focus on their performance in ODQA and related retrieval-augmented tasks. Our results show that dense retrievers, particularly DPR, achieve strong performance in ODQA with a top-1 accuracy of 50.17\% on NQ, while hybrid models improve nDCG@10 scores on BEIR from 43.42 (BM25) to 52.59, demonstrating their strength in document reranking. Additionally, we analyze language modeling tasks using WikiText-103, showing that retrieval-based approaches like BM25 achieve lower perplexity compared to generative and hybrid methods, highlighting their utility in retrieval-augmented generation. By providing detailed comparisons and practical insights into the conditions where each approach excels, we aim to facilitate future optimizations in retrieval, reranking, and generative models for ODQA and related knowledge-intensive applications.

Many challenging reasoning tasks require not just rapid, intuitive responses, but a more deliberate, multi-step approach. Recent progress in large language models (LLMs) highlights an important shift from the "System 1" way of quick reactions to the "System 2" style of reflection-and-correction problem solving. However, current benchmarks heavily rely on the final-answer accuracy, leaving much of a model's intermediate reasoning steps unexamined. This fails to assess the model's ability to reflect and rectify mistakes within the reasoning process. To bridge this gap, we introduce FINEREASON, a logic-puzzle benchmark for fine-grained evaluation of LLMs' reasoning capabilities. Each puzzle can be decomposed into atomic steps, making it ideal for rigorous validation of intermediate correctness. Building on this, we introduce two tasks: state checking, and state transition, for a comprehensive evaluation of how models assess the current situation and plan the next move. To support broader research, we also provide a puzzle training set aimed at enhancing performance on general mathematical tasks. We show that models trained on our state checking and transition data demonstrate gains in math reasoning by up to 5.1% on GSM8K.

While existing studies have recognised explicit biases in generative models, including occupational gender biases, the nuances of gender stereotypes and expectations of relationships between users and AI companions remain underexplored. In the meantime, AI companions have become increasingly popular as friends or gendered romantic partners to their users. This study bridges the gap by devising three experiments tailored for romantic, gender-assigned AI companions and their users, effectively evaluating implicit biases across various-sized LLMs. Each experiment looks at a different dimension: implicit associations, emotion responses, and sycophancy. This study aims to measure and compare biases manifested in different companion systems by quantitatively analysing persona-assigned model responses to a baseline through newly devised metrics. The results are noteworthy: they show that assigning gendered, relationship personas to Large Language Models significantly alters the responses of these models, and in certain situations in a biased, stereotypical way.

We introduce the Granite Embedding models, a family of encoder-based embedding models designed for retrieval tasks, spanning dense-retrieval and sparse-retrieval architectures, with both English and Multilingual capabilities. This report provides the technical details of training these highly effective 12 layer embedding models, along with their efficient 6 layer distilled counterparts. Extensive evaluations show that the models, developed with techniques like retrieval oriented pretraining, contrastive finetuning, knowledge distillation, and model merging significantly outperform publicly available models of similar sizes on both internal IBM retrieval and search tasks, and have equivalent performance on widely-used information retrieval benchmarks, while being trained on high-quality data suitable for enterprise use. We publicly release all our Granite Embedding models under the Apache 2.0 license, allowing both research and commercial use at https://huggingface.co/collections/ibm-granite . Figure 1: Average performance on the Granite embedding models (in blue) vs BGE, GTE, Snowflake, E5, and Nomic models on 5 QA and IR datasets: BEIR, ClapNQ, CoIR, RedHat, and UnifiedSearch (the last 2 are internal IBM datasets). The goal of text embedding models is to convert variable length text into a fixed vector, encoding the text semantics into a multidimensional vector in such a way that semantically close texts are close in the vector space, while dissimilar texts have a low similarity. These embeddings can then be used in a variety of tasks, most commonly in retrieval applications, where the relevance of a document to a given query can be determined by the similarity of their embeddings (Dunn et al., 2017; Xiong et al., 2020; Neelakantan et al., 2022)(Zamani et al., 2018; Zhao et al., 2020), but also in document clustering (Angelov, 2020) and text classification (Sun et al., 2019). See Contributions section for full author list.

With the increasing use of Large Language Models (LLMs) in fields such as e-commerce, domain-specific concept evaluation benchmarks are crucial for assessing their domain capabilities. Existing LLMs may generate factually incorrect information within the complex e-commerce applications. Therefore, it is necessary to build an e-commerce concept benchmark. Existing benchmarks encounter two primary challenges: (1) handle the heterogeneous and diverse nature of tasks, (2) distinguish between generality and specificity within the e-commerce field. To address these problems, we propose \textbf{ChineseEcomQA}, a scalable question-answering benchmark focused on fundamental e-commerce concepts. ChineseEcomQA is built on three core characteristics: \textbf{Focus on Fundamental Concept}, \textbf{E-commerce Generality} and \textbf{E-commerce Expertise}. Fundamental concepts are designed to be applicable across a diverse array of e-commerce tasks, thus addressing the challenge of heterogeneity and diversity. Additionally, by carefully balancing generality and specificity, ChineseEcomQA effectively differentiates between broad e-commerce concepts, allowing for precise validation of domain capabilities. We achieve this through a scalable benchmark construction process that combines LLM validation, Retrieval-Augmented Generation (RAG) validation, and rigorous manual annotation. Based on ChineseEcomQA, we conduct extensive evaluations on mainstream LLMs and provide some valuable insights. We hope that ChineseEcomQA could guide future domain-specific evaluations, and facilitate broader LLM adoption in e-commerce applications.

Large language models (LLMs) demonstrate strong task-specific capabilities through fine-tuning, but merging multiple fine-tuned models often leads to degraded performance due to overlapping instruction-following components. Task Arithmetic (TA), which combines task vectors derived from fine-tuning, enables multi-task learning and task forgetting but struggles to isolate task-specific knowledge from general instruction-following behavior. To address this, we propose Layer-Aware Task Arithmetic (LATA), a novel approach that assigns layer-specific weights to task vectors based on their alignment with instruction-following or task-specific components. By amplifying task-relevant layers and attenuating instruction-following layers, LATA improves task learning and forgetting performance while preserving overall model utility. Experiments on multiple benchmarks, including WikiText-2, GSM8K, and HumanEval, demonstrate that LATA outperforms existing methods in both multi-task learning and selective task forgetting, achieving higher task accuracy and alignment with minimal degradation in output quality. Our findings highlight the importance of layer-wise analysis in disentangling task-specific and general-purpose knowledge, offering a robust framework for efficient model merging and editing.