Recent retrieval-augmented models enhance basic methods by building a hierarchical structure over retrieved text chunks through recursive embedding, clustering, and summarization. The most relevant information is then retrieved from both the original text and generated summaries. However, such approaches face limitations with dynamic datasets, where adding or removing documents over time complicates the updating of hierarchical representations formed through clustering. We propose a new algorithm to efficiently maintain the recursive-abstractive tree structure in dynamic datasets, without compromising performance. Additionally, we introduce a novel post-retrieval method that applies query-focused recursive abstractive processing to substantially improve context quality. Our method overcomes the limitations of other approaches by functioning as a black-box post-retrieval layer compatible with any retrieval algorithm. Both algorithms are validated through extensive experiments on real-world datasets, demonstrating their effectiveness in handling dynamic data and improving retrieval performance.

Reward Models (RMs) play a crucial role in aligning LLMs with human preferences, enhancing their performance by ranking outputs during inference or iterative training. However, the degree to which an RM generalizes to new tasks is often not known a priori (e.g. some RMs may excel at scoring creative writing vs. math reasoning). Therefore, using only one fixed RM while training LLMs can be suboptimal. Moreover, optimizing LLMs with multiple RMs simultaneously can be prohibitively computationally-intensive and challenging due to conflicting signals from different RMs, potentially degrading performance. To address these challenges, we introduce LASeR (Learning to Adaptively Select Rewards), which iteratively trains LLMs using multiple RMs, selecting and utilizing the most well-suited RM for each instance to rank outputs and generate preference data, framed as a multi-armed bandit problem. Our results on commonsense and math reasoning tasks demonstrate that LASeR can boost iterative LLM optimization by optimizing for multiple RMs, improving the absolute average accuracy of Llama-3-8B over three datasets by 2.67% over training with ensemble RM scores while also showing superior training efficiency (e.g., a 2x speedup). Moreover, on WildChat, a benchmark of instruction-following prompts, we find that using Llama-3-8B LASeR leads to a 71.45% AlpacaEval win rate over sequentially optimizing multiple RMs. Extending to long-context generation tasks, we find that on Llama-3-8B, LASeR achieves an average improvement of 2.64 F1 and 2.42 F1 on single- and multi-document QA over random RM selection when used with best-of-n sampling. LASeR is robust to noisy rewards and generalizes to multiple settings. Finally, LASeR's RM selection changes depending on the underlying task or instance and we verify the presence of conflicting preferences from multiple RMs that can be mitigated using LASeR.

Understanding visual semantics embedded in consecutive characters is a crucial capability for both large language models (LLMs) and multi-modal large language models (MLLMs). This type of artifact possesses the unique characteristic that identical information can be readily formulated in both texts and images, making them a significant proxy for analyzing modern LLMs' and MLLMs' capabilities in modality-agnostic vision understanding. In this work, we select ASCII art as a representative artifact, where the lines and brightness used to depict each concept are rendered by characters, and we frame the problem as an ASCII art recognition task. We benchmark model performance on this task by constructing an evaluation dataset with an elaborate categorization tree and also collect a training set to elicit the models' visual perception ability. Through a comprehensive analysis of dozens of models, results reveal that although humans can achieve nearly 100% accuracy, the state-of-the-art LLMs and MLLMs lag far behind. Models are capable of recognizing concepts depicted in the ASCII arts given only text inputs indicated by over 60% accuracy for some concepts, but most of them achieves merely around 30% accuracy when averaged across all categories. When provided with images as inputs, GPT-4o gets 82.68%, outperforming the strongest open-source MLLM by 21.95%. Although models favor different kinds of ASCII art depending on the modality provided, none of the MLLMs successfully benefit when both modalities are supplied simultaneously. Moreover, supervised fine-tuning helps improve models' accuracy especially when provided with the image modality, but also highlights the need for better training techniques to enhance the information fusion among modalities. While conventional wisdom suggests that texts primarily function as carriers of linguistic information and images as conveyors of visual information, real-world scenarios often involve the integration of multiple information formats.

The practical use of text-to-image generation has evolved from simple, monolithic models to complex workflows that combine multiple specialized components. While workflow-based approaches can lead to improved image quality, crafting effective workflows requires significant expertise, owing to the large number of available components, their complex inter-dependence, and their dependence on the generation prompt. Here, we introduce the novel task of prompt-adaptive workflow generation, where the goal is to automatically tailor a workflow to each user prompt. We propose two LLM-based approaches to tackle this task: a tuning-based method that learns from user-preference data, and a training-free method that uses the LLM to select existing flows. Both approaches lead to improved image quality when compared to monolithic models or generic, prompt-independent workflows. Our work shows that prompt-dependent flow prediction offers a new pathway to improving text-to-image generation quality, complementing existing research directions in the field.

Reward models are key in reinforcement learning from human feedback (RLHF) systems, aligning the model behavior with human preferences. Particularly in the math domain, there have been plenty of studies using reward models to align policies for improving reasoning capabilities. Recently, as the importance of reward models has been emphasized, RewardBench is proposed to understand their behavior. However, we figure out that the math subset of RewardBench has different representations between chosen and rejected completions, and relies on a single comparison, which may lead to unreliable results as it only see an isolated case. Therefore, it fails to accurately present the robustness of reward models, leading to a misunderstanding of its performance and potentially resulting in reward hacking. In this work, we introduce a new design for reliable evaluation of reward models, and to validate this, we construct RewardMATH, a benchmark that effectively represents the robustness of reward models in mathematical reasoning tasks. We demonstrate that the scores on RewardMATH strongly correlate with the results of optimized policy and effectively estimate reward overoptimization, whereas the existing benchmark shows almost no correlation. The results underscore the potential of our design to enhance the reliability of evaluation, and represent the robustness of reward model. We make our code and data publicly available.

Batch prompting is a common technique in large language models (LLMs) used to process multiple inputs simultaneously, aiming to improve computational efficiency. However, as batch sizes increase, performance degradation often occurs due to the model's difficulty in handling lengthy context inputs. Existing methods that attempt to mitigate these issues rely solely on batch data arrangement and majority voting rather than improving the design of the batch prompt itself. In this paper, we address these limitations by proposing "Auto-Demo Prompting," a novel approach that leverages the question-output pairs from earlier questions within a batch as demonstrations for subsequent answer inference. We provide a formal theoretical analysis of how Auto-Demo Prompting functions within the autoregressive generation process of LLMs, illustrating how it utilizes prior outputs to optimize the model's internal representations. Experimental results across five NLP tasks demonstrate its effectiveness in mitigating performance degradation and occasionally outperforming single prompts. Furthermore, it opens new avenues for applying few-shot learning techniques, such as demonstration selection, within batch prompting, making it a robust solution for real-world applications. Large language models (LLMs), such as GPT (Brown et al., 2020), and PaLM (Chowdhery et al., 2023), have demonstrated an extraordinary ability to perform in-context learning (ICL), where they utilize provided examples or contextual information to adapt and solve a wide range of downstream tasks. This capability enables LLMs to generalize from few-shot or even zero-shot examples without requiring task-specific fine-tuning, significantly enhancing their versatility across diverse applications (Song et al., 2023). The success of ICL in these models highlights their potential as powerful tools for natural language processing and as adaptable frameworks for learning in dynamic, dataconstrained environments, offering broader implications for machine learning and AI research.

The rapid development and deployment of Generative AI in social settings raise important questions about how to optimally personalize them for users while maintaining accuracy and realism. Based on a Facebook public post-comment dataset, this study evaluates the ability of Llama 3.0 (70B) to predict the semantic tones across different combinations of a commenter's and poster's gender, age, and friendship closeness and to replicate these differences in LLM-generated comments. The study consists of two parts: Part I assesses differences in semantic tones across social relationship categories, and Part II examines the similarity between comments generated by Llama 3.0 (70B) and human comments from Part I given public Facebook posts as input. Part I results show that including social relationship information improves the ability of a model to predict the semantic tone of human comments. However, Part II results show that even without including social context information in the prompt, LLM-generated comments and human comments are equally sensitive to social context, suggesting that LLMs can comprehend semantics from the original post alone. When we include all social relationship information in the prompt, the similarity between human comments and LLM-generated comments decreases. This inconsistency may occur because LLMs did not include social context information as part of their training data. Together these results demonstrate the ability of LLMs to comprehend semantics from the original post and respond similarly to human comments, but also highlights their limitations in generalizing personalized comments through prompting alone.

Large language models (LLMs) have demonstrated limitations in handling combinatorial optimization problems involving long-range reasoning, partially due to causal hallucinations and huge search space. As for causal hallucinations, i.e., the inconsistency between reasoning and corresponding state transition, this paper introduces the Causal Relationship Enhancement (CRE) mechanism combining cause-effect interventions and the Individual Treatment Effect (ITE) to guarantee the solid causal rightness between each step of reasoning and state transition. As for the long causal range and huge search space limiting the performances of existing models featuring single-direction search, a Dual-End Searching (DES) approach is proposed to seek solutions by simultaneously starting from both the initial and goal states on the causal probability tree. By integrating CRE and DES (CreDes), our model has realized simultaneous multi-step reasoning, circumventing the inefficiencies from cascading multiple one-step reasoning like the Chain-of-Thought (CoT). Experiments demonstrate that CreDes significantly outperforms existing State-Of-The-Art (SOTA) solutions in long-range reasoning tasks in terms of both accuracy and time efficiency.

Large language models (LLMs) have been shown to exhibit emergent abilities in some downstream tasks, where performance seems to stagnate at first and then improve sharply and unpredictably with scale beyond a threshold. By dividing questions in the datasets according to difficulty level by average performance, we observe U-shaped scaling for hard questions, and inverted-U scaling followed by steady improvement for easy questions. Moreover, the emergence threshold roughly coincides with the point at which performance on easy questions reverts from inverse scaling to standard scaling. Capitalizing on the observable though opposing scaling trend on easy and hard questions, we propose a simple yet effective pipeline, called Slice-and-Sandwich, to predict both the emergence threshold and model performance beyond the threshold.

Large language models have demonstrated significant potential as the next-generation information access engines. However, their reliability is hindered by issues of hallucination and generating non-factual content. This is particularly problematic in long-form responses, where assessing and ensuring factual accuracy is complex. In this paper, we address this gap by proposing FactAlign, a novel alignment framework designed to enhance the factuality of LLMs' long-form responses while maintaining their helpfulness. We introduce fKTO, a fine-grained, sentence-level alignment algorithm that extends the Kahneman-Tversky Optimization (KTO) alignment method. Leveraging recent advances in automatic factuality evaluation, FactAlign utilizes fine-grained factuality assessments to guide the alignment process. Our experiments on open-domain prompts and information-seeking questions demonstrate that FactAlign significantly improves the factual accuracy of LLM responses while also improving their helpfulness. Further analyses identify that FactAlign is capable of training LLMs to provide more information without losing factual precision, thus improving the factual F1 score. Our source code, datasets, and trained models are publicly available at https://github.com/MiuLab/FactAlign