Attention sink (AS) is a consistent pattern in transformer attention maps where certain tokens (often special tokens or positional anchors) disproportionately attract attention from other tokens. We show that in transformers, AS is not an architectural artifact, but it is the manifestation of a fundamental geometric principle: the establishment of reference frames that anchor representational spaces. We analyze several architectures and identify three distinct reference frame types, centralized, distributed, and bidirectional, that correlate with the attention sink phenomenon. We show that they emerge during the earliest stages of training as optimal solutions to the problem of establishing stable coordinate systems in high-dimensional spaces. We show the influence of architecture components, particularly position encoding implementations, on the specific type of reference frame. This perspective transforms our understanding of transformer attention mechanisms and provides insights for both architecture design and the relationship with AS.

With the rapid evolution of multimodal large language models, the capacity to deeply understand and interpret human intentions has emerged as a critical capability, which demands detailed and thoughtful reasoning. In recent studies, Reinforcement Learning (RL) has demonstrated potential in enhancing the reasoning capabilities of Large Language Models (LLMs). Nonetheless, the challenges associated with adapting RL to multimodal data and formats remain largely unaddressed. In this paper, we identify two issues in existing multimodal reasoning models: insufficient global context understanding and shortcut problems. Insufficient context understanding can happen when a model misinterprets multimodal context, resulting in incorrect answers. The shortcut problem occurs when the model overlooks crucial clues in multimodal inputs, directly addressing the query without considering the multimodal information. To tackle these issues, we emphasize the necessity for the model to reason with a clear understanding of the global context within multimodal inputs. This global context understanding can effectively prevent the model from overlooking key multimodal cues and ensure a thorough reasoning process. To ensure the accurate interpretation of multimodal context information, we implement a context reward judged by a large language model, alongside format and accuracy rewards. Additionally, to improve complex reasoning capability, we employ the LLM to assess the logical reward, determining whether the reasoning process successfully integrates multimodal information with logical methods. We also introduce a reasoning omni-modal benchmark, IntentBench, aimed at evaluating models in understanding complex human intentions and emotions. Our proposed method demonstrates advanced performance across multiple omni-modal benchmarks compared to other open-source omni-modal models.

Large language models (LLMs) have shown significant potential to change how we write, communicate, and create, leading to rapid adoption across society. This dissertation examines how individuals and institutions are adapting to and engaging with this emerging technology through three research directions. First, I demonstrate how the institutional adoption of AI detectors introduces systematic biases, particularly disadvantaging writers of non-dominant language varieties, highlighting critical equity concerns in AI governance. Second, I present novel population-level algorithmic approaches that measure the increasing adoption of LLMs across writing domains, revealing consistent patterns of AI-assisted content in academic peer reviews, scientific publications, consumer complaints, corporate communications, job postings, and international organization press releases. Finally, I investigate LLMs' capability to provide feedback on research manuscripts through a large-scale empirical analysis, offering insights into their potential to support researchers who face barriers in accessing timely manuscript feedback, particularly early-career researchers and those from under-resourced settings.

The general public often encounters complex texts but does not have the time or expertise to fully understand them, leading to the spread of misinformation. Automatic Text Simplification (ATS) helps make information more accessible, but its evaluation methods have not kept up with advances in text generation, especially with Large Language Models (LLMs). In particular, recent studies have shown that current ATS metrics do not correlate with the presence of errors. Manual inspections have further revealed a variety of errors, underscoring the need for a more nuanced evaluation framework, which is currently lacking. This resource paper addresses this gap by introducing a test collection for detecting and classifying errors in simplified texts. First, we propose a taxonomy of errors, with a formal focus on information distortion. Next, we introduce a parallel dataset of automatically simplified scientific texts. This dataset has been human-annotated with labels based on our proposed taxonomy. Finally, we analyze the quality of the dataset, and we study the performance of existing models to detect and classify errors from that taxonomy. These contributions give researchers the tools to better evaluate errors in ATS, develop more reliable models, and ultimately improve the quality of automatically simplified texts.

Purpose: We present a virtual model to optimize point of entry (POE) in lung biopsy planning systems. Our model allows to compute the quality of a biopsy sample taken from potential POE, taking into account the margin of error that arises from discrepancies between the orientation in the planning simulation and the actual orientation during the operation. Additionally, the study examines the impact of the characteristics of the lesion. Methods: The quality of the biopsy is given by a heatmap projected onto the skeleton of a patient-specific model of airways. The skeleton provides a 3D representation of airways structure, while the heatmap intensity represents the potential amount of tissue that it could be extracted from each POE. This amount of tissue is determined by the intersection of the lesion with a cone that represents the uncertainty area in the introduction of biopsy instruments. The cone, lesion, and skeleton are modelled as graphical objects that define a 3D scene of the intervention. Results: We have simulated different settings of the intervention scene from a single anatomy extracted from a CT scan and two lesions with regular and irregular shapes. The different scenarios are simulated by systematic rotation of each lesion placed at different distances from airways. Analysis of the heatmaps for the different settings show a strong impact of lesion orientation for irregular shape and the distance for both shapes. Conclusion: The proposed heatmaps help to visually assess the optimal POE and identify whether multiple optimal POEs exist in different zones of the bronchi. They also allow us to model the maximum allowable error in navigation systems and study which variables have the greatest influence on the success of the operation. Additionally, they help determine at what point this influence could potentially jeopardize the operation.

Many important scientific problems involve multivariate optimization coupled with slow and laborious experimental measurements. These complex, high-dimensional searches can be defined by non-convex optimization landscapes that resemble needle-in-a-haystack surfaces, leading to entrapment in local minima. Contextualizing optimizers with human domain knowledge is a powerful approach to guide searches to localized fruitful regions. However, this approach is susceptible to human confirmation bias and it is also challenging for domain experts to keep track of the rapidly expanding scientific literature. Here, we propose the use of Large Language Models (LLMs) for contextualizing Bayesian optimization (BO) via a hybrid optimization framework that intelligently and economically blends stochastic inference with domain knowledge-based insights from the LLM, which is used to suggest new, better-performing areas of the search space for exploration. Our method fosters user engagement by offering real-time commentary on the optimization progress, explaining the reasoning behind the search strategies. We validate the effectiveness of our approach on synthetic benchmarks with up to 15 independent variables and demonstrate the ability of LLMs to reason in four real-world experimental tasks where context-aware suggestions boost optimization performance substantially.

This paper introduces CaseSumm, a novel dataset for long-context summarization in the legal domain that addresses the need for longer and more complex datasets for summarization evaluation. We collect 25.6K U.S. Supreme Court (SCOTUS) opinions and their official summaries, known as "syllabuses." Our dataset is the largest open legal case summarization dataset, and is the first to include summaries of SCOTUS decisions dating back to 1815. We also present a comprehensive evaluation of LLM-generated summaries using both automatic metrics and expert human evaluation, revealing discrepancies between these assessment methods. Our evaluation shows Mistral 7b, a smaller open-source model, outperforms larger models on most automatic metrics and successfully generates syllabus-like summaries. In contrast, human expert annotators indicate that Mistral summaries contain hallucinations. The annotators consistently rank GPT-4 summaries as clearer and exhibiting greater sensitivity and specificity. Further, we find that LLM-based evaluations are not more correlated with human evaluations than traditional automatic metrics. Furthermore, our analysis identifies specific hallucinations in generated summaries, including precedent citation errors and misrepresentations of case facts. These findings demonstrate the limitations of current automatic evaluation methods for legal summarization and highlight the critical role of human evaluation in assessing summary quality, particularly in complex, high-stakes domains. CaseSumm is available at https://huggingface.co/datasets/ChicagoHAI/CaseSumm

Tool wear conditions impact the surface quality of the workpiece Tool wear is an inevitable phenomenon in the actual machining and its final geometric precision. In this research, we process. It leads to alterations in the cutting zone's process propose an efficient tool wear segmentation approach based on variables like the forces and temperatures exerted on both the tool Segment Anything Model, which integrates U-Net as an automated and workpiece. These conditions not only influence the rate of prompt generator to streamline the processes of tool wear tool wear but also affect the surface quality and geometric precision detection. Our evaluation covered three Point-of-Interest generation of the workpiece [1]. Therefore, tool wear is one of the methods and further investigated the effects of variations in key determinants of both tool costs and the quality of the finished training dataset sizes and U-Net training intensities on resultant workpiece, emphasizing the necessity for monitoring during the wear segmentation outcomes. The results consistently highlight machining process to ensure optimal outcomes [1, 2].

Research on dialogue constructiveness assessment focuses on (i) analysing conversational factors that influence individuals to take specific actions, win debates, change their perspectives or broaden their open-mindedness and (ii) predicting constructive outcomes following dialogues for such use cases. These objectives can be achieved by training either interpretable feature-based models (which often involve costly human annotations) or neural models such as pre-trained language models (which have empirically shown higher task accuracy but lack interpretability). We propose a novel LLM feature-based framework that combines the strengths of feature-based and neural approaches while mitigating their downsides, in assessing dialogue constructiveness. The framework first defines a set of dataset-independent and interpretable linguistic features, which can be extracted by both prompting an LLM and simple heuristics. Such features are then used to train LLM feature-based models. We apply this framework to three datasets of dialogue constructiveness and find that our LLM feature-based models significantly outperform standard feature-based models and neural models, and tend to learn more robust prediction rules instead of relying on superficial shortcuts (as seen with neural models). Further, we demonstrate that interpreting these LLM feature-based models can yield valuable insights into what makes a dialogue constructive.

We present an approach for estimating the fraction of text in a large corpus which is likely to be substantially modified or produced by a large language model (LLM). Our maximum likelihood model leverages expert-written and AI-generated reference texts to accurately and efficiently examine real-world LLM-use at the corpus level. We apply this approach to a case study of scientific peer review in AI conferences that took place after the release of ChatGPT: ICLR 2024, NeurIPS 2023, CoRL 2023 and EMNLP 2023. Our results suggest that between 6.5% and 16.9% of text submitted as peer reviews to these conferences could have been substantially modified by LLMs, i.e. beyond spell-checking or minor writing updates. The circumstances in which generated text occurs offer insight into user behavior: the estimated fraction of LLM-generated text is higher in reviews which report lower confidence, were submitted close to the deadline, and from reviewers who are less likely to respond to author rebuttals. We also observe corpus-level trends in generated text which may be too subtle to detect at the individual level, and discuss the implications of such trends on peer review. We call for future interdisciplinary work to examine how LLM use is changing our information and knowledge practices.