Although large language models (LLMs) are increasingly capable, these capabilities are unevenly distributed: they excel at formal linguistic tasks, such as producing fluent, grammatical text, but struggle more with functional linguistic tasks like reasoning and consistent fact retrieval. Inspired by neuroscience, recent work suggests that to succeed on both formal and functional linguistic tasks, LLMs should use different mechanisms for each; such localization could either be built-in or emerge spontaneously through training. In this paper, we ask: do current models, with fast-improving functional linguistic abilities, exhibit distinct localization of formal and functional linguistic mechanisms? We answer this by finding and comparing the "circuits", or minimal computational subgraphs, responsible for various formal and functional tasks. Comparing 5 LLMs across 10 distinct tasks, we find that while there is indeed little overlap between circuits for formal and functional tasks, there is also little overlap between formal linguistic tasks, as exists in the human brain. Thus, a single formal linguistic network, unified and distinct from functional task circuits, remains elusive. However, in terms of cross-task faithfulness - the ability of one circuit to solve another's task - we observe a separation between formal and functional mechanisms, suggesting that shared mechanisms between formal tasks may exist.

Evaluating creative text such as human-written stories using language models has always been a challenging task -- owing to the subjectivity of multi-annotator ratings. To mimic the thinking process of humans, chain of thought (CoT) generates free-text explanations that help guide a model's predictions and Self-Consistency (SC) marginalizes predictions over multiple generated explanations. In this study, we discover that the widely-used self-consistency reasoning methods cause suboptimal results due to an objective mismatch between generating 'fluent-looking' explanations vs. actually leading to a good rating prediction for an aspect of a story. To overcome this challenge, we propose $\textbf{C}$hain-$\textbf{o}$f-$\textbf{Ke}$ywords (CoKe), that generates a sequence of keywords $\textit{before}$ generating a free-text rationale, that guide the rating prediction of our evaluation language model. Then, we generate a diverse set of such keywords, and aggregate the scores corresponding to these generations. On the StoryER dataset, CoKe based on our small fine-tuned evaluation models not only reach human-level performance and significantly outperform GPT-4 with a 2x boost in correlation with human annotators, but also requires drastically less number of parameters.

The recruitment process is undergoing a significant transformation with the increasing use of machine learning and natural language processing techniques. While previous studies have focused on automating candidate selection, the role of multiple vacancies in this process remains understudied. This paper addresses this gap by proposing a novel pipeline that leverages Large Language Models and graph similarity measures to suggest ideal candidates for specific job openings. Our approach represents candidate profiles as multimodal embeddings, enabling the capture of nuanced relationships between job requirements and candidate attributes. The proposed approach has significant implications for the recruitment industry, enabling companies to streamline their hiring processes and identify top talent more efficiently. Our work contributes to the growing body of research on the application of machine learning in human resources, highlighting the potential of LLMs and graph-based methods in revolutionizing the recruitment landscape.

Caching has the potential to be of significant benefit for accessing large language models (LLMs) due to their high latencies which typically range from a small number of seconds to well over a minute. Furthermore, many LLMs charge money for queries; caching thus has a clear monetary benefit. This paper presents a new caching system for improving user experiences with LLMs. In addition to reducing both latencies and monetary costs for accessing LLMs, our system also provides important features that go beyond the performance benefits typically associated with caches. A key feature we provide is generative caching, wherein multiple cached responses can be synthesized to provide answers to queries which have never been seen before. Our generative caches function as repositories of valuable information which can be mined and analyzed. We also improve upon past semantic caching techniques by tailoring the caching algorithms to optimally balance cost and latency reduction with the quality of responses provided. Performance tests indicate that our caches are considerably faster than GPTcache.

Assertion status detection is a critical yet often overlooked component of clinical NLP, essential for accurately attributing extracted medical facts. Past studies have narrowly focused on negation detection, leading to underperforming commercial solutions such as AWS Medical Comprehend, Azure AI Text Analytics, and GPT-4o due to their limited domain adaptation. To address this gap, we developed state-of-the-art assertion detection models, including fine-tuned LLMs, transformer-based classifiers, few-shot classifiers, and deep learning (DL) approaches. We evaluated these models against cloud-based commercial API solutions, the legacy rule-based NegEx approach, and GPT-4o. Our fine-tuned LLM achieves the highest overall accuracy (0.962), outperforming GPT-4o (0.901) and commercial APIs by a notable margin, particularly excelling in Present (+4.2%), Absent (+8.4%), and Hypothetical (+23.4%) assertions. Our DL-based models surpass commercial solutions in Conditional (+5.3%) and Associated-with-Someone-Else (+10.1%) categories, while the few-shot classifier offers a lightweight yet highly competitive alternative (0.929), making it ideal for resource-constrained environments. Integrated within Spark NLP, our models consistently outperform black-box commercial solutions while enabling scalable inference and seamless integration with medical NER, Relation Extraction, and Terminology Resolution. These results reinforce the importance of domain-adapted, transparent, and customizable clinical NLP solutions over general-purpose LLMs and proprietary APIs.

Indoor navigation presents unique challenges due to complex layouts, lack of GPS signals, and accessibility concerns. Existing solutions often struggle with real-time adaptability and user-specific needs. In this work, we explore the potential of a Large Language Model (LLM), i.e., ChatGPT, to generate natural, context-aware navigation instructions from indoor map images. We design and evaluate test cases across different real-world environments, analyzing the effectiveness of LLMs in interpreting spatial layouts, handling user constraints, and planning efficient routes. Our findings demonstrate the potential of LLMs for supporting personalized indoor navigation, with an average of 52% correct indications and a maximum of 62%. The results do not appear to depend on the complexity of the layout or the complexity of the expected path, but rather on the number of points of interest and the abundance of visual information, which negatively affect the performance.

Fully comprehending scientific papers by machines reflects a high level of Artificial General Intelligence, requiring the ability to reason across fragmented and heterogeneous sources of information, presenting a complex and practically significant challenge. While Vision-Language Models (VLMs) have made remarkable strides in various tasks, particularly those involving reasoning with evidence source from single image or text page, their ability to use cross-source information for reasoning remains an open problem. This work presents MMCR, a high-difficulty benchmark designed to evaluate VLMs' capacity for reasoning with cross-source information from scientific papers. The benchmark comprises 276 high-quality questions, meticulously annotated by humans across 7 subjects and 10 task types. Experiments with 18 VLMs demonstrate that cross-source reasoning presents a substantial challenge for existing models. Notably, even the top-performing model, GPT-4o, achieved only 48.55% overall accuracy, with only 20% accuracy in multi-table comprehension tasks, while the second-best model, Qwen2.5-VL-72B, reached 39.86% overall accuracy. Furthermore, we investigated the impact of the Chain-of-Thought (CoT) technique on cross-source reasoning and observed a detrimental effect on small models, whereas larger models demonstrated substantially enhanced performance. These results highlight the pressing need to develop VLMs capable of effectively utilizing cross-source information for reasoning.

Fine-tuning large language models (LLMs) on downstream tasks can inadvertently erode their safety alignment, even for benign fine-tuning datasets. It achieves this by selectively merging fine-tuned and safety-aligned model layers only when those deviate from safe behavior, measured by a cosine similarity criterion. We evaluate SafeMERGE against other fine-tuning-and post-fine-tuning-stage approaches for Llama-2-7B-Chat and Qwen-2-7B-Instruct models on GSM8K and PubMedQA tasks while exploring different merging strategies. We find that SafeMERGE consistently reduces harmful outputs compared to other baselines without significantly sacrificing performance, sometimes even enhancing it. The results suggest that our selective, subspace-guided, and per-layer merging method provides an effective safeguard against the inadvertent loss of safety in fine-tuned LLMs while outperforming simpler post-fine-tuning-stage defenses. Large language models (LLMs) have demonstrated remarkable capabilities in text generation and understanding while becoming increasingly accessible to AI practitioners. Safety tuning is critical to ensure that advanced LLMs align with human values and security policies, making them safe for deployment (Ouyang et al., 2022; Bai et al., 2022; Chiang et al., 2023; Zhang et al., 2024). However, the safety alignment of current LLMs has been shown to be vulnerable (Wei et al., 2023; Huang et al., 2024e; Yang et al., 2023; Zeng et al., 2024; Zhan et al., 2024; Qi et al., 2023; 2024a).

The widespread deployment of general-purpose AI (GPAI) systems introduces significant new risks. Yet the infrastructure, practices, and norms for reporting flaws in GPAI systems remain seriously underdeveloped, lagging far behind more established fields like software security. Based on a collaboration between experts from the fields of software security, machine learning, law, social science, and policy, we identify key gaps in the evaluation and reporting of flaws in GPAI systems. We call for three interventions to advance system safety. First, we propose using standardized AI flaw reports and rules of engagement for researchers in order to ease the process of submitting, reproducing, and triaging flaws in GPAI systems. Second, we propose GPAI system providers adopt broadly-scoped flaw disclosure programs, borrowing from bug bounties, with legal safe harbors to protect researchers. Third, we advocate for the development of improved infrastructure to coordinate distribution of flaw reports across the many stakeholders who may be impacted. These interventions are increasingly urgent, as evidenced by the prevalence of jailbreaks and other flaws that can transfer across different providers' GPAI systems. By promoting robust reporting and coordination in the AI ecosystem, these proposals could significantly improve the safety, security, and accountability of GPAI systems.

Knowledge distillation is a technique used to train a small student network using the output generated by a large teacher network, and has many empirical advantages~\citep{Hinton2015DistillingTK}. While the standard one-shot approach to distillation only uses the output of the final teacher network, recent work~\citep{panigrahi2024progressive} has shown that using intermediate checkpoints from the teacher's training process as an implicit ``curriculum'' for progressive distillation can significantly speed up training. However, such schemes require storing these checkpoints, and often require careful selection of the intermediate checkpoints to train on, which can be impractical for large-scale training. In this paper, we show that a curriculum can be \emph{extracted} from just the fully trained teacher network, and that this extracted curriculum can give similar efficiency benefits to those of progressive distillation. Our extraction scheme is natural; we use a random projection of the hidden representations of the teacher network to progressively train the student network, before training using the output of the full network. We show that our scheme significantly outperforms one-shot distillation and achieves a performance similar to that of progressive distillation for learning sparse parities with two-layer networks, and provide theoretical guarantees for this setting. Additionally, we show that our method outperforms one-shot distillation even when using transformer-based architectures, both for sparse-parity learning, and language modeling tasks.