Stance detection holds great potential for enhancing the quality of online political discussions, as it has shown to be useful for summarizing discussions, detecting misinformation, and evaluating opinion distributions. Usually, transformer-based models are used directly for stance detection, which require large amounts of data. However, the broad range of debate questions in online political discussion creates a variety of possible scenarios that the model is faced with and thus makes data acquisition for model training difficult. In this work, we show how to leverage LLM-generated synthetic data to train and improve stance detection agents for online political discussions:(i) We generate synthetic data for specific debate questions by prompting a Mistral-7B model and show that fine-tuning with the generated synthetic data can substantially improve the performance of stance detection. (ii) We examine the impact of combining synthetic data with the most informative samples from an unlabelled dataset. First, we use the synthetic data to select the most informative samples, second, we combine both these samples and the synthetic data for fine-tuning. This approach reduces labelling effort and consistently surpasses the performance of the baseline model that is trained with fully labeled data. Overall, we show in comprehensive experiments that LLM-generated data greatly improves stance detection performance for online political discussions.

Pairwise evaluation using large language models (LLMs) is widely used for evaluating natural language generation (NLG) tasks. However, the reliability of LLMs is often compromised by biases, such as favoring verbosity and authoritative tone. In the study, we focus on the comparison of two LLM-based evaluation approaches, pointwise and pairwise. Our findings demonstrate that pointwise evaluators exhibit more robustness against undesirable preferences. Further analysis reveals that pairwise evaluators can accurately identify the shortcomings of low-quality outputs even when their judgment is incorrect. These results indicate that LLMs are more severely influenced by their bias in a pairwise evaluation setup. To mitigate this, we propose a hybrid method that integrates pointwise reasoning into pairwise evaluation. Experimental results show that our method enhances the robustness of pairwise evaluators against adversarial samples while preserving accuracy on normal samples.

This paper's primary goal is to provoke thoughtful discussion about the relationship between bias and fundamental properties of large language models. We do this by seeking to convince the reader that harmful biases are an inevitable consequence arising from the design of any large language model as LLMs are currently formulated. To the extent that this is true, it suggests that the problem of harmful bias cannot be properly addressed without a serious reconsideration of AI driven by LLMs, going back to the foundational assumptions underlying their design.

The rapid development of large language models (LLMs) has shown promising practical results. However, their low interpretability often leads to errors in unforeseen circumstances, limiting their utility. Many works have focused on creating comprehensive evaluation systems, but previous benchmarks have primarily assessed problem-solving abilities while neglecting the response's uncertainty, which may result in unreliability. Recent methods for measuring LLM reliability are resource-intensive and unable to test black-box models. To address this, we propose UBENCH, a comprehensive benchmark for evaluating LLM reliability. UBENCH includes 3,978 multiple-choice questions covering knowledge, language, understanding, and reasoning abilities. Experimental results show that UBENCH has achieved state-of-the-art performance, while its single-sampling method significantly saves computational resources compared to baseline methods that require multiple samplings. Additionally, based on UBENCH, we evaluate the reliability of 15 popular LLMs, finding GLM4 to be the most outstanding, closely followed by GPT-4. We also explore the impact of Chain-of-Thought prompts, role-playing prompts, option order, and temperature on LLM reliability, analyzing the varying effects on different LLMs.

Despite their remarkable successes, state-of-the-art large language models (LLMs), including vision-and-language models (VLMs) and unimodal language models (ULMs), fail to understand precise semantics. For example, semantically equivalent sentences expressed using different lexical compositions elicit diverging representations. The degree of this divergence and its impact on encoded semantics is not very well understood. In this paper, we introduce the SUGARCREPE++ dataset to analyze the sensitivity of VLMs and ULMs to lexical and semantic alterations. Each sample in SUGARCREPE++ dataset consists of an image and a corresponding triplet of captions: a pair of semantically equivalent but lexically different positive captions and one hard negative caption. This poses a 3-way semantic (in)equivalence problem to the language models. We comprehensively evaluate VLMs and ULMs that differ in architecture, pre-training objectives and datasets to benchmark the performance of SUGARCREPE++ dataset. Experimental results highlight the difficulties of VLMs in distinguishing between lexical and semantic variations, particularly in object attributes and spatial relations. Although VLMs with larger pre-training datasets, model sizes, and multiple pre-training objectives achieve better performance on SUGARCREPE++, there is a significant opportunity for improvement. We show that all the models which achieve better performance on compositionality datasets need not perform equally well on SUGARCREPE++, signifying that compositionality alone may not be sufficient for understanding semantic and lexical alterations. Given the importance of the property that the SUGARCREPE++ dataset targets, it serves as a new challenge to the vision-and-language community.

Safe learning is central to AI-enabled robots where a single failure may lead to catastrophic results. Barrier-based method is one of the dominant approaches for safe robot learning. However, this method is not scalable, hard to train, and tends to generate unstable signals under noisy inputs that are challenging to be deployed for robots. To address these challenges, we propose a novel Attention BarrierNet (ABNet) that is scalable to build larger foundational safe models in an incremental manner. Each head of BarrierNet in the ABNet could learn safe robot control policies from different features and focus on specific part of the observation. In this way, we do not need to one-shotly construct a large model for complex tasks, which significantly facilitates the training of the model while ensuring its stable output. Most importantly, we can still formally prove the safety guarantees of the ABNet. We demonstrate the strength of ABNet in 2D robot obstacle avoidance, safe robot manipulation, and vision-based end-to-end autonomous driving, with results showing much better robustness and guarantees over existing models.

A dog whistle is a form of coded communication that carries a secondary meaning to specific audiences and is often weaponized for racial and socioeconomic discrimination. Dog whistling historically originated from United States politics, but in recent years has taken root in social media as a means of evading hate speech detection systems and maintaining plausible deniability. In this paper, we present an approach for word-sense disambiguation of dog whistles from standard speech using Large Language Models (LLMs), and leverage this technique to create a dataset of 16,550 high-confidence coded examples of dog whistles used in formal and informal communication. Silent Signals is the largest dataset of disambiguated dog whistle usage, created for applications in hate speech detection, neology, and political science. The dataset can be found at https://huggingface.co/datasets/SALT-NLP/silent_signals.

We present SciRIFF (Scientific Resource for Instruction-Following and Finetuning), a dataset of 137K instruction-following demonstrations for 54 tasks covering five essential scientific literature understanding capabilities: information extraction, summarization, question answering, claim verification, and classification. SciRIFF demonstrations are notable for their long input contexts, detailed task specifications, and complex structured outputs. While instruction-following resources are available in specific domains such as clinical medicine and chemistry, SciRIFF is the first dataset focused on extracting and synthesizing information from research literature across a wide range of scientific fields. To demonstrate the utility of SciRIFF, we develop a sample-efficient strategy to adapt a general instruction-following model for science by performing additional finetuning on a mix of general-domain and SciRIFF demonstrations. In evaluations on nine held-out scientific tasks, our model -- called SciTulu -- improves over a strong LLM baseline by 28.1% and 6.5% at the 7B and 70B scales respectively, while maintaining general instruction-following performance within 2% of the baseline. We are optimistic that SciRIFF will facilitate the development and evaluation of LLMs to help researchers navigate the ever-growing body of scientific literature. We release our dataset, model checkpoints, and data processing and evaluation code to enable further research.

The integration of Large Language Models (LLMs) into healthcare applications offers promising advancements in medical diagnostics, treatment recommendations, and patient care. However, the susceptibility of LLMs to adversarial attacks poses a significant threat, potentially leading to harmful outcomes in delicate medical contexts. This study investigates the vulnerability of LLMs to two types of adversarial attacks in three medical tasks. Utilizing real-world patient data, we demonstrate that both open-source and proprietary LLMs are susceptible to manipulation across multiple tasks. This research further reveals that domain-specific tasks demand more adversarial data in model fine-tuning than general domain tasks for effective attack execution, especially for more capable models. We discover that while integrating adversarial data does not markedly degrade overall model performance on medical benchmarks, it does lead to noticeable shifts in fine-tuned model weights, suggesting a potential pathway for detecting and countering model attacks. This research highlights the urgent need for robust security measures and the development of defensive mechanisms to safeguard LLMs in medical applications, to ensure their safe and effective deployment in healthcare settings.

In recent years, predictive maintenance (PMx) has gained prominence for its potential to enhance efficiency, automation, accuracy, and cost-effectiveness while reducing human involvement. Importantly, PMx has evolved in tandem with digital advancements, such as Big Data and the Internet of Things (IOT). These technological strides have enabled Artificial Intelligence (AI) to revolutionize PMx processes, with increasing capacities for real-time automation of monitoring, analysis, and prediction tasks. However, PMx still faces challenges such as poor explainability and sample inefficiency in data-driven methods and high complexity in physics-based models, hindering broader adoption. This paper posits that Digital Twins (DTs) can be integrated into PMx to overcome these challenges, paving the way for more automated PMx applications across various stakeholders. Despite their potential, current DTs have not fully matured to bridge existing gaps. Our paper provides a comprehensive roadmap for DT evolution, addressing current limitations to foster large-scale automated PMx progression. We structure our approach in three stages: First, we reference prior work where we identified and defined the Information Requirements (IRs) and Functional Requirements (FRs) for PMx, forming the blueprint for a unified framework. Second, we conduct a literature review to assess current DT applications integrating these IRs and FRs, revealing standardized DT models and tools that support automated PMx. Lastly, we highlight gaps in current DT implementations, particularly those IRs and FRs not fully supported, and outline the necessary components for a comprehensive, automated PMx system. Our paper concludes with research directions aimed at seamlessly integrating DTs into the PMx paradigm to achieve this ambitious vision.