We introduce WorkBench: a benchmark dataset for evaluating agents' ability to execute tasks in a workplace setting. WorkBench contains a sandbox environment with five databases, 26 tools, and 690 tasks. These tasks represent common business activities, such as sending emails and scheduling meetings. The tasks in WorkBench are challenging as they require planning, tool selection, and often multiple actions. If a task has been successfully executed, one (or more) of the database values may change. The correct outcome for each task is unique and unambiguous, which allows for robust, automated evaluation. We call this key contribution outcome-centric evaluation. We evaluate five existing ReAct agents on WorkBench, finding they successfully complete as few as 3% of tasks (Llama2-70B), and just 43% for the best-performing (GPT-4). We further find that agents' errors can result in the wrong action being taken, such as an email being sent to the wrong person. WorkBench reveals weaknesses in agents' ability to undertake common business activities, raising questions about their use in high-stakes workplace settings. WorkBench is publicly available as a free resource at https://github.com/olly-styles/WorkBench.

This paper introduces TVB-HKSL-News, a new Hong Kong sign language (HKSL) dataset collected from a TV news program over a period of 7 months. The dataset is collected to enrich resources for HKSL and support research in large-vocabulary continuous sign language recognition (SLR) and translation (SLT). It consists of 16.07 hours of sign videos of two signers with a vocabulary of 6,515 glosses (for SLR) and 2,850 Chinese characters or 18K Chinese words (for SLT). One signer has 11.66 hours of sign videos and the other has 4.41 hours. One objective in building the dataset is to support the investigation of how well large-vocabulary continuous sign language recognition/translation can be done for a single signer given a (relatively) large amount of his/her training data, which could potentially lead to the development of new modeling methods. Besides, most parts of the data collection pipeline are automated with little human intervention; we believe that our collection method can be scaled up to collect more sign language data easily for SLT in the future for any sign languages if such sign-interpreted videos are available. We also run a SOTA SLR/SLT model on the dataset and get a baseline SLR word error rate of 34.08% and a baseline SLT BLEU-4 score of 23.58 for benchmarking future research on the dataset.

One-on-one tutoring is widely acknowledged as an effective instructional method, conditioned on qualified tutors. However, the high demand for qualified tutors remains a challenge, often necessitating the training of novice tutors (i.e., trainees) to ensure effective tutoring. Research suggests that providing timely explanatory feedback can facilitate the training process for trainees. However, it presents challenges due to the time-consuming nature of assessing trainee performance by human experts. Inspired by the recent advancements of large language models (LLMs), our study employed the GPT-4 model to build an explanatory feedback system. This system identifies trainees' responses in binary form (i.e., correct/incorrect) and automatically provides template-based feedback with responses appropriately rephrased by the GPT-4 model. We conducted our study on 410 responses from trainees across three training lessons: Giving Effective Praise, Reacting to Errors, and Determining What Students Know. Our findings indicate that: 1) using a few-shot approach, the GPT-4 model effectively identifies correct/incorrect trainees' responses from three training lessons with an average F1 score of 0.84 and an AUC score of 0.85; and 2) using the few-shot approach, the GPT-4 model adeptly rephrases incorrect trainees' responses into desired responses, achieving performance comparable to that of human experts.

As Large Language Models (LLMs) have become more advanced, they have outpaced our abilities to accurately evaluate their quality. Not only is finding data to adequately probe particular model properties difficult, but evaluating the correctness of a model's freeform generation alone is a challenge. To address this, many evaluations now rely on using LLMs themselves as judges to score the quality of outputs from other LLMs. Evaluations most commonly use a single large model like GPT4. While this method has grown in popularity, it is costly, has been shown to introduce intramodel bias, and in this work, we find that very large models are often unnecessary. We propose instead to evaluate models using a Panel of LLm evaluators (PoLL). Across three distinct judge settings and spanning six different datasets, we find that using a PoLL composed of a larger number of smaller models outperforms a single large judge, exhibits less intra-model bias due to its composition of disjoint model families, and does so while being over seven times less expensive.

Graph neural networks (GNN) have emerged as a popular tool for modelling functional magnetic resonance imaging (fMRI) datasets. Many recent studies have reported significant improvements in disorder classification performance via more sophisticated GNN designs and highlighted salient features that could be potential biomarkers of the disorder. In this review, we provide an overview of how GNN and model explainability techniques have been applied on fMRI datasets for disorder prediction tasks, with a particular emphasis on the robustness of biomarkers produced for neurodegenerative diseases and neuropsychiatric disorders. We found that while most studies have performant models, salient features highlighted in these studies vary greatly across studies on the same disorder and little has been done to evaluate their robustness. To address these issues, we suggest establishing new standards that are based on objective evaluation metrics to determine the robustness of these potential biomarkers. We further highlight gaps in the existing literature and put together a prediction-attribution-evaluation framework that could set the foundations for future research on improving the robustness of potential biomarkers discovered via GNNs.

Astronomers and data scientists working with a boost in computer power on loan from Google have discovered 27,500 new asteroids, some perilously close to Earth. Their collaboration hopes to speed up the development of'a comprehensive map of the solar system' needed for'planetary defense,' according to one Harvard astrophysicist, Matthew Holman, who helped develop the asteroid-hunting software. Nearly 1.7 billion points of light, documented in 412,000 infrared images from the US National Optical-Infrared Astronomy Research Laboratory (NOIRLab) archives, were scanned by the project's novel'killer asteroid'-hunting algorithm. 'This is super important,' as one former NASA astronaut leading the project put it. 'This is the key to protecting the Earth from being hit by asteroids: knowing where all these are.'

Real-world misinformation can be partially correct and even factual but misleading. It undermines public trust in science and democracy, particularly on social media, where it can spread rapidly. High-quality and timely correction of misinformation that identifies and explains its (in)accuracies has been shown to effectively reduce false beliefs. Despite the wide acceptance of manual correction, it is difficult to be timely and scalable, a concern as technologies like large language models (LLMs) make misinformation easier to produce. LLMs also have versatile capabilities that could accelerate misinformation correction-however, they struggle due to a lack of recent information, a tendency to produce false content, and limitations in addressing multimodal information. We propose MUSE, an LLM augmented with access to and credibility evaluation of up-to-date information. By retrieving evidence as refutations or contexts, MUSE identifies and explains (in)accuracies in a piece of content-not presupposed to be misinformation-with references. It also describes images and conducts multimodal searches to verify and correct multimodal content. Fact-checking experts evaluate responses to social media content that are not presupposed to be (non-)misinformation but broadly include incorrect, partially correct, and correct posts, that may or may not be misleading. We propose and evaluate 13 dimensions of misinformation correction quality, ranging from the accuracy of identifications and factuality of explanations to the relevance and credibility of references. The results demonstrate MUSE's ability to promptly write high-quality responses to potential misinformation on social media-overall, MUSE outperforms GPT-4 by 37% and even high-quality responses from laypeople by 29%. This work reveals LLMs' potential to help combat real-world misinformation effectively and efficiently.

The implications of backdoor attacks on English-centric large language models (LLMs) have been widely examined - such attacks can be achieved by embedding malicious behaviors during training and activated under specific conditions that trigger malicious outputs. However, the impact of backdoor attacks on multilingual models remains under-explored. Our research focuses on cross-lingual backdoor attacks against multilingual LLMs, particularly investigating how poisoning the instruction-tuning data in one or two languages can affect the outputs in languages whose instruction-tuning data was not poisoned. Despite its simplicity, our empirical analysis reveals that our method exhibits remarkable efficacy in models like mT5, BLOOM, and GPT-3.5-turbo, with high attack success rates, surpassing 95% in several languages across various scenarios. Alarmingly, our findings also indicate that larger models show increased susceptibility to transferable cross-lingual backdoor attacks, which also applies to LLMs predominantly pre-trained on English data, such as Llama2, Llama3, and Gemma. Moreover, our experiments show that triggers can still work even after paraphrasing, and the backdoor mechanism proves highly effective in cross-lingual response settings across 25 languages, achieving an average attack success rate of 50%. Our study aims to highlight the vulnerabilities and significant security risks present in current multilingual LLMs, underscoring the emergent need for targeted security measures.

This paper explores the intricate relationship between capitalism, racial injustice, and artificial intelligence (AI), arguing that AI acts as a contemporary vehicle for age-old forms of exploitation. By linking historical patterns of racial and economic oppression with current AI practices, this study illustrates how modern technology perpetuates and deepens societal inequalities. It specifically examines how AI is implicated in the exploitation of marginalized communities through underpaid labor in the gig economy, the perpetuation of biases in algorithmic decision-making, and the reinforcement of systemic barriers that prevent these groups from benefiting equitably from technological advances. Furthermore, the paper discusses the role of AI in extending and intensifying the social, economic, and psychological burdens faced by these communities, highlighting the problematic use of AI in surveillance, law enforcement, and mental health contexts. The analysis concludes with a call for transformative changes in how AI is developed and deployed. Advocating for a reevaluation of the values driving AI innovation, the paper promotes an approach that integrates social justice and equity into the core of technological design and policy. This shift is crucial for ensuring that AI serves as a tool for societal improvement, fostering empowerment and healing rather than deepening existing divides.

Large Language Models are transforming NLP for a variety of tasks. However, how LLMs perform NLP tasks for low-resource languages (LRLs) is less explored. In line with the goals of the AmericasNLP workshop, we focus on 12 LRLs from Brazil, 2 LRLs from Africa and 2 high-resource languages (HRLs) (e.g., English and Brazilian Portuguese). Our results indicate that the LLMs perform worse for the part of speech (POS) labeling of LRLs in comparison to HRLs. We explain the reasons behind this failure and provide an error analysis through examples observed in our data set.