The advent of Large Language Models (LLMs) such as Generative Pre-trained Transformers (GPT-4) mark a transformative era in Natural Language Generation (NLG). These models demonstrate the ability to generate coherent text that closely resembles human-authored content. They are easily accessible and have become invaluable tools in handling various text-based tasks, such as data annotation, report generation, and question answering. In this paper, we investigate GPT-4’s ability to discern between data it has annotated and data annotated by humans, specifically within the context of tweets in the medical domain. Through experimental analysis, we observe GPT-4 outperform other state-of-the-art models. The dataset used in this study was provided by the SMM4H (Social Media Mining for Health Research and Applications) shared task. Our model achieved an accuracy of 0.51, securing a second rank in the shared task.

Determining the duration and length of a news event’s impact on a company’s performance remains elusive for financial analysts. The complexity arises from the fact that the effects of these news articles are influenced by various extraneous factors and can change over time. As a result, in this work, we investigate our ability to predict 1) the duration (length) of a news event’s impact, and 2) level of impact on companies. The datasets used in this study are provided as part of the Multi-Lingual ESG Impact Duration Inference (ML-ESG-3) shared task. To handle the data scarcity, we explored data augmentation techniques to augment our training data. To address each of the research objectives stated above, we employ an ensemble approach combining transformer model, a variant of Convolutional Neural Networks (CNNs), specifically the KimCNN model and contextual embeddings. The model’s performance is assessed across a multilingual dataset encompassing English, French, Japanese, and Korean news articles. For the first task of determining impact duration, our model ranked in first, fifth, seventh, and eight place for Japanese, French, Korean and English texts respectively (with respective macro F1 scores of 0.256, 0.458, 0.552, 0.441). For the second task of assessing impact level, our model ranked in sixth, and eight place for French and English texts, respectively (with respective macro F1 scores of 0.488 and 0.550).

The rapid advancement of Generative Artificial Intelligence (AI), particularly Large Language Models (LLMs), has led to their widespread adoption for various natural language processing (NLP) tasks. One crucial domain ripe for innovation is the Technology-Assisted Review (TAR) process in Electronic discovery (eDiscovery). Traditionally, TAR involves manual review and classification of documents for relevance over large document collections for litigations and investigations. This process is aided by machine learning and NLP tools which require extensive training and fine-tuning. In this paper, we explore the application of LLMs to TAR, specifically for predictive coding. We experiment with out-of-the-box prompting and fine-tuning of LLMs using parameter-efficient techniques. We conduct experiments using open LLMs and compare them to commercially-licensed ones. Our experiments demonstrate that open LLMs lag behind commercially-licensed models in relevance classification using out-of-the-box prompting. However, topic-specific instruction tuning of open LLMs not only improve their effectiveness but can often outperform their commercially-licensed counterparts in performance evaluations. Additionally, we conduct a user study to gauge the preferences of our eDiscovery Subject Matter Specialists (SMS) regarding human-authored versus model-generated reasoning. We demonstrate that instruction-tuned open LLMs can generate high quality reasonings that are comparable to commercial LLMs.

Recent Large Language Models (LLMs) have demonstrated remarkable capabilities in generating text that closely resembles human writing across wide range of styles and genres. However, such capabilities are prone to potential abuse, such as fake news generation, spam email creation, and misuse in academic assignments. Hence, it is essential to build automated approaches capable of distinguishing between artificially generated text and human-authored text. In this paper, we propose a simple yet efficient solution to this problem by ensembling predictions from multiple constituent LLMs. Compared to previous state-of-the-art approaches, which are perplexity-based or uses ensembles with a large number of LLMs, our condensed ensembling approach uses only two constituent LLMs to achieve comparable performance. Experiments conducted on four benchmark datasets for generative text classification show performance improvements in the range of 0.5 to 100% compared to previous state-of-the-art approaches. We also study that the influence the training data from individual LLMs have on model performance. We found that substituting commercially-restrictive Generative Pre-trained Transformer (GPT) data with data generated from other open language models such as Falcon, Large Language Model Meta AI (LLaMA2), and Mosaic Pretrained Transformers (MPT) is a feasible alternative when developing generative text detectors. Furthermore, to demonstrate zero-shot generalization, we experimented with an English essays dataset, and results suggest that our ensembling approach can handle new data effectively.