Sam Lin, Wenyue Hua, Zhenting Wang, Mingyu Jin, Lizhou Fan, Yongfeng Zhang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Complex reasoning ability is one of the most important features of Large Language Models (LLMs). Numerous benchmarks have been established to assess the reasoning abilities of LLMs. However, they are inadequate in offering a rigorous evaluation and prone to the risk of overfitting, as these publicly accessible and static benchmarks allow models to potentially tailor their responses to specific benchmark metrics, thereby inflating their performance. Addressing these limitations, we introduce a new benchmark NPHardEval. It contains a broad spectrum of 900 algorithmic questions belonging up to the NP-Hard complexity class, offering a rigorous measure of the reasoning ability of LLMs utilizing computational complexity. Moreover, this benchmark is designed with a dynamic update mechanism, where the datapoints are refreshed on a monthly basis. Such regular updates play a crucial role in mitigating the risk of LLMs overfitting to the benchmark, promoting a more accurate and reliable assessment of their reasoning capabilities. The benchmark dataset and code of NPHardEval are available at https://github.com/casmlab/NPHardEval.

Large language models (LLMs), such as OpenAI's Generative Pre-trained Transformer (GPT), are a class of language models that have demonstrated outstanding performance across a range of natural language processing (NLP) tasks. LLMs have become a highly sought-after research area because of their ability to generate human-like language and their potential to revolutionize science and technology. In this study, we conduct bibliometric and discourse analyses of scholarly literature on LLMs. Synthesizing over 5,000 publications, this article serves as a roadmap for researchers, practitioners, and policymakers to navigate the current landscape of LLMs research. We present the research trends from 2017 to early 2023, identifying patterns in research paradigms and collaborations. We start with analyzing the core algorithm developments and NLP tasks that are fundamental in LLMs research. We then investigate the applications of LLMs in various fields and domains, including medicine, engineering, social science, and humanities. Our review also reveals the dynamic, fast-paced evolution of LLMs research. Overall, this article offers valuable insights into the current state, impact, and potential of LLMs research and its applications.

Harmful textual content is pervasive on social media, poisoning online communities and negatively impacting participation. A common approach to this issue is developing detection models that rely on human annotations. However, the tasks required to build such models expose annotators to harmful and offensive content and may require significant time and cost to complete. Generative AI models have the potential to understand and detect harmful textual content. We used ChatGPT to investigate this potential and compared its performance with MTurker annotations for three frequently discussed concepts related to harmful textual content on social media: Hateful, Offensive, and Toxic (HOT). We designed five prompts to interact with ChatGPT and conducted four experiments eliciting HOT classifications. Our results show that ChatGPT can achieve an accuracy of approximately 80% when compared to MTurker annotations. Specifically, the model displays a more consistent classification for non-HOT comments than HOT comments compared to human annotations. Our findings also suggest that ChatGPT classifications align with the provided HOT definitions. However, ChatGPT classifies “hateful” and “offensive” as subsets of “toxic.” Moreover, the choice of prompts used to interact with ChatGPT impacts its performance. Based on these insights, our study provides several meaningful implications for employing ChatGPT to detect HOT content, particularly regarding the reliability and consistency of its performance, its understanding and reasoning of the HOT concept, and the impact of prompts on its performance. Overall, our study provides guidance on the potential of using generative AI models for moderating large volumes of user-generated textual content on social media.

Addressing increasing calls to surface hidden and counter-narratives from within archival collections, this paper reports on a study that provides proof-of-concept of automatic methods that could be used on archived social media collections. Using a test collection of 3,457,434 unique tweets relating to COVID-19, China and Chinese people, it sought to identify instances of Hate Speech as well as hard-to-pinpoint trends in anti-Chinese racist sentiment. The study, part of a larger archival research effort investigating automatic methods for appraisal and description of very large digital archival collections, used a Three-step Social Media Similarity (TSMS) mapping method that aggregates hashtag mapping, TF-IDF Similarity Selection, and Emotion Similarity Calculation on the test collection. Compared to using a purely lexicon-based method to identify and analyze controversial speech, this method successfully expanded the amount of controversial contents detected from 21,050 tweets to 212,605, and the detection rate from 0.6% to 6.1%. We argue that the TSMS method could be similarly applied by archives in automatically identifying, analyzing, describing other controversial content on social media and in other rapidly evolving and complex contexts in order to increase public awareness and facilitate public policy responses.