Large Language Models (LLMs) enable a future in which certain types of legal documents may be generated automatically. This has a great potential to streamline legal processes, lower the cost of legal services, and dramatically increase access to justice. While many researchers focus their efforts on proposing and evaluating LLM-based applications supporting tasks in the legal domain, there is a notable lack of investigations into how legal professionals perceive content if they believe it has been generated by an LLM. Yet, this is a critical point as over-reliance or unfounded skepticism may influence whether such documents bring about appropriate legal consequences. This study is the necessary analysis in the context of the ongoing transition towards mature generative AI systems. Specifically, we examined whether the perception of legal documents' by lawyers (n=75) varies based on their assumed origin (human-crafted vs AI-generated). The participants evaluated the documents focusing on their correctness and language quality. Our analysis revealed a clear preference for documents perceived as crafted by a human over those believed to be generated by AI. At the same time, most of the participants are expecting the future in which documents will be generated automatically. These findings could be leveraged by legal practitioners, policy makers and legislators to implement and adopt legal document generation technology responsibly, and to fuel the necessary discussions into how legal processes should be updated to reflect the recent technological developments.

Due to the large number of parameters, the inference phase of Large Language Models (LLMs) is resource-intensive. However, not all requests posed to LLMs are equally difficult to handle. Through analysis, we show that for some tasks, LLMs can achieve results comparable to the final output at some intermediate layers. That is, not all layers of LLMs are necessary during inference. If we can predict at which layer the inferred results match the final results (produced by evaluating all layers), we could significantly reduce the inference cost. To this end, we propose a simple yet effective algorithm named AdaInfer to adaptively terminate the inference process for an input instance. AdaInfer relies on easily obtainable statistical features and classic classifiers like SVM. Experiments on well-known LLMs like the Llama2 series and OPT, show that AdaInfer can achieve an average of 17.8% pruning ratio, and up to 43% on sentiment tasks, with nearly no performance drop (<1%). Because AdaInfer does not alter LLM parameters, the LLMs incorporated with AdaInfer maintain generalizability across tasks.

In multi-session dialog system, it is essential to continuously update the memory as the session progresses. Simply accumulating memory can make it difficult to focus on the content of the conversation for inference due to the limited input sentence size. Therefore, efficient and accurate conversation model that is capable of managing memory to reflect the conversation history continuously is necessary. This paper presents a conversation model that efficiently manages memory as sessions progress and incorporates this into the model to reflect the conversation history accurately with 3 methodologies: SFT, DPO and DPO with SFT model. Our model using DPO algorithm shows an improvement about 0.0591 of BERTScore in memory accuracy, and the rate of responses reflecting the memory increased as well. Also, response generation performance enhanced about 4.292 in fluency, 3.935 in coherence, and 2.896 in consistency. This paper describes a training method that yields better performance than models with more than twice the parameter size, even when the model size is smaller. Thus, our model demonstrates efficiency not only in terms of accuracy but also in resource utilization.

In recent years, Natural Language Processing (NLP) has played a significant role in various Artificial Intelligence (AI) applications such as chatbots, text generation, and language translation. The emergence of large language models (LLMs) has greatly improved the performance of these applications, showing astonishing results in language understanding and generation. However, they still show some disadvantages, such as hallucinations and lack of domain-specific knowledge, that affect their performance in real-world tasks. These issues can be effectively mitigated by incorporating knowledge graphs (KGs), which organise information in structured formats that capture relationships between entities in a versatile and interpretable fashion. Likewise, the construction and validation of KGs present challenges that LLMs can help resolve. The complementary relationship between LLMs and KGs has led to a trend that combines these technologies to achieve trustworthy results. This work collected 28 papers outlining methods for KG-powered LLMs, LLM-based KGs, and LLM-KG hybrid approaches. We systematically analysed and compared these approaches to provide a comprehensive overview highlighting key trends, innovative techniques, and common challenges. This synthesis will benefit researchers new to the field and those seeking to deepen their understanding of how KGs and LLMs can be effectively combined to enhance AI applications capabilities.

Reinforcement learning with human feedback (RLHF), as a widely adopted approach in current large language model pipelines, is \textit{bottlenecked by the size of human preference data}. While traditional methods rely on offline preference dataset constructions, recent approaches have shifted towards online settings, where a learner uses a small amount of labeled seed data and a large pool of unlabeled prompts to iteratively construct new preference data through self-generated responses and high-quality reward/preference feedback. However, most current online algorithms still focus on preference labeling during policy model updating with given feedback oracles, which incurs significant expert query costs. \textit{We are the first to explore cost-effective proxy reward oracles construction strategies for further labeling preferences or rewards with extremely limited labeled data and expert query budgets}. Our approach introduces two key innovations: (1) on-policy query to avoid OOD and imbalance issues in seed data, and (2) active learning to select the most informative data for preference queries. Using these methods, we train a evaluation model with minimal expert-labeled data, which then effectively labels nine times more preference pairs for further RLHF training. For instance, our model using Direct Preference Optimization (DPO) gains around over 1% average improvement on AlpacaEval2, MMLU-5shot and MMLU-0shot, with only 1.7K query cost. Our methodology is orthogonal to other direct expert query-based strategies and therefore might be integrated with them to further reduce query costs.

Large language models (LLMs) have greatly impacted the natural language processing (NLP) field, particularly for the English language. These models have demonstrated capabilities in understanding and generating human-like text. The success of language models largely depends on the availability of high-quality instruction datasets, which consist of detailed task descriptions and corresponding responses that are essential for training the models to address a variety of prompts accurately. However, the availability and quality of these resources vary by language. While models perform well in English, they often need help with languages like Arabic, due to the lack of datasets for fine-tuning Arabic-specific tasks. To address this issue, we introduce InstAr-500k, a new Arabic instruction dataset created by generating and collecting content that covers several domains and instruction types. We assess this dataset by fine-tuning an open-source Gemma-7B model on several downstream tasks to improve its functionality. Based on multiple evaluations, our fine-tuned model achieves excellent performance on several Arabic NLP benchmarks. These outcomes emphasize the effectiveness of our dataset in elevating the capabilities of language models for Arabic. Our instruction dataset bridges the performance gap between English and Arabic language models by providing resources that amplify Arabic NLP development. Building on this foundation, we developed a model, GemmAr-7B-V1, specifically tuned to excel at a wide range of Arabic NLP tasks.

The study of Twitter as a means for analyzing social phenomena has gained interest in recent years due to the availability of large amounts of data in a relatively spontaneous environment. Within opinion-mining tasks, emotion detection is specially relevant, as it allows for the identification of people's subjective responses to different social events in a more granular way than traditional sentiment analysis based on polarity. In the particular case of political events, the analysis of emotions in social networks can provide valuable information on the perception of candidates, proposals, and other important aspects of the public debate. In spite of this importance, there are few studies on emotion detection in Spanish and, to the best of our knowledge, few resources are public for opinion mining in Colombian Spanish, highlighting the need for generating resources addressing the specific cultural characteristics of this variety. In this work, we present a small corpus of tweets in Spanish related to the 2022 Colombian presidential elections, manually labeled with emotions using a fine-grained taxonomy. We perform classification experiments using supervised state-of-the-art models (BERT models) and compare them with GPT-3.5 in few-shot learning settings. We make our dataset and code publicly available for research purposes.

Reinforcement learning from human feedback (RLHF) provides a principled framework for aligning AI systems with human preference data. For various reasons, e.g., personal bias, context ambiguity, lack of training, etc, human annotators may give incorrect or inconsistent preference labels. To tackle this challenge, we propose a robust RLHF approach -- $R^3M$, which models the potentially corrupted preference label as sparse outliers. Accordingly, we formulate the robust reward learning as an $\ell_1$-regularized maximum likelihood estimation problem. Computationally, we develop an efficient alternating optimization algorithm, which only incurs negligible computational overhead compared with the standard RLHF approach. Theoretically, we prove that under proper regularity conditions, $R^3M$ can consistently learn the underlying reward and identify outliers, provided that the number of outlier labels scales sublinearly with the preference sample size. Furthermore, we remark that $R^3M$ is versatile and can be extended to various preference optimization methods, including direct preference optimization (DPO). Our experiments on robotic control and natural language generation with large language models (LLMs) show that $R^3M$ improves robustness of the reward against several types of perturbations to the preference data.

While humans naturally develop theory of mind (ToM), the capability to understand other people's mental states and beliefs, state-of-the-art large language models (LLMs) underperform on simple ToM benchmarks. We posit that we can extend our understanding of LLMs' ToM abilities by evaluating key human ToM precursors -- perception inference and perception-to-belief inference -- in LLMs. We introduce two datasets, Percept-ToMi and Percept-FANToM, to evaluate these precursory inferences for ToM in LLMs by annotating characters' perceptions on ToMi and FANToM, respectively. Our evaluation of eight state-of-the-art LLMs reveals that the models generally perform well in perception inference while exhibiting limited capability in perception-to-belief inference (e.g., lack of inhibitory control). Based on these results, we present PercepToM, a novel ToM method leveraging LLMs' strong perception inference capability while supplementing their limited perception-to-belief inference. Experimental results demonstrate that PercepToM significantly enhances LLM's performance, especially in false belief scenarios.

Large language models (LLMs) have transformed the way we interact with cyber technologies. In this paper, we study the possibility of connecting LLM with wireless sensor networks (WSN). A successful design will not only extend LLM's knowledge landscape to the physical world but also revolutionize human interaction with WSN. To the end, we present ChatTracer, an LLM-powered real-time Bluetooth device tracking system. ChatTracer comprises three key components: an array of Bluetooth sniffing nodes, a database, and a fine-tuned LLM. ChatTracer was designed based on our experimental observation that commercial Apple/Android devices always broadcast hundreds of BLE packets per minute even in their idle status. Its novelties lie in two aspects: i) a reliable and efficient BLE packet grouping algorithm; and ii) an LLM fine-tuning strategy that combines both supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF). We have built a prototype of ChatTracer with four sniffing nodes. Experimental results show that ChatTracer not only outperforms existing localization approaches, but also provides an intelligent interface for user interaction.