What counts is not what people actually know, but what people believe that everyone knows, and which is thus taken as a common background. The rising popularity of AI systems over the past few months is remarkable. Large language models (LLMs), once confined to little more than curiosities within AI labs or the topic of research papers, have now been deployed by corporations, making them available for public consumption in the form of various chat-like applications.

Until now, the only free and easy part had been the data. Widely used services like the nonprofit Common Crawl charge Google, Meta, OpenAI and others nothing to use its service, which crawls the internet in search of troves of online text and archives the information for others to download. To assemble the vast quantities of natural language and specialized information needed to train large AI systems, tech companies have combined those archives with online data sets, accessing information made available for research purposes, and increasingly straying from information clearly in the public domain.

Data Synthesis is a promising way to train a small model with very little labeled data. One approach for data synthesis is to leverage the rich knowledge from large language models to synthesize pseudo training examples for small models, making it possible to achieve both data and compute efficiency at the same time. However, a key challenge in data synthesis is that the synthesized dataset often suffers from a large distributional discrepancy from the real task data distribution. Thus, in this paper, we propose Synthesis Step by Step (S3), a data synthesis framework that shrinks this distribution gap by iteratively extrapolating the errors made Figure 1: Training and testing accuracy of DistilBert by a small model trained on the synthesized with ZeroGen (Ye et al., 2022b) on the IMDb dataset dataset on a small real-world validation dataset with 200k training datapoints. Also shown are the training using a large language model. Extensive experiments and testing accuracy of the model trained on Gold-on multiple NLP tasks show that our Data. We can see here that ZeroGen's training accuracy approach improves the performance of a small quickly reaches nearly 100%, but testing accuracy remains model by reducing the gap between the synthetic low.

This study investigated how perceived workload, satisfaction, performance expectancy, and risk-benefit perception influenced users' trust in Chat Generative Pre-Trained Transformer (ChatGPT). We aimed to understand the nuances of user engagement and provide insights to improve future design and adoption strategies for similar technologies. A semi-structured, web-based survey was conducted among adults in the United States who actively use ChatGPT at least once a month. The survey was conducted from 22nd February 2023 through 24th March 2023. We used structural equation modeling to understand the relationships among the constructs of perceived workload, satisfaction, performance expectancy, risk-benefit, and trust. The analysis of 607 survey responses revealed a significant negative relationship between perceived workload and user satisfaction, a negative but insignificant relationship between perceived workload and trust, and a positive relationship between user satisfaction and trust. Trust was also found to increase with performance expectancy. In contrast, the relationship between the benefit-to-risk ratio of using ChatGPT and trust was insignificant. The findings underscore the importance of ensuring user-friendly design and functionality in AI-based applications to reduce workload and enhance user satisfaction, thereby increasing user trust. Future research should further explore the relationship between the benefit-to-risk ratio and trust in the context of AI chatbots.

There is a lack of research into capabilities of recent LLMs to generate convincing text in languages other than English and into performance of detectors of machine-generated text in multilingual settings. This is also reflected in the available benchmarks which lack authentic texts in languages other than English and predominantly cover older generators. To fill this gap, we introduce MULTITuDE, a novel benchmarking dataset for multilingual machine-generated text detection comprising of 74,081 authentic and machine-generated texts in 11 languages (ar, ca, cs, de, en, es, nl, pt, ru, uk, and zh) generated by 8 multilingual LLMs. Using this benchmark, we compare the performance of zero-shot (statistical and black-box) and fine-tuned detectors. Considering the multilinguality, we evaluate 1) how these detectors generalize to unseen languages (linguistically similar as well as dissimilar) and unseen LLMs and 2) whether the detectors improve their performance when trained on multiple languages.

Financial cybercrime prevention is an increasing issue with many organisations and governments. As deep learning models have progressed to identify illicit activity on various financial and social networks, the explainability behind the model decisions has been lacklustre with the investigative analyst at the heart of any deep learning platform. In our paper, we present a state-of-the-art, novel multimodal proactive approach to addressing XAI in financial cybercrime detection. We leverage a triad of deep learning models designed to distill essential representations from transaction sequencing, subgraph connectivity, and narrative generation to significantly streamline the analyst's investigative process. Our narrative generation proposal leverages LLM to ingest transaction details and output contextual narrative for an analyst to understand a transaction and its metadata much further.

Large-scale deployment of generative AI tools often depends on costly API calls to a Large Language Model (LLM) to fulfil user queries. To curtail the frequency of these calls, one can employ a smaller language model -- a student -- which is continuously trained on the responses of the LLM. This student gradually gains proficiency in independently handling an increasing number of user requests, a process we term neural caching. The crucial element in neural caching is a policy that decides which requests should be processed by the student alone and which should be redirected to the LLM, subsequently aiding the student's learning. In this study, we focus on classification tasks, and we consider a range of classic active learning-based selection criteria as the policy. Our experiments suggest that Margin Sampling and Query by Committee bring consistent benefits across tasks and budgets.

Aligning large language models (LLMs) to human values has become increasingly important as it enables sophisticated steering of LLMs. However, it requires significant human demonstrations and feedback or distillation from proprietary LLMs such as ChatGPT. In this work, we propose a novel alignment learning framework with synthetic feedback not dependent on extensive human annotations and proprietary LLMs. First, we perform reward modeling (RM) with synthetic feedback by contrasting responses from vanilla LLMs with various sizes and prompts. Then, we use the RM to simulate high-quality demonstrations to train a supervised policy and further optimize the model with reinforcement learning. Our resulting model, Aligned Language Model with Synthetic Training dataset (ALMoST), outperforms recent open-sourced models, which are trained on the outputs of InstructGPT or human-annotated demonstrations, in alignment benchmarks. In human evaluation, our model is preferred to Alpaca and Dolly-v2, 55.0% and 58.5% of the time, respectively. Further analyses demonstrate the efficacy and importance of synthetic feedback in our framework. The code is available at https://github.com/naver-ai/almost

Social cognitive theory explains how people learn and acquire knowledge through observing others. Recent years have witnessed the rapid development of large language models (LLMs), which suggests their potential significance as agents in the society. LLMs, as AI agents, can observe external information, which shapes their cognition and behaviors. However, the extent to which external information influences LLMs' cognition and behaviors remains unclear. This study investigates how external statements and opinions influence LLMs' thoughts and behaviors from a social cognitive perspective. Three experiments were conducted to explore the effects of external information on LLMs' memories, opinions, and social media behavioral decisions. Sociocognitive factors, including source authority, social identity, and social role, were analyzed to investigate their moderating effects. Results showed that external information can significantly shape LLMs' memories, opinions, and behaviors, with these changes mirroring human social cognitive patterns such as authority bias, in-group bias, emotional positivity, and emotion contagion. This underscores the challenges in developing safe and unbiased LLMs, and emphasizes the importance of understanding the susceptibility of LLMs to external influences.

In task-oriented dialogue scenarios, cross-domain zero-shot slot filling plays a vital role in leveraging source domain knowledge to learn a model with high generalization ability in unknown target domain where annotated data is unavailable. However, the existing state-of-the-art zero-shot slot filling methods have limited generalization ability in target domain, they only show effective knowledge transfer on seen slots and perform poorly on unseen slots. To alleviate this issue, we present a novel Hierarchical Contrastive Learning Framework (HiCL) for zero-shot slot filling. Specifically, we propose a coarse- to fine-grained contrastive learning based on Gaussian-distributed embedding to learn the generalized deep semantic relations between utterance-tokens, by optimizing inter- and intra-token distribution distance. This encourages HiCL to generalize to the slot types unseen at training phase. Furthermore, we present a new iterative label set semantics inference method to unbiasedly and separately evaluate the performance of unseen slot types which entangled with their counterparts (i.e., seen slot types) in the previous zero-shot slot filling evaluation methods. The extensive empirical experiments on four datasets demonstrate that the proposed method achieves comparable or even better performance than the current state-of-the-art zero-shot slot filling approaches.