Public release of the weights of pretrained foundation models, otherwise known as downloadable access [Solaiman, 2023], enables fine-tuning without the prohibitive expense of pretraining. Our work argues that increasingly accessible fine-tuning of downloadable models may increase hazards. First, we highlight research to improve the accessibility of fine-tuning. We split our discussion into research that A) reduces the computational cost of fine-tuning and B) improves the ability to share that cost across more actors. Second, we argue that increasingly accessible finetuning methods may increase hazard through facilitating malicious use and making oversight of models with potentially dangerous capabilities more difficult. Third, we discuss potential mitigatory measures, as well as benefits of more accessible fine-tuning. Given substantial remaining uncertainty about hazards, we conclude by emphasizing the urgent need for the development of mitigations.

We evaluate the ability of large language models (LLMs) to infer causal relations from natural language. Compared to traditional natural language processing and deep learning techniques, LLMs show competitive performance in a benchmark of pairwise relations without needing (explicit) training samples. This motivates us to extend our approach to extrapolating causal graphs through iterated pairwise queries. We perform a preliminary analysis on a benchmark of biomedical abstracts with ground-truth causal graphs validated by experts. The results are promising and support the adoption of LLMs for such a crucial step in causal inference, especially in medical domains, where the amount of scientific text to analyse might be huge, and the causal statements are often implicit.

In today's digital world, Generative Artificial Intelligence (GenAI) such as Large Language Models (LLMs) is becoming increasingly prevalent, extending its reach across diverse applications. This surge in adoption has sparked a significant increase in demand for data-centric GenAI models, highlighting the necessity for robust data communication infrastructures. Central to this need are message brokers, which serve as essential channels for data transfer within various system components. This survey aims to delve into a comprehensive analysis of traditional and modern message brokers, offering a comparative study of prevalent platforms. Our study considers numerous criteria including, but not limited to, open-source availability, integrated monitoring tools, message prioritization mechanisms, capabilities for parallel processing, reliability, distribution and clustering functionalities, authentication processes, data persistence strategies, fault tolerance, and scalability. Furthermore, we explore the intrinsic constraints that the design and operation of each message broker might impose, recognizing that these limitations are crucial in understanding their real-world applicability. We then leverage these insights to propose a sophisticated message broker framework -- one designed with the adaptability and robustness necessary to meet the evolving requisites of GenAI applications. Finally, this study examines the enhancement of message broker mechanisms specifically for GenAI contexts, emphasizing the criticality of developing a versatile message broker framework. Such a framework would be poised for quick adaptation, catering to the dynamic and growing demands of GenAI in the foreseeable future. Through this dual-pronged approach, we intend to contribute a foundational compendium that can guide future innovations and infrastructural advancements in the realm of GenAI data communication.

As humans, we consistently engage in interactions with our peers and receive feedback in the form of natural language. This language feedback allows us to reflect on our actions, maintain appropriate behavior, and rectify our errors. The question arises naturally: can we use language feedback to align large language models (LLMs)? In contrast to previous research that aligns LLMs with reward or preference data, we present the first systematic exploration of alignment through the lens of language feedback (i.e., judgment). We commence with an in-depth investigation of potential methods that can be adapted for aligning LLMs with judgments, revealing that these methods are unable to fully capitalize on the judgments. To facilitate more effective utilization of judgments, we propose a novel framework, Contrastive Unlikelihood Training (CUT), that allows for fine-grained inappropriate content detection and correction based on judgments. Our offline alignment results show that, with merely 1317 off-the-shelf judgment data, CUT (LLaMA2-13b) can beat the 175B DaVinci003 and surpass the best baseline by 52.34 points on AlpacaEval. The online alignment results demonstrate that CUT can align LLMs (LLaMA2-chat-13b) in an iterative fashion using model-specific judgment data, with a steady performance improvement from 81.09 to 91.36 points on AlpacaEval. Our analysis further suggests that judgments exhibit greater potential than rewards for LLM alignment and warrant future research.

Identifying speakers of quotations in narratives is an important task in literary analysis, with challenging scenarios including the out-of-domain inference for unseen speakers, and non-explicit cases where there are no speaker mentions in surrounding context. In this work, we propose a simple and effective approach SIG, a generation-based method that verbalizes the task and quotation input based on designed prompt templates, which also enables easy integration of other auxiliary tasks that further bolster the speaker identification performance. The prediction can either come from direct generation by the model, or be determined by the highest generation probability of each speaker candidate. Based on our approach design, SIG supports out-of-domain evaluation, and achieves open-world classification paradigm that is able to accept any forms of candidate input. We perform both cross-domain evaluation and in-domain evaluation on PDNC, the largest dataset of this task, where empirical results suggest that SIG outperforms previous baselines of complicated designs, as well as the zero-shot ChatGPT, especially excelling at those hard non-explicit scenarios by up to 17% improvement. Additional experiments on another dataset WP further corroborate the efficacy of SIG.

Prompt learning has demonstrated impressive efficacy in the fine-tuning of multimodal large models to a wide range of downstream tasks. Nonetheless, applying existing prompt learning methods for the diagnosis of neurological disorder still suffers from two issues: (i) existing methods typically treat all patches equally, despite the fact that only a small number of patches in neuroimaging are relevant to the disease, and (ii) they ignore the structural information inherent in the brain connection network which is crucial for understanding and diagnosing neurological disorders. To tackle these issues, we introduce a novel prompt learning model by learning graph prompts during the fine-tuning process of multimodal large models for diagnosing neurological disorders. Specifically, we first leverage GPT-4 to obtain relevant disease concepts and compute semantic similarity between these concepts and all patches. Secondly, we reduce the weight of irrelevant patches according to the semantic similarity between each patch and disease-related concepts. Moreover, we construct a graph among tokens based on these concepts and employ a graph convolutional network layer to extract the structural information of the graph, which is used to prompt the pre-trained multimodal large models for diagnosing neurological disorders. Extensive experiments demonstrate that our method achieves superior performance for neurological disorder diagnosis compared with state-of-the-art methods and validated by clinicians.

The global surge in AI applications is transforming industries, leading to displacement and complementation of existing jobs, while also giving rise to new employment opportunities. Human oversight of AI is an emerging task in which human workers interact with an AI model to improve its performance, safety, and compliance with normative principles. Data annotation, encompassing the labelling of images or annotating of texts, serves as a critical human oversight process, as the quality of a dataset directly influences the quality of AI models trained on it. Therefore, the efficiency of human oversight work stands as an important competitive advantage for AI developers. This paper delves into the foundational economics of human oversight, with a specific focus on the impact of norm design and monetary incentives on data quality and costs. An experimental study involving 307 data annotators examines six groups with varying task instructions (norms) and monetary incentives. Results reveal that annotators provided with clear rules exhibit higher accuracy rates, outperforming those with vague standards by 14%. Similarly, annotators receiving an additional monetary incentive perform significantly better, with the highest accuracy rate recorded in the group working with both clear rules and incentives (87.5% accuracy). However, both groups require more time to complete tasks, with a 31% increase in average task completion time compared to those working with standards and no incentives. These empirical findings underscore the trade-off between data quality and efficiency in data curation, shedding light on the nuanced impact of norm design and incentives on the economics of AI development. The paper contributes experimental insights to discussions on the economical, ethical, and legal considerations of AI technologies.

Large language models (LLMs) are increasingly being used in Metaverse environments to generate dynamic and realistic content and to control the behavior of non-player characters (NPCs). However, the cybersecurity concerns associated with LLMs have become increasingly prominent. Previous research has primarily focused on patching system vulnerabilities to enhance cybersecurity, but these approaches are not well-suited to the Metaverse, where the virtual space is more complex, LLMs are vulnerable, and ethical user interaction is critical. Moreover, the scope of cybersecurity in the Metaverse is expected to expand significantly. This paper proposes a method for enhancing cybersecurity through the simulation of user interaction with LLMs. Our goal is to educate users and strengthen their defense capabilities through exposure to a comprehensive simulation system. This system includes extensive Metaverse cybersecurity Q&A and attack simulation scenarios. By engaging with these, users will improve their ability to recognize and withstand risks. Additionally, to address the ethical implications of user input, we propose using LLMs as evaluators to assess user content across five dimensions. We further adapt the models through vocabulary expansion training to better understand personalized inputs and emoticons. We conduct experiments on multiple LLMs and find that our approach is effective.

Following the standard supervised fine-tuning (SFT) paradigm, in-context learning (ICL) has become an efficient approach propelled by the recent advancements in large language models (LLMs), yielding promising performance across various tasks in few-shot data setups. However, both paradigms are prone to suffer from the critical problem of overconfidence (i.e., miscalibration), especially in such limited data setups. In this work, we deliver an in-depth analysis of the behavior across different choices of learning methods from the perspective of both performance and calibration, as well as their interplay. Through extensive controlled experiments, we find that simultaneous gains for both task performance and calibration are difficult to achieve, and the problem of miscalibration exists across all learning methods in low-resource scenarios. To address this challenging trade-off between performance and calibration, we then investigate the potential of self-ensembling techniques applied at different modeling stages (e.g., variations of in-context examples or variations in prompts or different ensembling strategies). We justify the feasibility of self-ensembling on SFT in addition to ICL, to make the predictions more calibrated and have comparable or even better performance. Our work sheds light on which learning paradigm to choose and how to enhance both task performance and calibration of LLMs.

This paper studies how to configure powerful In-Context Demonstration (ICD) sequences for a Large Vision-Language Model (LVLM) to solve Vision-Language tasks through In-Context Learning (ICL). After observing that configuring an ICD sequence is a mirror process of composing a sentence, i.e., just as a sentence can be composed word by word via a Language Model, an ICD sequence can also be configured one by one. Consequently, we introduce an ICD Language Model (ICD-LM) specifically designed to generate effective ICD sequences. This involves creating a dataset of hand-crafted ICD sequences for various query samples and using it to train the ICD-LM. Our approach, diverging from traditional methods in NLP that select and order ICDs separately, enables to simultaneously learn how to select and order ICDs, enhancing the effect of the sequences. Moreover, during data construction, we use the LVLM intended for ICL implementation to validate the strength of each ICD sequence, resulting in a model-specific dataset and the ICD-LM trained by this dataset is also model-specific. We validate our methodology through experiments in Visual Question Answering and Image Captioning, confirming the viability of using a Language Model for ICD configuration. Our comprehensive ablation studies further explore the impact of various dataset construction and ICD-LM development settings on the outcomes. The code is given in https://github.com/ForJadeForest/ICD-LM.