Compositionality is a fundamental aspect of vision-language understanding and is especially required for videos since they contain multiple entities (e.g. persons, actions, and scenes) interacting dynamically over time. Existing benchmarks focus primarily on perception capabilities. However, they do not study binding, the ability of a model to associate entities through appropriate relationships. To this end, we propose VELOCITI, a new benchmark building on complex movie clips and dense semantic role label annotations to test perception and binding in video language models (contrastive and Video-LLMs). Our perception-based tests require discriminating video-caption pairs that share similar entities, and the binding tests require models to associate the correct entity to a given situation while ignoring the different yet plausible entities that also appear in the same video. While current state-of-the-art models perform moderately well on perception tests, accuracy is near random when both entities are present in the same video, indicating that they fail at binding tests. Even the powerful Gemini 1.5 Flash has a substantial gap (16-28%) with respect to human accuracy in such binding tests.

Large language models (LLMs) inevitably memorize sensitive, copyrighted, and harmful knowledge from the training corpus; therefore, it is crucial to erase this knowledge from the models. Machine unlearning is a promising solution for efficiently removing specific knowledge by post hoc modifying models. In this paper, we propose a Real-World Knowledge Unlearning benchmark (RWKU) for LLM unlearning. RWKU is designed based on the following three key factors: (1) For the task setting, we consider a more practical and challenging unlearning setting, where neither the forget corpus nor the retain corpus is accessible. (2) For the knowledge source, we choose 200 real-world famous people as the unlearning targets and show that such popular knowledge is widely present in various LLMs. (3) For the evaluation framework, we design the forget set and the retain set to evaluate the model's capabilities across various real-world applications. Regarding the forget set, we provide four four membership inference attack (MIA) methods and nine kinds of adversarial attack probes to rigorously test unlearning efficacy. Regarding the retain set, we assess locality and utility in terms of neighbor perturbation, general ability, reasoning ability, truthfulness, factuality, and fluency. We conduct extensive experiments across two unlearning scenarios, two models and six baseline methods and obtain some meaningful findings. We release our benchmark and code publicly at http://rwku-bench.github.io for future work.

State-of-the-art large language models are sometimes distributed as open-source software but are also increasingly provided as a closed-source service. These closed-source large-language models typically see the widest usage by the public, however, they often do not provide an estimate of their uncertainty when responding to queries. As even the best models are prone to ``hallucinating" false information with high confidence, a lack of a reliable estimate of uncertainty limits the applicability of these models in critical settings. We explore estimating the uncertainty of closed-source LLMs via multiple rephrasings of an original base query. Specifically, we ask the model, multiple rephrased questions, and use the similarity of the answers as an estimate of uncertainty. We diverge from previous work in i) providing rules for rephrasing that are simple to memorize and use in practice ii) proposing a theoretical framework for why multiple rephrased queries obtain calibrated uncertainty estimates. Our method demonstrates significant improvements in the calibration of uncertainty estimates compared to the baseline and provides intuition as to how query strategies should be designed for optimal test calibration.

Large language models (LLMs) are raging over the medical domain, and their momentum has carried over into the mental health domain, leading to the emergence of few mental health LLMs. Although such mental health LLMs could provide reasonable suggestions for psychological counseling, how to develop an authentic and effective doctor-patient relationship (DPR) through LLMs is still an important problem. To fill this gap, we dissect DPR into two key attributes, i.e., the psychologist's empathy and proactive guidance. We thus present WundtGPT, an empathetic and proactive mental health large language model that is acquired by fine-tuning it with instruction and real conversation between psychologists and patients. It is designed to assist psychologists in diagnosis and help patients who are reluctant to communicate face-to-face understand their psychological conditions. Its uniqueness lies in that it could not only pose purposeful questions to guide patients in detailing their symptoms but also offer warm emotional reassurance. In particular, WundtGPT incorporates Collection of Questions, Chain of Psychodiagnosis, and Empathy Constraints into a comprehensive prompt for eliciting LLMs' questions and diagnoses. Additionally, WundtGPT proposes a reward model to promote alignment with empathetic mental health professionals, which encompasses two key factors: cognitive empathy and emotional empathy. We offer a comprehensive evaluation of our proposed model. Based on these outcomes, we further conduct the manual evaluation based on proactivity, effectiveness, professionalism and coherence. We notice that WundtGPT can offer professional and effective consultation. The model is available at huggingface.

Adopting human and large language models (LLM) as judges (a.k.a human- and LLM-as-a-judge) for evaluating the performance of LLMs has recently gained attention. Nonetheless, this approach concurrently introduces potential biases from human and LLMs, questioning the reliability of the evaluation results. In this paper, we propose a novel framework that is free from referencing groundtruth annotations for investigating Misinformation Oversight Bias, Gender Bias, Authority Bias and Beauty Bias on LLM and human judges. We curate a dataset referring to the revised Bloom's Taxonomy and conduct thousands of evaluations. Results show that human and LLM judges are vulnerable to perturbations to various degrees, and that even the cutting-edge judges possess considerable biases. We further exploit these biases to conduct attacks on LLM judges. We hope that our work can notify the community of the bias and vulnerability of human- and LLM-as-a-judge, as well as the urgency of developing robust evaluation systems.

Most large language models (LLMs) are sensitive to prompts, and another synonymous expression or a typo may lead to unexpected results for the model. Composing an optimal prompt for a specific demand lacks theoretical support and relies entirely on human experimentation, which poses a considerable obstacle to popularizing generative artificial intelligence. However, there is no systematic analysis of the stability of LLMs in resisting prompt perturbations in real-world scenarios. In this work, we propose to evaluate the ease-of-use of LLMs and construct E-Bench, simulating the actual situation of human use from synonymous perturbation (including paraphrasing, simplification, and colloquialism) and typographical perturbation (such as typing). On this basis, we also discuss the combination of these two types of perturbation and analyze the main reasons for performance degradation. Experimental results indicate that with the increase of model size, although the ease-of-use are significantly improved, there is still a long way to go to build a sufficiently user-friendly model.

Recently, there has been a growing availability of pre-trained text models on various model repositories. These models greatly reduce the cost of training new models from scratch as they can be fine-tuned for specific tasks or trained on large datasets. However, these datasets may not be publicly accessible due to the privacy, security, or intellectual property issues. In this paper, we aim to develop a lightweight student network that can learn from multiple teacher models without accessing their original training data. Hence, we investigate Data-Free Knowledge Amalgamation (DFKA), a knowledge-transfer task that combines insights from multiple pre-trained teacher models and transfers them effectively to a compact student network. To accomplish this, we propose STRATANET, a modeling framework comprising: (a) a steerable data generator that produces text data tailored to each teacher and (b) an amalgamation module that implements a self-regulative strategy using confidence estimates from the teachers' different layers to selectively integrate their knowledge and train a versatile student. We evaluate our method on three benchmark text classification datasets with varying labels or domains. Empirically, we demonstrate that the student model learned using our STRATANET outperforms several baselines significantly under data-driven and data-free constraints.

Large language models (LLMs) have become increasingly sophisticated, leading to widespread deployment in sensitive applications where safety and reliability are paramount. However, LLMs have inherent risks accompanying them, including bias, potential for unsafe actions, dataset poisoning, lack of explainability, hallucinations, and non-reproducibility. These risks necessitate the development of "guardrails" to align LLMs with desired behaviors and mitigate potential harm. This work explores the risks associated with deploying LLMs and evaluates current approaches to implementing guardrails and model alignment techniques. We examine intrinsic and extrinsic bias evaluation methods and discuss the importance of fairness metrics for responsible AI development. The safety and reliability of agentic LLMs (those capable of real-world actions) are explored, emphasizing the need for testability, fail-safes, and situational awareness. Technical strategies for securing LLMs are presented, including a layered protection model operating at external, secondary, and internal levels. System prompts, Retrieval-Augmented Generation (RAG) architectures, and techniques to minimize bias and protect privacy are highlighted. Effective guardrail design requires a deep understanding of the LLM's intended use case, relevant regulations, and ethical considerations. Striking a balance between competing requirements, such as accuracy and privacy, remains an ongoing challenge. This work underscores the importance of continuous research and development to ensure the safe and responsible use of LLMs in real-world applications.

Large Language Models (LLMs) have been applied to Math Word Problems (MWPs) with transformative impacts, revolutionizing how these complex problems are approached and solved in various domains including educational settings. However, the evaluation of these models often prioritizes final accuracy, overlooking the crucial aspect of reasoning capabilities. This work addresses this gap by focusing on the ability of LLMs to detect and correct reasoning mistakes. We introduce a novel dataset MWP-MISTAKE, incorporating MWPs with both correct and incorrect reasoning steps generated through rule-based methods and smaller language models. Our comprehensive benchmarking reveals significant insights into the strengths and weaknesses of state-of-the-art models, such as GPT-4o, GPT-4, GPT-3.5Turbo, and others. We highlight GPT-$o's superior performance in mistake detection and rectification and the persistent challenges faced by smaller models. Additionally, we identify issues related to data contamination and memorization, impacting the reliability of LLMs in real-world applications. Our findings emphasize the importance of rigorous evaluation of reasoning processes and propose future directions to enhance the generalization and robustness of LLMs in mathematical problem-solving.

Privacy issues arise prominently during the inappropriate transmission of information between entities. Existing research primarily studies privacy by exploring various privacy attacks, defenses, and evaluations within narrowly predefined patterns, while neglecting that privacy is not an isolated, context-free concept limited to traditionally sensitive data (e.g., social security numbers), but intertwined with intricate social contexts that complicate the identification and analysis of potential privacy violations. The advent of Large Language Models (LLMs) offers unprecedented opportunities for incorporating the nuanced scenarios outlined in privacy laws to tackle these complex privacy issues. However, the scarcity of open-source relevant case studies restricts the efficiency of LLMs in aligning with specific legal statutes. To address this challenge, we introduce a novel framework, GoldCoin, designed to efficiently ground LLMs in privacy laws for judicial assessing privacy violations. Our framework leverages the theory of contextual integrity as a bridge, creating numerous synthetic scenarios grounded in relevant privacy statutes (e.g., HIPAA), to assist LLMs in comprehending the complex contexts for identifying privacy risks in the real world. Extensive experimental results demonstrate that GoldCoin markedly enhances LLMs' capabilities in recognizing privacy risks across real court cases, surpassing the baselines on different judicial tasks.