Adobe, the maker of Photoshop, Premiere, and other industry-standard tools in the Creative Suite package, has its foot in its mouth. Last week an update to the Creative Cloud terms of service set off alarms across the web as users interpreted the new wording to mean that the company was using their cloud storage files to train its generative AI systems. Not true, says Adobe in a non-apology post. According to the message from Creative Cloud design leader Scott Belsky and legal, security, and policy lead Dana Rao, it's all been a big misunderstanding. The language that customers had noticed, which said that the company's automated systems can "access, view, or listen to your Content," sure seems like the kind of thing that enables generative AI systems to be trained.

Elon Musk took aim at Apple after it announced a partnership to use OpenAI's technology on its devices. Musk took to his social media platform X (formerly Twitter) to voice concerns about Apple potentially integrating OpenAI at the operating system level, after the tie-up was announced Monday, calling the deal a security risk. "If Apple integrates OpenAI at the OS level, then Apple devices will be banned at my companies," Musk wrote. "Visitors will have to check their Apple devices at the door, where they will be stored in a Faraday cage." Apple CEO Tim Cook announced the company's new AI offering, which the company dubs "Apple Intelligence," during the tech giant's developers conference on June 10.

We present Expert-Token-Routing, a unified generalist framework that facilitates seamless integration of multiple expert LLMs. Our framework represents expert LLMs as special expert tokens within the vocabulary of a meta LLM. The meta LLM can route to an expert LLM like generating new tokens. Expert-Token-Routing not only supports learning the implicit expertise of expert LLMs from existing instruction dataset but also allows for dynamic extension of new expert LLMs in a plug-and-play manner. It also conceals the detailed collaboration process from the user's perspective, facilitating interaction as though it were a singular LLM. Our framework outperforms various existing multi-LLM collaboration paradigms across benchmarks that incorporate six diverse expert domains, demonstrating effectiveness and robustness in building generalist LLM system via synergizing multiple expert LLMs.

Internet memes, channels for humor, social commentary, and cultural expression, are increasingly used to spread toxic messages. Studies on the computational analyses of toxic memes have significantly grown over the past five years, and the only three surveys on computational toxic meme analysis cover only work published until 2022, leading to inconsistent terminology and unexplored trends. Our work fills this gap by surveying content-based computational perspectives on toxic memes, and reviewing key developments until early 2024. Employing the PRISMA methodology, we systematically extend the previously considered papers, achieving a threefold result. First, we survey 119 new papers, analyzing 158 computational works focused on content-based toxic meme analysis. We identify over 30 datasets used in toxic meme analysis and examine their labeling systems. Second, after observing the existence of unclear definitions of meme toxicity in computational works, we introduce a new taxonomy for categorizing meme toxicity types. We also note an expansion in computational tasks beyond the simple binary classification of memes as toxic or non-toxic, indicating a shift towards achieving a nuanced comprehension of toxicity. Third, we identify three content-based dimensions of meme toxicity under automatic study: target, intent, and conveyance tactics. We develop a framework illustrating the relationships between these dimensions and meme toxicities. The survey analyzes key challenges and recent trends, such as enhanced cross-modal reasoning, integrating expert and cultural knowledge, the demand for automatic toxicity explanations, and handling meme toxicity in low-resource languages. Also, it notes the rising use of Large Language Models (LLMs) and generative AI for detecting and generating toxic memes. Finally, it proposes pathways for advancing toxic meme detection and interpretation.

There is growing consensus that language model (LM) developers should not be the sole deciders of LM behavior, creating a need for methods that enable the broader public to collectively shape the behavior of LM systems that affect them. To address this need, we present Collective Constitutional AI (CCAI): a multi-stage process for sourcing and integrating public input into LMs-from identifying a target population to sourcing principles to training and evaluating a model. We demonstrate the real-world practicality of this approach by creating what is, to our knowledge, the first LM fine-tuned with collectively sourced public input and evaluating this model against a baseline model trained with established principles from a LM developer. Our quantitative evaluations demonstrate several benefits of our approach: the CCAI-trained model shows lower bias across nine social dimensions compared to the baseline model, while maintaining equivalent performance on language, math, and helpful-harmless evaluations. Qualitative comparisons of the models suggest that the models differ on the basis of their respective constitutions, e.g., when prompted with contentious topics, the CCAI-trained model tends to generate responses that reframe the matter positively instead of a refusal. These results demonstrate a promising, tractable pathway toward publicly informed development of language models.

Large Language Models (LLMs) have demonstrated impressive capabilities across a range of scientific tasks including mathematics, physics, and chemistry. Despite their successes, the effectiveness of LLMs in handling complex statistical tasks remains systematically under-explored. To bridge this gap, we introduce StatQA, a new benchmark designed for statistical analysis tasks. StatQA comprises 11,623 examples tailored to evaluate LLMs' proficiency in specialized statistical tasks and their applicability assessment capabilities, particularly for hypothesis testing methods. We systematically experiment with representative LLMs using various prompting strategies and show that even state-of-the-art models such as GPT-4o achieve a best performance of only 64.83%, indicating significant room for improvement. Notably, while open-source LLMs (e.g., LLaMA-3) show limited capability, those fine-tuned ones exhibit marked improvements, outperforming all in-context learning-based methods (e.g., GPT-4o). Moreover, our comparative human experiments highlight a striking contrast in error types between LLMs and humans: LLMs primarily make applicability errors, whereas humans mostly make statistical task confusion errors. This divergence highlights distinct areas of proficiency and deficiency, suggesting that combining LLM and human expertise could lead to complementary strengths, inviting further investigation into their collaborative potential.

Recent years have seen two trends emerge simultaneously: large language models (LLMs) trained on increasing amounts of user data (generally scraped indiscriminately from the web), in parallel with increasing legal protections on digital data use including data revocation ("right to be forgotten") laws. In order to support data revocation for models that have already been trained on potentially sensitive data, a number of works have proposed approaches for data "unlearning" (Bourtoule et al., 2021; Gupta et al., 2021; Ginart et al., 2019), which aims to remove the influence of specific subsets of training data without entirely retraining a model. Unlearning in LLMs is particularly challenging because individuals' information may not be contained to specific data points (Brown et al., 2022; Tramèr et al., 2022). Nevertheless, recent work has shown that model finetuning is a promising approach to forget, for example, information corresponding to the book series Harry Potter (Eldan and Russinovich, 2023); information about specific individuals in a synthetic dataset (Maini et al., 2024); or knowledge that could give information to malicious agents Li et al. (2024). While finetuning is a promising approach, a number of recent works have shown that simple modifications to the input prompt or output postprocessing filters (which we collectively call "guardrails") can also be effective for generating a desirable output distribution from a model (Pawelczyk et al., 2023; Brown et al., 2020; Chowdhery et al., 2023; Wei et al., 2021; Kim et al., 2024). Prompt prefixes and postprocessing filters do not update the model weights, so the resulting model itself would not satisfy definitions of unlearning that require the distribution of model weights to match a model retrained from scratch Bourtoule et al. (2021). However, in practical settings where users can only access the model through an API, modifying the output distribution alone can suffice. In fact, most existing unlearning benchmarks (Eldan and Russinovich, 2023; Maini et al., 2024; unl, 2023; Li et al., 2024) only examine the model outputs when evaluating unlearning, which is consistent with a threat model in which users have only API access (see Section 3). In this paper, we investigate how existing benchmarks fare under guardrail-based approaches, and show that in three popular unlearning benchmarks, guardrails not only give strong performance comparable to finetuning baselines, but can also surface weaknesses or inconsistencies in the benchmarks or metrics themselves.

Large Language Models (LLMs) have significantly advanced the field of Natural Language Processing (NLP), achieving remarkable performance across diverse tasks and enabling widespread real-world applications. However, LLMs are prone to hallucination, generating content that either conflicts with established knowledge or is unfaithful to the original sources. Existing hallucination benchmarks primarily focus on sentence- or passage-level hallucination detection, neglecting dialogue-level evaluation, hallucination localization, and rationale provision. They also predominantly target factuality hallucinations while underestimating faithfulness hallucinations, often relying on labor-intensive or non-specialized evaluators. To address these limitations, we propose HalluDial, the first comprehensive large-scale benchmark for automatic dialogue-level hallucination evaluation. HalluDial encompasses both spontaneous and induced hallucination scenarios, covering factuality and faithfulness hallucinations. The benchmark includes 4,094 dialogues with a total of 146,856 samples. Leveraging HalluDial, we conduct a comprehensive meta-evaluation of LLMs' hallucination evaluation capabilities in information-seeking dialogues and introduce a specialized judge language model, HalluJudge. The high data quality of HalluDial enables HalluJudge to achieve superior or competitive performance in hallucination evaluation, facilitating the automatic assessment of dialogue-level hallucinations in LLMs and providing valuable insights into this phenomenon. The dataset and the code are available at https://github.com/FlagOpen/HalluDial.

Predictive models are often introduced to decision-making tasks under the rationale that they improve performance over an existing decision-making policy. However, it is challenging to compare predictive performance against an existing decision-making policy that is generally under-specified and dependent on unobservable factors. These sources of uncertainty are often addressed in practice by making strong assumptions about the data-generating mechanism. In this work, we propose a method to compare the predictive performance of decision policies under a variety of modern identification approaches from the causal inference and off-policy evaluation literatures (e.g., instrumental variable, marginal sensitivity model, proximal variable). Key to our method is the insight that there are regions of uncertainty that we can safely ignore in the policy comparison. We develop a practical approach for finite-sample estimation of regret intervals under no assumptions on the parametric form of the status quo policy. We verify our framework theoretically and via synthetic data experiments. We conclude with a real-world application using our framework to support a pre-deployment evaluation of a proposed modification to a healthcare enrollment policy.

The rapid advancements in the development of multimodal large language models (MLLMs) have consistently led to new breakthroughs on various benchmarks. In response, numerous challenging and comprehensive benchmarks have been proposed to more accurately assess the capabilities of MLLMs. However, there is a dearth of exploration of the higher-order perceptual capabilities of MLLMs. To fill this gap, we propose the Image Implication understanding Benchmark, II-Bench, which aims to evaluate the model's higher-order perception of images. Through extensive experiments on II-Bench across multiple MLLMs, we have made significant findings. Initially, a substantial gap is observed between the performance of MLLMs and humans on II-Bench. The pinnacle accuracy of MLLMs attains 74.8%, whereas human accuracy averages 90%, peaking at an impressive 98%. Subsequently, MLLMs perform worse on abstract and complex images, suggesting limitations in their ability to understand high-level semantics and capture image details. Finally, it is observed that most models exhibit enhanced accuracy when image sentiment polarity hints are incorporated into the prompts. This observation underscores a notable deficiency in their inherent understanding of image sentiment. We believe that II-Bench will inspire the community to develop the next generation of MLLMs, advancing the journey towards expert artificial general intelligence (AGI). II-Bench is publicly available at https://huggingface.co/datasets/m-a-p/II-Bench.