Later this year, millions of Apple devices will begin running Apple Intelligence, Cupertino's take on generative AI that, among other things, lets people create images from text prompts. But some members of the creative community are unhappy about what they say is the company's lack of transparency around the raw information powering the AI model that makes this possible. "I wish Apple would have explained to the public in a more transparent way how they collected their training data," Jon Lam, a video games artist and a creators' rights activist based in Vancouver, told Engadget. "I think their announcement could not have come at a worse time." Creatives have historically been some of the most loyal customers of Apple, a company whose founder famously positioned it at the "intersection of technology and liberal arts."

We all want to be able to speak our minds online--to be heard by our friends and talk (back) to our opponents. At the same time, we don't want to be exposed to unpleasant speech. Technology companies address this conundrum by setting standards for free speech, a practice protected under federal law, hiring in-house moderators to examine individual pieces of content and removing them if posts violate predefined rules. The approach clearly has problems: harassment, misinformation about topics like public health, and false descriptions of legitimate elections run rampant. But even if content moderation were implemented perfectly, it would still miss a whole host of issues. We need a new strategy: treat social media companies as potential polluters of the social fabric, and directly measure and mitigate the effects their choices have on us.

The rapid advancements in artificial intelligence, big data analytics, and cloud computing have precipitated an unprecedented demand for computational resources. However, the current landscape of computational resource allocation is characterized by significant inefficiencies, including underutilization and price volatility. This paper addresses these challenges by introducing a novel global platform for the commodification of compute hours, termed the Global Compute Exchange (GCX) (Patent Pending). The GCX leverages blockchain technology and smart contracts to create a secure, transparent, and efficient marketplace for buying and selling computational power. The GCX is built in a layered fashion, comprising Market, App, Clearing, Risk Management, Exchange (Offchain), and Blockchain (Onchain) layers, each ensuring a robust and efficient operation. This platform aims to revolutionize the computational resource market by fostering a decentralized, efficient, and transparent ecosystem that ensures equitable access to computing power, stimulates innovation, and supports diverse user needs on a global scale. By transforming compute hours into a tradable commodity, the GCX seeks to optimize resource utilization, stabilize pricing, and democratize access to computational resources. This paper explores the technological infrastructure, market potential, and societal impact of the GCX, positioning it as a pioneering solution poised to drive the next wave of innovation in commodities and compute.

Diffusion models have marked a significant milestone in the enhancement of image and video generation technologies. However, generating videos that precisely retain the shape and location of moving objects such as robots remains a challenge. This paper presents diffusion models specifically tailored to generate videos that accurately maintain the shape and location of mobile robots. This development offers substantial benefits to those working on detecting dangerous interactions between humans and robots by facilitating the creation of training data for collision detection models, circumventing the need for collecting data from the real world, which often involves legal and ethical issues. Our models incorporate techniques such as embedding accessible robot pose information and applying semantic mask regulation within the ConvNext backbone network. These techniques are designed to refine intermediate outputs, therefore improving the retention performance of shape and location. Through extensive experimentation, our models have demonstrated notable improvements in maintaining the shape and location of different robots, as well as enhancing overall video generation quality, compared to the benchmark diffusion model. Codes will be opensourced at \href{https://github.com/PengPaulWang/diffusion-robots}{Github}.

Jailbreaking attacks can enable Large Language Models (LLMs) to bypass the safeguard and generate harmful content. Existing jailbreaking defense methods have failed to address the fundamental issue that harmful knowledge resides within the model, leading to potential jailbreak risks for LLMs. In this paper, we propose a novel defense method called Eraser, which mainly includes three goals: unlearning harmful knowledge, retaining general knowledge, and maintaining safety alignment. The intuition is that if an LLM forgets the specific knowledge required to answer a harmful question, it will no longer have the ability to answer harmful questions. The training of Erase does not actually require the model's own harmful knowledge, and it can benefit from unlearning general answers related to harmful queries, which means it does not need assistance from the red team. The experimental results show that Eraser can significantly reduce the jailbreaking success rate for various attacks without compromising the general capabilities of the model. Our codes are available at https://github.com/ZeroNLP/Eraser.

Using language models as a remote service entails sending private information to an untrusted provider. In addition, potential eavesdroppers can intercept the messages, thereby exposing the information. In this work, we explore the prospects of avoiding such data exposure at the level of text manipulation. We focus on text classification models, examining various token mapping and contextualized manipulation functions in order to see whether classifier accuracy may be maintained while keeping the original text unrecoverable. We find that although some token mapping functions are easy and straightforward to implement, they heavily influence performance on the downstream task, and via a sophisticated attacker can be reconstructed. In comparison, the contextualized manipulation provides an improvement in performance.

Large language models (LLMs) appear to bias their survey answers toward certain values. Nonetheless, some argue that LLMs are too inconsistent to simulate particular values. Are they? To answer, we first define value consistency as the similarity of answers across (1) paraphrases of one question, (2) related questions under one topic, (3) multiple-choice and open-ended use-cases of one question, and (4) multilingual translations of a question to English, Chinese, German, and Japanese. We apply these measures to a few large ($>=34b$), open LLMs including llama-3, as well as gpt-4o, using eight thousand questions spanning more than 300 topics. Unlike prior work, we find that models are relatively consistent across paraphrases, use-cases, translations, and within a topic. Still, some inconsistencies remain. Models are more consistent on uncontroversial topics (e.g., in the U.S., "Thanksgiving") than on controversial ones ("euthanasia"). Base models are both more consistent compared to fine-tuned models and are uniform in their consistency across topics, while fine-tuned models are more inconsistent about some topics ("euthanasia") than others ("women's rights") like our human subjects (n=165).

Is explainability a false promise? This debate has emerged from the insufficient evidence that explanations aid people in situations they are introduced for. More human-centered, application-grounded evaluations of explanations are needed to settle this. Yet, with no established guidelines for such studies in NLP, researchers accustomed to standardized proxy evaluations must discover appropriate measurements, tasks, datasets, and sensible models for human-AI teams in their studies. To help with this, we first review fitting existing metrics. We then establish requirements for datasets to be suitable for application-grounded evaluations. Among over 50 datasets available for explainability research in NLP, we find that 4 meet our criteria. By finetuning Flan-T5-3B, we demonstrate the importance of reassessing the state of the art to form and study human-AI teams. Finally, we present the exemplar studies of human-AI decision-making for one of the identified suitable tasks -- verifying the correctness of a legal claim given a contract.

Vision-language models (VLMs) pre-trained on extensive datasets can inadvertently learn biases by correlating gender information with specific objects or scenarios. Current methods, which focus on modifying inputs and monitoring changes in the model's output probability scores, often struggle to comprehensively understand bias from the perspective of model components. We propose a framework that incorporates causal mediation analysis to measure and map the pathways of bias generation and propagation within VLMs. This approach allows us to identify the direct effects of interventions on model bias and the indirect effects of interventions on bias mediated through different model components. Our results show that image features are the primary contributors to bias, with significantly higher impacts than text features, specifically accounting for 32.57% and 12.63% of the bias in the MSCOCO and PASCAL-SENTENCE datasets, respectively. Notably, the image encoder's contribution surpasses that of the text encoder and the deep fusion encoder. Further experimentation confirms that contributions from both language and vision modalities are aligned and non-conflicting. Consequently, focusing on blurring gender representations within the image encoder, which contributes most to the model bias, reduces bias efficiently by 22.03% and 9.04% in the MSCOCO and PASCAL-SENTENCE datasets, respectively, with minimal performance loss or increased computational demands.

With the rapid advancements in Multimodal Large Language Models (MLLMs), securing these models against malicious inputs while aligning them with human values has emerged as a critical challenge. In this paper, we investigate an important and unexplored question of whether techniques that successfully jailbreak Large Language Models (LLMs) can be equally effective in jailbreaking MLLMs. To explore this issue, we introduce JailBreakV-28K, a pioneering benchmark designed to assess the transferability of LLM jailbreak techniques to MLLMs, thereby evaluating the robustness of MLLMs against diverse jailbreak attacks. Utilizing a dataset of 2, 000 malicious queries that is also proposed in this paper, we generate 20, 000 text-based jailbreak prompts using advanced jailbreak attacks on LLMs, alongside 8, 000 image-based jailbreak inputs from recent MLLMs jailbreak attacks, our comprehensive dataset includes 28, 000 test cases across a spectrum of adversarial scenarios. Our evaluation of 10 open-source MLLMs reveals a notably high Attack Success Rate (ASR) for attacks transferred from LLMs, highlighting a critical vulnerability in MLLMs that stems from their text-processing capabilities. Our findings underscore the urgent need for future research to address alignment vulnerabilities in MLLMs from both textual and visual inputs.