We present a new challenge to examine whether large language models understand social norms. In contrast to existing datasets, our dataset requires a fundamental understanding of social norms to solve. Our dataset features the largest set of social norm skills, consisting of 402 skills and 12,383 questions covering a wide set of social norms ranging from opinions and arguments to culture and laws. We design our dataset according to the K-12 curriculum. This enables the direct comparison of the social understanding of large language models to humans, more specifically, elementary students. While prior work generates nearly random accuracy on our benchmark, recent large language models such as GPT3.5-Turbo and LLaMA2-Chat are able to improve the performance significantly, only slightly below human performance. We then propose a multi-agent framework based on large language models to improve the models' ability to understand social norms. This method further improves large language models to be on par with humans. Given the increasing adoption of large language models in real-world applications, our finding is particularly important and presents a unique direction for future improvements.

Latent generative models (e.g., Stable Diffusion) have become more and more popular, but concerns have arisen regarding potential misuse related to images generated by these models. It is, therefore, necessary to analyze the origin of images by inferring if a particular image was generated by a specific latent generative model. Most existing methods (e.g., image watermark and model fingerprinting) require extra steps during training or generation. These requirements restrict their usage on the generated images without such extra operations, and the extra required operations might compromise the quality of the generated images. In this work, we ask whether it is possible to effectively and efficiently trace the images generated by a specific latent generative model without the aforementioned requirements. To study this problem, we design a latent inversion based method called LatentTracer to trace the generated images of the inspected model by checking if the examined images can be well-reconstructed with an inverted latent input. We leverage gradient based latent inversion and identify a encoder-based initialization critical to the success of our approach. Our experiments on the state-of-the-art latent generative models, such as Stable Diffusion, show that our method can distinguish the images generated by the inspected model and other images with a high accuracy and efficiency. Our findings suggest the intriguing possibility that today's latent generative generated images are naturally watermarked by the decoder used in the source models. Code: https://github.com/ZhentingWang/LatentTracer.

The recent breakthrough in large language models (LLMs) such as ChatGPT has revolutionized every industry at an unprecedented pace. Alongside this progress also comes mounting concerns about LLMs' susceptibility to jailbreaking attacks, which leads to the generation of harmful or unsafe content. While safety alignment measures have been implemented in LLMs to mitigate existing jailbreak attempts and force them to become increasingly complicated, it is still far from perfect. In this paper, we analyze the common pattern of the current safety alignment and show that it is possible to exploit such patterns for jailbreaking attacks by simultaneous obfuscation in queries and responses. Specifically, we propose WordGame attack, which replaces malicious words with word games to break down the adversarial intent of a query and encourage benign content regarding the games to precede the anticipated harmful content in the response, creating a context that is hardly covered by any corpus used for safety alignment. Extensive experiments demonstrate that WordGame attack can break the guardrails of the current leading proprietary and open-source LLMs, including the latest Claude 3, GPT 4, and Llama 3 models more effectively than existing attacks efficiently. Further ablation studies on such simultaneous obfuscation in query and response provide evidence of the merits of the attack strategy beyond an individual attack. Warning: The paper contains unfiltered text generated by LLMs which can be offensive.

Large language models (LLMs) are trained on a vast amount of human-written data, but data providers often remain uncredited. In response to this issue, data valuation (or data attribution), which quantifies the contribution or value of each data to the model output, has been discussed as a potential solution. Nevertheless, applying existing data valuation methods to recent LLMs and their vast training datasets has been largely limited by prohibitive compute and memory costs. In this work, we focus on influence functions, a popular gradient-based data valuation method, and significantly improve its scalability with an efficient gradient projection strategy called LoGra that leverages the gradient structure in backpropagation. We then provide a theoretical motivation of gradient projection approaches to influence functions to promote trust in the data valuation process. Lastly, we lower the barrier to implementing data valuation systems by introducing LogIX, a software package that can transform existing training code into data valuation code with minimal effort. In our data valuation experiments, LoGra achieves competitive accuracy against more expensive baselines while showing up to 6,500x improvement in throughput and 5x reduction in GPU memory usage when applied to Llama3-8B-Instruct and the 1B-token dataset.

We initiate the study of locally differentially private (LDP) learning with public features. We define semi-feature LDP, where some features are publicly available while the remaining ones, along with the label, require protection under local differential privacy. Under semi-feature LDP, we demonstrate that the mini-max convergence rate for non-parametric regression is significantly reduced compared to that of classical LDP. Then we propose HistOfTree, an estimator that fully leverages the information contained in both public and private features. Theoretically, HistOfTree reaches the mini-max optimal convergence rate. Empirically, HistOfTree achieves superior performance on both synthetic and real data. We also explore scenarios where users have the flexibility to select features for protection manually. In such cases, we propose an estimator and a data-driven parameter tuning strategy, leading to analogous theoretical and empirical results.

In a statement Monday afternoon, the Justice Department said it had levied criminal charges against Steven Anderegg, 42, of Holmen, Wis., for using the AI image generator Stable Diffusion to create over 13,000 fake images of minors, many of which depicted fully or partially nude children touching their genitals or engaging in sexual intercourse with men.

The US Department of Justice arrested a Wisconsin man last week for generating and distributing AI-generated child sexual abuse material (CSAM). As far as we know, this is the first case of its kind as the DOJ looks to establish a judicial precedent that exploitative materials are still illegal even when no children were used to create them. "Put simply, CSAM generated by AI is still CSAM," Deputy Attorney General Lisa Monaco wrote in a press release. The DOJ says 42-year-old software engineer Steven Anderegg of Holmen, WI, used a fork of the open-source AI image generator Stable Diffusion to make the images, which he then used to try to lure an underage boy into sexual situations. The latter will likely play a central role in the eventual trial for the four counts of "producing, distributing, and possessing obscene visual depictions of minors engaged in sexually explicit conduct and transferring obscene material to a minor under the age of 16."

Scarlett Johannson has gone to war with OpenAI, and in the battle for public opinion, OpenAI is losing--badly. Last week, OpenAI released an update of its AI chatbot called ChatGPT-4o, which featured a female voice talking to its users. Many people pointed out that the voice, which sometimes seemed to veer into flirtation, was eerily similar to Scarlett Johannson's in the 2013 dystopian sci-fi film Her. OpenAI CEO Sam Altman has long talked about how much the movie inspired the company's products, and even made the connection clear last week by tweeting the title of the movie. But on Monday, Johannson released a statement saying OpenAI had asked her to be the voice of the chatbot, and when she refused, they found a soundalike.

The FBI has charged a US man with creating more than 10,000 sexually explicit and abusive images of children, which he allegedly generated using a popular artificial intelligence tool. Authorities also accused the man, 42-year-old Steven Anderegg, of sending pornographic AI-made images to a 15-year-old boy over Instagram. Anderegg crafted about 13,000 "hyper-realistic images of nude and semi-clothed prepubescent children", prosecutors stated in an indictment released on Monday, often images depicting children touching their genitals or being sexually abused by adult men. Evidence from the Wisconsin man's laptop allegedly showed he used the popular Stable Diffusion AI model, which turns text descriptions into images. Anderegg's charges came after the National Center for Missing & Exploited Children (NCMEC) received two reports last year that flagged his Instagram account, which prompted law enforcement officials to monitor his activity on the social network, obtain information from Instagram and eventually obtain a search warrant.

'The CyberGuy' Kurt Knutsson joins'Fox & Friends Weekend' to discuss Elon Musk's lawsuit against OpenAI and its CEO over a contractual breach, saying hes right on this one. "Avengers" and "Her" actress Scarlett Johansson revealed that legal action was likely behind OpenAI removing a voice that sounded eerily like hers. A statement released by NPR on Monday explained that OpenAI CEO Sam Altman reached out to Johansson in September about possibly hiring her to voice the ChatGPT 4.0 system. She claimed he suggested her "comforting" voice "could bridge the gap between tech companies and creatives" and help with the "seismic shift concerning humans and Al." Though she rejected the offer after "much consideration and for personal reasons," Johansson was furious to hear the public discuss how the "Sky" voice system resembled hers. Scarlett Johansson said in a statement that she took legal action against OpenAI CEO Sam Altman and the company.