Large language models (LLMs), while memorizing parts of their training data scraped from the Internet, may also inadvertently encode societal preferences and norms. As these models are integrated into sociotechnical systems, it is crucial that the norms they encode align with societal expectations. These norms could vary across models, hyperparameters, optimization techniques, and datasets. This is especially challenging due to prompt sensitivity$-$small variations in prompts yield different responses, rendering existing assessment methodologies unreliable. There is a need for a comprehensive framework covering various models, optimization, and datasets, along with a reliable methodology to assess encoded norms. We present LLM-CI, the first open-sourced framework to assess privacy norms encoded in LLMs. LLM-CI uses a Contextual Integrity-based factorial vignette methodology to assess the encoded norms across different contexts and LLMs. We propose the multi-prompt assessment methodology to address prompt sensitivity by assessing the norms from only the prompts that yield consistent responses across multiple variants. Using LLM-CI and our proposed methodology, we comprehensively evaluate LLMs using IoT and COPPA vignettes datasets from prior work, examining the impact of model properties (e.g., hyperparameters, capacity) and optimization strategies (e.g., alignment, quantization).

X will permanently avoid training its AI chatbot Grok on the public posts of users in the European Union and European Economic Area following pressure from a regulator in the region. Last month, the company temporarily suspended the practice after Ireland's Data Protection Commission (DPC) opened High Court proceedings against it. X has now made that commitment a permanent one, which prompted the DPC to end its legal action. The DPC, which is the chief EU regulator for X, raised concerns that X may have been violating data protection rules and users' rights. Since May, X had offered users the option to opt-out of having their public posts being used to train Grok, implying that the company had enabled that setting for public accounts by default.

In a statement to South Korea's Communications Standards Commission (KCSC), Telegram said the situation was "unfortunate", adding that it "apologised if there had been an element of misunderstanding". It also confirmed that it had taken down 25 such videos as requested by KCSC. In its latest statement to KCSC, Telegram also proposed an email address dedicated to future communication with the regulator. KCSC described the company's approach as "very forward-looking" and said Telegram has "acknowledged the seriousness" of the situation. Deepfakes are generated using artificial intelligence, and often combine the face of a real person with a fake body.

Shares in the AI chip designer Nvidia have continued to slide overnight after a report said US authorities were ramping up an investigation into whether the company had breached competition laws. The company's shares fell 2.4% in after-hours trading, exacerbating a near-10% drop in the regular trading session that slashed its value by 279bn ( 212bn) to 2.6tn, marking the largest one-day drop in history for a US company. Overnight, the US Department of Justice sent subpoenas to Nvidia and other tech companies, in a move that will force recipients to provide information under law, Bloomberg reported. Officials are said to be concerned the company has made it harder for clients to switch to other semiconductor suppliers and is penalising buyers that refuse to exclusively use Nvidia's AI chips. Such a move would signal an escalation of the US antitrust investigation, and brings the government a step closer to launching a formal complaint against Nvidia. The sell-off on Tuesday came amid a wider sell-off on markets sparked by weak US manufacturing data that raised broader concerns about the outlook for the country's economy among investors.

In the pursuit of an effective spam detection system, the focus has often been on identifying known spam patterns either through rule-based detection systems or machine learning (ML) solutions that rely on keywords. However, both systems are susceptible to evasion techniques and zero-day attacks that can be achieved at low cost. Therefore, an email that bypassed the defense system once can do it again in the following days, even though rules are updated or the ML models are retrained. The recurrence of failures to detect emails that exhibit layout similarities to previously undetected spam is concerning for customers and can erode their trust in a company. Our observations show that threat actors reuse email kits extensively and can bypass detection with little effort, for example, by making changes to the content of emails. In this work, we propose an email visual similarity detection approach, named Pisco, to improve the detection capabilities of an email threat defense system. We apply our proof of concept to some real-world samples received from different sources. Our results show that email kits are being reused extensively and visually similar emails are sent to our customers at various time intervals. Therefore, this method could be very helpful in situations where detection engines that rely on textual features and keywords are bypassed, an occurrence our observations show happens frequently.

We propose misogyny detection as an Argumentative Reasoning task and we investigate the capacity of large language models (LLMs) to understand the implicit reasoning used to convey misogyny in both Italian and English. The central aim is to generate the missing reasoning link between a message and the implied meanings encoding the misogyny. Our study uses argumentation theory as a foundation to form a collection of prompts in both zero-shot and few-shot settings. These prompts integrate different techniques, including chain-of-thought reasoning and augmented knowledge. Our findings show that LLMs fall short on reasoning capabilities about misogynistic comments and that they mostly rely on their implicit knowledge derived from internalized common stereotypes about women to generate implied assumptions, rather than on inductive reasoning.

In this article, we propose the R2GQA system, a Retriever-Reader-Generator Question Answering system, consisting of three main components: Document Retriever, Machine Reader, and Answer Generator. The Retriever module employs advanced information retrieval techniques to extract the context of articles from a dataset of legal regulation documents. The Machine Reader module utilizes state-of-the-art natural language understanding algorithms to comprehend the retrieved documents and extract answers. Finally, the Generator module synthesizes the extracted answers into concise and informative responses to questions of students regarding legal regulations. Furthermore, we built the ViRHE4QA dataset in the domain of university training regulations, comprising 9,758 question-answer pairs with a rigorous construction process. This is the first Vietnamese dataset in the higher regulations domain with various types of answers, both extractive and abstractive. In addition, the R2GQA system is the first system to offer abstractive answers in Vietnamese. This paper discusses the design and implementation of each module within the R2GQA system on the ViRHE4QA dataset, highlighting their functionalities and interactions. Furthermore, we present experimental results demonstrating the effectiveness and utility of the proposed system in supporting the comprehension of students of legal regulations in higher education settings. In general, the R2GQA system and the ViRHE4QA dataset promise to contribute significantly to related research and help students navigate complex legal documents and regulations, empowering them to make informed decisions and adhere to institutional policies effectively. Our dataset is available for research purposes.

With the rapid advancement of Large Language Models (LLMs), long-context information understanding and processing have become a hot topic in academia and industry. However, benchmarks for evaluating the ability of LLMs to handle long-context information do not seem to have kept pace with the development of LLMs. Despite the emergence of various long-context evaluation benchmarks, the types of capability assessed are still limited, without new capability dimensions. In this paper, we introduce DetectiveQA, a narrative reasoning benchmark featured with an average context length of over 100K tokens. DetectiveQA focuses on evaluating the long-context reasoning ability of LLMs, which not only requires a full understanding of context but also requires extracting important evidences from the context and reasoning according to extracted evidences to answer the given questions. This is a new dimension of capability evaluation, which is more in line with the current intelligence level of LLMs. We use detective novels as data sources, which naturally have various reasoning elements. Finally, we manually annotated 600 questions in Chinese and then also provided an English edition of the context information and questions. We evaluate many long-context LLMs on DetectiveQA, including commercial and open-sourced models, and the results indicate that existing long-context LLMs still require significant advancements to effectively process true long-context dependency questions.

In this paper, we investigate the presence of additive bias in Large Language Models (LLMs), drawing a parallel to the cognitive bias observed in humans where individuals tend to favor additive over subtractive changes. Using a series of controlled experiments, we tested various LLMs, including GPT-3.5 Turbo, Claude 3.5 Sonnet, Mistral, Math$\Sigma$tral, and Llama 3.1, on tasks designed to measure their propensity for additive versus subtractive modifications. Our findings demonstrate a significant preference for additive changes across all tested models. For example, in a palindrome creation task, Llama 3.1 favored adding letters 97.85% of the time over removing them. Similarly, in a Lego tower balancing task, GPT-3.5 Turbo chose to add a brick 76.38% of the time rather than remove one. In a text summarization task, Mistral 7B produced longer summaries in 59.40% to 75.10% of cases when asked to improve its own or others' writing. These results indicate that, similar to humans, LLMs exhibit a marked additive bias, which might have implications when LLMs are used on a large scale. Addittive bias might increase resource use and environmental impact, leading to higher economic costs due to overconsumption and waste. This bias should be considered in the development and application of LLMs to ensure balanced and efficient problem-solving approaches.

International AI governance agreements and institutions may play an important role in reducing global security risks from advanced AI. To inform the design of such agreements and institutions, we conducted case studies of historical and contemporary international security agreements. We focused specifically on those arrangements around dual-use technologies, examining agreements in nuclear security, chemical weapons, biosecurity, and export controls. For each agreement, we examined four key areas: (a) purpose, (b) core powers, (c) governance structure, and (d) instances of non-compliance. From these case studies, we extracted lessons for the design of international AI agreements and governance institutions. We discuss the importance of robust verification methods, strategies for balancing power between nations, mechanisms for adapting to rapid technological change, approaches to managing trade-offs between transparency and security, incentives for participation, and effective enforcement mechanisms.