As modern Large Language Models (LLMs) shatter many state-of-the-art benchmarks in a variety of domains, this paper investigates their behavior in the domains of ethics and fairness, focusing on protected group bias. We conduct a two-part study: first, we solicit sentence continuations describing the occupations of individuals from different protected groups, including gender, sexuality, religion, and race. Second, we have the model generate stories about individuals who hold different types of occupations. We collect >10k sentence completions made by a publicly available LLM, which we subject to human annotation. We find bias across minoritized groups, but in particular in the domains of gender and sexuality, as well as Western bias, in model generations. The model not only reflects societal biases, but appears to amplify them. The model is additionally overly cautious in replies to queries relating to minoritized groups, providing responses that strongly emphasize diversity and equity to an extent that other group characteristics are overshadowed. This suggests that artificially constraining potentially harmful outputs may itself lead to harm, and should be applied in a careful and controlled manner.

In this paper, we present PARAMANU-AYN, a language model based exclusively on case documents of the Supreme Court of India, the Constitution of India, and the Indian Penal Code. The novel Auto Regressive (AR) decoder based model is pretrained from scratch at a context size of 8192. We evaluated our pretrained legal model on perplexity metrics. We also instruction-tuned our pretrained model on a set of 10,763 instructions covering various legal tasks such as legal reasoning, judgement explanation, legal clause generation, legal drafting, legal contract drafting, case summarization, constitutional question-answering, etc. We also evaluated the responses of prompts for instruction-tuned models by GPT-3.5-Turbo on clarity, relevance, completeness, and legal reasoning metrics in a scale of 10. Our model can be run on CPU and achieved 42.46 tokens/sec CPU inference speed. We found that our models, despite not being pretrained on legal books, various legal contracts, and legal documents, were able to learn the domain knowledge required for drafting various legal contracts and legal clauses, and generalize to draft legal contracts and legal clauses with limited instruction tuning. Hence, we conclude that for a strong domain-specialized generative language model (such as legal), very large amounts of data are not required to develop models from scratch. We believe that this work is the first attempt to make a dedicated generative legal language model from scratch for Indian Supreme Court jurisdiction or in legal NLP overall. We plan to release our Paramanu-Ayn model at https://www.bharatgpts.com.

With the advent of the AI Act and other regulations, there is now an urgent need for algorithms that repair unfairness in training data. In this paper, we define fairness in terms of conditional independence between protected attributes ($S$) and features ($X$), given unprotected attributes ($U$). We address the important setting in which torrents of archival data need to be repaired, using only a small proportion of these data, which are $S|U$-labelled (the research data). We use the latter to design optimal transport (OT)-based repair plans on interpolated supports. This allows {\em off-sample}, labelled, archival data to be repaired, subject to stationarity assumptions. It also significantly reduces the size of the supports of the OT plans, with correspondingly large savings in the cost of their design and of their {\em sequential\/} application to the off-sample data. We provide detailed experimental results with simulated and benchmark real data (the Adult data set). Our performance figures demonstrate effective repair -- in the sense of quenching conditional dependence -- of large quantities of off-sample, labelled (archival) data.

Network slicing, a cornerstone technology for future networks, enables the creation of customized virtual networks on a shared physical infrastructure. This fosters innovation and agility by providing dedicated resources tailored to specific applications. However, current orchestration and management approaches face limitations in handling the complexity of new service demands within multi-administrative domain environments. This paper proposes a future vision for network slicing powered by Large Language Models (LLMs) and multi-agent systems, offering a framework that can be integrated with existing Management and Orchestration (MANO) frameworks. This framework leverages LLMs to translate user intent into technical requirements, map network functions to infrastructure, and manage the entire slice lifecycle, while multi-agent systems facilitate collaboration across different administrative domains. We also discuss the challenges associated with implementing this framework and potential solutions to mitigate them.

The politics of AI regulation became a little clearer this weekend, after an influential Labour thinktank laid out its framework for how the party should approach the topic in its manifesto. The policy paper, produced by the centre-left Labour Together thinktank, proposes a legal ban on dedicated nudification tools that allow users to generate explicit content by uploading images of real people. It would also create an obligation for developers of general-purpose AI tools and web hosting companies to take reasonable steps to ensure they are not involved in the production of such images, or other harmful deepfakes. Labour Together's suggestions aren't party policy yet, but they point at the sort of issues Westminster wonks think a campaign can be built on. For the last few decades, technology has been a curiously apolitical realm in the UK, with all parties agreeing on the vague idea that it's important to support British technology as a driver of growth and soft power, and little active campaigning beyond that.

NVIDIA's H100 chips are used by nearly every AI company in the world to train large language models hooked into services like ChatGPT. It's been great for business. Now, the company is ready to make those chips look terrible, announcing a next-generation platform called Blackwell. Named for David Harold Blackwell, a mathematician who specialized in game theory and statistics, NVIDIA claims Blackwell is the world's most powerful chip, reaching speeds of 20 petaflops compared to just 4 petaflops the H100 provided. Yeah, throw it in the trash.

YouTube has updated its rulebook for the era of deepfakes. Starting today, anyone uploading video to the platform must disclose certain uses of synthetic media, including generative AI, so viewers know what they're seeing isn't real. YouTube says it applies to "realistic" altered media such as "making it appear as if a real building caught fire" or swapping "the face of one individual with another's." The new policy shows YouTube taking steps that could help curb the spread of AI-generated misinformation as the US presidential election approaches. It is also striking for what it permits: AI-generated animations aimed at kids are not subject to the new synthetic content disclosure rules.

Given the uncertainty surrounding how existing explainability methods for autonomous vehicles (AVs) meet the diverse needs of stakeholders, a thorough investigation is imperative to determine the contexts requiring explanations and suitable interaction strategies. A comprehensive review becomes crucial to assess the alignment of current approaches with the varied interests and expectations within the AV ecosystem. This study presents a review to discuss the complexities associated with explanation generation and presentation to facilitate the development of more effective and inclusive explainable AV systems. Our investigation led to categorising existing literature into three primary topics: explanatory tasks, explanatory information, and explanatory information communication. Drawing upon our insights, we have proposed a comprehensive roadmap for future research centred on (i) knowing the interlocutor, (ii) generating timely explanations, (ii) communicating human-friendly explanations, and (iv) continuous learning. Our roadmap is underpinned by principles of responsible research and innovation, emphasising the significance of diverse explanation requirements. To effectively tackle the challenges associated with implementing explainable AV systems, we have delineated various research directions, including the development of privacy-preserving data integration, ethical frameworks, real-time analytics, human-centric interaction design, and enhanced cross-disciplinary collaborations. By exploring these research directions, the study aims to guide the development and deployment of explainable AVs, informed by a holistic understanding of user needs, technological advancements, regulatory compliance, and ethical considerations, thereby ensuring safer and more trustworthy autonomous driving experiences.

Payroll management is a critical business task that is subject to a large number of rules, which vary widely between companies, sectors, and countries. Moreover, the rules are often complex and change regularly. Therefore, payroll management systems must be flexible in design. In this paper, we suggest an approach based on a flexible Answer Set Programming (ASP) model and an easy-to-read tabular representation based on the Decision Model and Notation (DMN) standard. It allows HR consultants to represent complex rules without the need for a software engineer, and to ultimately design payroll systems for a variety of different scenarios. We show how the multi-shot solving capabilities of the clingo ASP system can be used to reach the performance that is necessary to handle real-world instances.

The current societal challenges exceed the capacity of human individual or collective effort alone. As AI evolves, its role within human collectives is poised to vary from an assistive tool to a participatory member. Humans and AI possess complementary capabilities that, when synergized, can achieve a level of collective intelligence that surpasses the collective capabilities of either humans or AI in isolation. However, the interactions in human-AI systems are inherently complex, involving intricate processes and interdependencies. This review incorporates perspectives from network science to conceptualize a multilayer representation of human-AI collective intelligence, comprising a cognition layer, a physical layer, and an information layer. Within this multilayer network, humans and AI agents exhibit varying characteristics; humans differ in diversity from surface-level to deep-level attributes, while AI agents range in degrees of functionality and anthropomorphism. The interplay among these agents shapes the overall structure and dynamics of the system. We explore how agents' diversity and interactions influence the system's collective intelligence. Furthermore, we present an analysis of real-world instances of AI-enhanced collective intelligence. We conclude by addressing the potential challenges in AI-enhanced collective intelligence and offer perspectives on future developments in this field.