Large language models (LLMs) have significantly advanced artificial intelligence (AI) and natural language processing (NLP) by excelling in tasks like text generation, machine translation, question answering and sentiment analysis, often rivaling human performance. This paper reviews LLMs’ foundations, advancements and applications, beginning with the transformative transformer architecture, which improved on earlier models like recurrent neural networks and convolutional neural networks through self-attention mechanisms that capture long-range dependencies and contextual relationships. Key innovations such as masked language modeling and causal language modeling underpin leading models like Bidirectional encoder representations from transformers (BERT) and the Generative Pre-trained Transformer (GPT) series. The paper highlights scaling laws, model size increases and advanced training techniques that have driven LLMs’ growth. It also explores methodologies to enhance their precision and adaptability, including parameter-efficient fine-tuning and prompt engineering. Challenges like high computational demands, biases and hallucinations are addressed, with solutions such as retrieval-augmented generation to improve factual accuracy. By discussing LLMs’ strengths, limitations and transformative potential, this paper provides researchers, practitioners and students with a comprehensive understanding. It underscores the importance of ongoing research to improve efficiency, manage ethical concerns and shape the future of AI and language technologies.

Large or “foundation” models are now being widely used to generate not just text and images but also video, music and code from prompts. Although this “generative AI” revolution is clearly driving new opportunities for innovation and creativity, it is also enabling easy and rapid dissemination of harmful speech and potentially infringing existing laws. Much attention has been paid recently to how we can draft bespoke legislation to control these risks and harms; however, private ordering by generative AI providers, via user contracts, licenses and privacy policies, has so far attracted less attention. Drawing on the extensive history of study of the terms and conditions (T&C) and privacy policies of social media companies, this paper reports the results of pilot empirical work conducted in January–March 2023, in which T&C were mapped across a representative sample of generative AI. With the focus on copyright and data protection, our early findings indicate the emergence of a “platformisation paradigm,” in which providers of generative AI attempt to position themselves as neutral intermediaries. This study concludes that new laws targeting “big tech” must be carefully reconsidered to avoid repeating past power imbalances between users and platforms.

This paper explores the regulatory awakening regarding generative AI (GenAI) in the United States and European Union (EU) institutions with the release of ChatGPT. Based on a thematic analysis of regulatory documents, it investigates how governments have approached the deployment and use of this emerging technology within their classic government activities. The analysis shows several layers of regulatory approaches, ranging from command-and-control to an experimental approach, combined with risk- and management-based approaches. It also reveals different perspectives. The EU institutions have notably adopted more restrictive guidelines on the use of publicly available Large Language Models (LLMs) - a type o GenAI that is trained on vast amounts of text data to understand, generate, and respond in human-like language. This approach reflects greater caution about data security and confidentiality and the risks of foreign interference. However, the American and EU documents share a common concern about the risk of reinforcing discrimination and the protection of human rights. Interestingly, considering the administrative environment, neither the administrative activities in which GenAI may be used nor the key legal principles embedded by the rule of law are explicitly used for guiding administration in their development and use of GenAI. In this context, the paper calls for future research that could help contribute to the renewal of administrative law theory in the context of the digital transformation of public administration.

Generative artificial intelligence (AI) has catapulted into the legal debate through the popular applications ChatGPT, Bard, Dall-E and others. While the predominant focus has hitherto centred on issues of copyright infringement and regulatory strategies, particularly in the context of the AI Act, a critical but often overlooked issue lies in the friction between generative AI and data protection laws. The rise of these technologies highlights unresolved tension between safeguarding fundamental protection rights and and the vast, almost universal, of scale of data processing required for machine learning. Large language models, which scrape nearly the whole Internet rely on and may even generate personal data falling under the GDPR. This tension manifests across multiple dimensions, encompassing data subjects’ rights, the foundational principles of data protection and the fundamental categories of data protection. Drawing on ongoing investigations by data protection authorities in Europe, this paper undertakes a comprehensive analysis of the intricate interplay between generative AI and data protection within the European legal framework.

Generative artificial intelligence (GenAI) systems generate artwork likely to be copyrightable if made by human. In lieu of discussing whether machine-enabled artworks should be copyrighted, a question that has received a large attention in academic scholarship, this paper proposes to analyse the economic implications of GenAI. Although the valuation of their outputs is still mostly unknown, preliminary studies show that, all other things being equal, humans’ works are evaluated at a significant higher value than machine-enabled ones. Yet, to be properly valued, human-made and machine-enabled products must be distinguishable. They are not. This indistinguishability creates an asymmetry in information that in turn leads to a lemons problem, defined as a market erosion of good-quality products. Against that background, this paper proposes a solution in light of European Union rules of origin and the so-called substantial transformation test. This solution can then be used in copyright law to identify where the human author has been able to make free and creative choices (or lack thereof) required to satisfy the criterion of originality and, hence, copyrightability.

Several criminal offenses can originate from or culminate with the creation of content. Sexual abuse can be committed by producing intimate materials without the subject’s consent, while incitement to violence or self-harm can begin with a conversation. When the task of generating content is entrusted to artificial intelligence (AI), it becomes necessary to explore the risks of this technology. AI changes criminal affordances because it creates new kinds of harmful content, it amplifies the range of recipients, and it can exploit cognitive vulnerabilities to manipulate user behavior. Given this evolving landscape, the question is whether policies aimed at fighting Generative AI-related harms should include criminal law. The bulk of criminal law scholarship to date would not criminalize AI harms on the theory that AI lacks moral agency. Even so, the field of AI might need criminal law, precisely because it entails a moral responsibility. When a serious harm occurs, responsibility needs to be distributed considering the guilt of the agents involved, and, if it is lacking, it needs to fall back because of their innocence. Thus, legal systems need to start exploring whether and how guilt can be preserved when the actus reus is completely or partially delegated to Generative AI.

The paper examines the legal regulation and governance of “generative artificial intelligence” (AI), “foundation AI,” “large language models” (LLMs), and the “general-purpose” AI models of the AI Act. Attention is drawn to two potential sorcerer’s apprentices, namely, in the spirit of J. W. Goethe’s poem, people who were unable to control a situation they created. Focus is on developers and producers of technologies, such as LLMs that bring about risks of discrimination and information hazards, malicious uses and environmental harms; furthermore, the analysis dwells on the normative attempt of European Union legislators to govern misuses and overuses of LLMs with the AI Act. Scholars, private companies, and organisations have stressed limits of such normative attempt. In addition to issues of competitiveness and legal certainty, bureaucratic burdens and standard development, the threat is the over-frequent revision of the law to tackle advancements of technology. The paper illustrates this threat since the inception of the AI Act and recommends some ways in which the law has not to be continuously amended to address the challenges of technological innovation.

The rapid development of generative artificial intelligence (AI) systems, particularly those fuelled by increasingly advanced large language models, has raised concerns of their potential risks among policymakers globally. In July 2023, Chinese regulators enacted the Interim Measures for the Management of Generative AI Services (“the Measures”). The Measures aim to mitigate various risks associated with public-facing generative AI services, particularly those concerning information content safety and security. China’s approach to regulating AI to date has sought to address the risks associated with rapidly advancing AI technologies while fostering innovation and development. Tensions between these policy objectives are reflected in the provisions of the Measures. As Beijing moves towards establishing a comprehensive legal framework for AI governance, there will be growing interest in how China’s approach may influence AI governance and regulation at a global level.