The European Parliament agreed to close a critical contentious point of the AI Act by adopting the definition used by the Organisation for Economic Cooperation and Development (OECD). Most other definitions have also been agreed upon, with new measures like a right to explanation also on the table of EU lawmakers. Last Friday (3 March), representatives of the European Parliament's political groups working on the AI Act reached a political agreement on one of the most politically sensitive parts of the file, the very definition of Artificial Intelligence, according to two European Parliament officials. The AI Act is a flagship legislative proposal to regulate this emerging technology based on its capacity to cause harm. What is defined as Artificial Intelligence will be highly consequential as it will also define the scope of the EU's AI rulebook.

The use of artificial intelligence (AI) is expanding rapidly. These technological breakthroughs present both opportunity and potential peril. AI technology offers great hope for increasing economic opportunity, boosting incomes, speeding life science research at reduced costs, and simplifying the lives of consumers. With so much potential for innovation, organizations investing in AI-oriented practices are already ramping up initiatives that boost productivity to remain competitive. Like most disruptive technologies, these investments can both create and displace jobs.

For International Women's Day, Sophia Ignatidou, Group Manager for AI and Data Science, discusses how bias can arise in AI, the importance of addressing AI-driven discrimination and how we can all work towards equity in these systems. Her blog also appears on the International Women's Day website. As a woman who also became an immigrant, the concepts of equity and inclusion have always been close to my heart. My career began as a journalist, working for newspapers across both Greece and the UK. I wanted to have a more meaningful impact on the world and in the hope that a career change would enable this, I decided to study international relations and diplomacy.

Beyond the Screen co-founder Frances Haugen discusses the emergence of ChatGPT and the ethical trap of advanced artificial intelligence on'The Claman Countdown.' Billionaire Mark Cuban is telling people to be careful when using artificial intelligence (AI) tools like ChatGPT and DaVinci, cautioning that there are very few guardrails in place to help determine fact from fiction. Cuban joined "The Problem with Jon Stewart," an Apple TV podcast, warning that technology's next "big battle" won't be over who's running operations at Twitter. "It's who controls the AI models and the information that goes in them," Cuban told Stewart in December. "Once these things start taking on a life of their own, and that's the foundation of a ChatGPT, a DaVinci 3.5 taking on a life of its own, so the machine itself will have an influence, and it'll be difficult for us to define why and how the machine makes the decisions that it makes and who controls the machine."

In recent years, Asian countries have stepped up efforts to support the rapid growth of their digital economies. These include measures to equip businesses with the necessary tools and infrastructure to use emerging technologies, such as AI, and support innovation and foster global confidence in them. "We have a belief that being an innovative regulator is not an oxymoron," says Lew Chuen Hong, chief executive of Singapore's Infocomm Media Development Authority (IMDA). "And the real role of the regulator is to build the foundations for trust, so that businesses, governments, and consumers have the trust to innovate and co-create in the digital domain." As AI fast becomes ubiquitous in day-to-day activities, calls for more robust governance to ensure AI systems are fair, transparent, and safe are increasing.

Artificial-intelligence systems that can churn out fluent text, such as OpenAI's ChatGPT, are the newest darlings of the technology industry. But when faced with mathematical queries that require reasoning to answer, these large language models (LLMs) often stumble. A line parallel to y 4x 6 passes through (5, 10). What is the y-coordinate of the point where this line crosses the y-axis? Although LLMs can sometimes answer these types of question correctly, they more often get them wrong. In one early test of its reasoning abilities, ChatGPT scored just 26% when faced with a sample of questions from the'MATH' data set of secondary-school-level mathematical problems1. This is to be expected: given input text, an LLM simply generates new text in accordance with statistical regularities in the words, symbols and sentences that make up the model's training data.

In a world first, Romania's prime minister unveiled a new honorary government adviser that will be joining his team – run entirely on artificial intelligence (AI). The AI is called Ion and consists of a mirror-like surface that displays text as well as at times a male or female face that responds in a calm voice. "Hi, you gave me life and my role is now to represent you, like a mirror," Ion's voice said at the launch. "What should I know about Romania?" The AI-powered adviser was developed by researchers to quickly analyse the opinions of Romanian citizens on key issues and policies.

Following the hype around OpenAI's ChatGPT conversational agent, the last straw in the recent development of Large Language Models (LLMs) that demonstrate emergent unprecedented zero-shot capabilities, we audit the latest OpenAI's GPT-3.5 model, `gpt-3.5-turbo', the first available ChatGPT model, in the LexGLUE benchmark in a zero-shot fashion providing examples in a templated instruction-following format. The results indicate that ChatGPT achieves an average micro-F1 score of 47.6% across LexGLUE tasks, surpassing the baseline guessing rates. Notably, the model performs exceptionally well in some datasets, achieving micro-F1 scores of 62.8% and 70.2% in the ECtHR B and LEDGAR datasets, respectively. The code base and model predictions are available for review on https://github.com/coastalcph/zeroshot_lexglue.

Large language models (LLMs) are used to generate content for a wide range of tasks, and are set to reach a growing audience in coming years due to integration in product interfaces like ChatGPT or search engines like Bing. This intensifies the need to ensure that models are aligned with human preferences and do not produce unsafe, inaccurate or toxic outputs. While alignment techniques like reinforcement learning with human feedback (RLHF) and red-teaming can mitigate some safety concerns and improve model capabilities, it is unlikely that an aggregate fine-tuning process can adequately represent the full range of users' preferences and values. Different people may legitimately disagree on their preferences for language and conversational norms, as well as on values or ideologies which guide their communication. Personalising LLMs through micro-level preference learning processes may result in models that are better aligned with each user. However, there are several normative challenges in defining the bounds of a societally-acceptable and safe degree of personalisation. In this paper, we ask how, and in what ways, LLMs should be personalised. First, we review literature on current paradigms for aligning LLMs with human feedback, and identify issues including (i) a lack of clarity regarding what alignment means; (ii) a tendency of technology providers to prescribe definitions of inherently subjective preferences and values; and (iii) a 'tyranny of the crowdworker', exacerbated by a lack of documentation in who we are really aligning to. Second, we present a taxonomy of benefits and risks associated with personalised LLMs, for individuals and society at large. Finally, we propose a three-tiered policy framework that allows users to experience the benefits of personalised alignment, while restraining unsafe and undesirable LLM-behaviours within (supra-)national and organisational bounds.

While large language models (LLMs) have demonstrated impressive capabilities across tasks in language understanding and interactive decision making, their abilities for reasoning (e.g. chain-of-thought prompting) and acting (e.g. action plan generation) have primarily been studied as separate topics. In this paper, we explore the use of LLMs to generate both reasoning traces and task-specific actions in an interleaved manner, allowing for greater synergy between the two: reasoning traces help the model induce, track, and update action plans as well as handle exceptions, while actions allow it to interface with external sources, such as knowledge bases or environments, to gather additional information. We apply our approach, named ReAct, to a diverse set of language and decision making tasks and demonstrate its effectiveness over state-of-the-art baselines, as well as improved human interpretability and trustworthiness over methods without reasoning or acting components. Concretely, on question answering (HotpotQA) and fact verification (Fever), ReAct overcomes issues of hallucination and error propagation prevalent in chain-of-thought reasoning by interacting with a simple Wikipedia API, and generates human-like task-solving trajectories that are more interpretable than baselines without reasoning traces. On two interactive decision making benchmarks (ALFWorld and WebShop), ReAct outperforms imitation and reinforcement learning methods by an absolute success rate of 34% and 10% respectively, while being prompted with only one or two in-context examples. Project site with code: https://react-lm.github.io