Chain-of-thought prompting (e.g., "Let's think step-by-step") primes large language models to verbalize rationalization for their predictions. While chain-of-thought can lead to dramatic performance gains, benefits appear to emerge only for sufficiently large models (beyond 50B parameters). We show that orders-of-magnitude smaller models (125M -- 1.3B parameters) can still benefit from chain-of-thought prompting. To achieve this, we introduce Symbolic Chain-of-Thought Distillation (SCoTD), a method to train a smaller student model on rationalizations sampled from a significantly larger teacher model. Experiments across several commonsense benchmarks show that: 1) SCoTD enhances the performance of the student model in both supervised and few-shot settings, and especially for challenge sets; 2) sampling many reasoning chains per instance from the teacher is paramount; and 3) after distillation, student chain-of-thoughts are judged by humans as comparable to the teacher, despite orders of magnitude fewer parameters. We test several hypotheses regarding what properties of chain-of-thought samples are important, e.g., diversity vs. teacher likelihood vs. open-endedness. We release our corpus of chain-of-thought samples and code.

Earlier this week, Senate majority leader Chuck Schumer unveiled his SAFE Innovation Framework for artificial intelligence (AI), calling on Congress to take swift, decisive action. Leaders in the AI industry have been calling out for regulation. But Schumer's proposal reveals how difficult it could be in practice for policymakers to regulate a technology that even experts struggle to fully understand. The SAFE Innovation Framework has a number of policy goals: make sure AI systems are secure against cyber attacks, protect jobs, ensure accountability for those deploying AI systems, and defend U.S. democratic values, all without stifling innovation. The part of Schumer's framework which comes closest to making a concrete policy proposal, rather than setting a policy goal, is his call for explainability.

It's something that's drilled into you from the first essay you write in school: Always check your sources. Yet, New York attorney Steven Schwartz relied on ChatGPT to find and review them for him -- a decision that's led a judge to issue a $5,000 fine to him, his associate Peter LoDuca and their law firm Levidow, Levidow and Oberman, The Guardian reports. Schwartz used it for a case in which a man was suing Colombian airline Avianca alleging he was injured on a flight to New York City. In this case, ChatGPT produced six cases as precedent, such as "Martinez v. Delta Airlines" and "Miller v. United Airlines," that were either inaccurate or simply didn't exist. In the decision to fine Schwartz and co., Judge P Kevin Castel explained, "Technological advances are commonplace and there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings."

A US judge has fined two lawyers and a law firm $5,000 (£3,935) after fake citations generated by ChatGPT were submitted in a court filing. A district judge in Manhattan ordered Steven Schwartz, Peter LoDuca and their law firm Levidow, Levidow & Oberman to pay the fine after fictitious legal research was used in an aviation injury claim. Schwartz had admitted that ChatGPT, a chatbot that churns out plausible text responses to human prompts, invented six cases he referred to in a legal brief in a case against the Colombian airline Avianca. The judge P Kevin Castel said in a written opinion there was nothing "inherently improper" about using artificial intelligence for assisting in legal work, but lawyers had to ensure their filings were accurate. "Technological advances are commonplace and there is nothing inherently improper about using a reliable artificial intelligence tool for assistance," Castel wrote.

Center for AI Safety Director Dan Hendrycks explains concerns about how the rapid growth of artificial intelligence could impact society. When Senate Majority Leader Chuck Schumer, D-N.Y., announced a "major effort" in April to put the Senate's imprint on artificial intelligence policy, he talked about having an "urgency to act" and said a legislative plan would start taking shape in a matter of weeks. "In the coming weeks, Leader Schumer plans to refine the proposal in conjunction with stakeholders from academia, advocacy organizations, industry, and the government," he said in an April 13 statement. But on Thursday, more than two months later, Schumer indicated that legislation may not be ready until 2024. In Wednesday remarks to the Center for Strategic and International Studies, Schumer said the process of getting input for the plan is still months away.

There is a data-driven revolution under way in science and society, disrupting every form of enterprise. We are collecting and storing data more rapidly than ever before. The value of data as a central asset in an organization is now well established and generally accepted. The Economist called data "the world's most valuable resource."40 The World Economic Forum's briefing paper, A New Paradigm for Business of Data, states "At the heart of digital economy and society is the explosion of insight, intelligence and information--data."5 The field of data science is expected to enable data to be leveraged for making better decisions and achieving more meaningful outcomes. Although the term data science has some history, in its current incarnation as a modern field of study, it has already had significant economic impact. A 2015 Organisation for Economic Co-operation and Development (OECD) report identified "data-driven innovation" (DDI) as having a central driving role in 21st century economies, defining DDI as "the use of data and analytics to improve and foster new products, processes, organisational methods and markets." Data science deployments are still what might be called first generation, but their impact is already being felt in many areas: global sustainability,11 power and energy systems,25 biological and biomedical systems,38 health sciences and health informatics,12 finance and insurance,8 smart cities,33 digital humanities,28 and more. The last decade has established the terms "big data," "data analytics," and "data science" into our lexicon, both as buzzwords and as important fields of study. Interest in the topic, as evidenced by Google Trends (see Figure 1), has exploded over the same period. An increasing number of countries have released policy statements related to data science. In academia, data-science programs and research institutes have been established with significant speed, while many industrial organizations have created data-science units.

Generative artificial intelligence (AI) has captured considerable popular attention recently. ChatGPT and DALL-E have given members of the general public opportunities to use AI systems to generate text and image outputs for fun and a wide range of other purposes. Google and Meta have announced their intentions to launch similar AI systems soon.

The rise of large-scale socio-technical systems in which humans interact with artificial intelligence (AI) systems (including assistants and recommenders, in short AIs) multiplies the opportunity for the emergence of collective phenomena and tipping points, with unexpected, possibly unintended, consequences. For example, navigation systems' suggestions may create chaos if too many drivers are directed on the same route, and personalised recommendations on social media may amplify polarisation, filter bubbles, and radicalisation. On the other hand, we may learn how to foster the "wisdom of crowds" and collective action effects to face social and environmental challenges. In order to understand the impact of AI on socio-technical systems and design next-generation AIs that team with humans to help overcome societal problems rather than exacerbate them, we propose to build the foundations of Social AI at the intersection of Complex Systems, Network Science and AI. In this perspective paper, we discuss the main open questions in Social AI, outlining possible technical and scientific challenges and suggesting research avenues.

Deductive coding is a widely used qualitative research method for determining the prevalence of themes across documents. While useful, deductive coding is often burdensome and time consuming since it requires researchers to read, interpret, and reliably categorize a large body of unstructured text documents. Large language models (LLMs), like ChatGPT, are a class of quickly evolving AI tools that can perform a range of natural language processing and reasoning tasks. In this study, we explore the use of LLMs to reduce the time it takes for deductive coding while retaining the flexibility of a traditional content analysis. We outline the proposed approach, called LLM-assisted content analysis (LACA), along with an in-depth case study using GPT-3.5 for LACA on a publicly available deductive coding data set. Additionally, we conduct an empirical benchmark using LACA on 4 publicly available data sets to assess the broader question of how well GPT-3.5 performs across a range of deductive coding tasks. Overall, we find that GPT-3.5 can often perform deductive coding at levels of agreement comparable to human coders. Additionally, we demonstrate that LACA can help refine prompts for deductive coding, identify codes for which an LLM is randomly guessing, and help assess when to use LLMs vs. human coders for deductive coding. We conclude with several implications for future practice of deductive coding and related research methods.

Despite recent efforts by the Artificial Intelligence (AI) community to move towards standardised procedures for documenting models, methods, systems or datasets, there is currently no methodology focused on use cases aligned with the risk-based approach of the European AI Act (AI Act). In this paper, we propose a new framework for the documentation of use cases, that we call "use case cards", based on the use case modelling included in the Unified Markup Language (UML) standard. Unlike other documentation methodologies, we focus on the intended purpose and operational use of an AI system. It consists of two main parts. Firstly, a UML-based template, tailored to allow implicitly assessing the risk level of the AI system and defining relevant requirements. Secondly, a supporting UML diagram designed to provide information about the system-user interactions and relationships. The proposed framework is the result of a co-design process involving a relevant team of EU policy experts and scientists. We have validated our proposal with 11 experts with different backgrounds and a reasonable knowledge of the AI Act as a prerequisite. We provide the 5 "use case cards" used in the co-design and validation process. "Use case cards" allows framing and contextualising use cases in an effective way, and we hope this methodology can be a useful tool for policy makers and providers for documenting use cases, assessing the risk level, adapting the different requirements and building a catalogue of existing usages of AI.