Synthese, Год журнала: 2025, Номер 205(3)
Опубликована: Март 12, 2025
Язык: Английский
Опубликована: Дек. 22, 2023
Recent advancements in artificial intelligence (AI) and specifically generative AI (GenAI) are threatening to fundamentally reshape computing society. Largely driven by large language models (LLMs), many tools now able interpret generate both natural instructions source code. These capabilities have sparked urgent questions the education community around how educators should adapt their pedagogy address challenges leverage opportunities presented this new technology. In working group report, we undertake a comprehensive exploration of context make five significant contributions. First, provide detailed review literature on LLMs synthesise findings from 71 primary articles, nearly 80% which been published first 8 months 2023. Second, report survey students instructors across 20 countries, capturing prevailing attitudes towards GenAI/LLMs use contexts. Third, understand is already changing, offer insights collected in-depth interviews with 22 continents. Fourth, ACM Code Ethics frame discussion ethical issues raised education, concrete advice for policy makers, educators, students. Finally, benchmark performance several current GenAI models/tools various datasets, highlight extent rapidly improving.
Язык: Английский
Процитировано
131
Public Choice, Год журнала: 2023, Номер 198(1-2), С. 3 - 23
Опубликована: Авг. 17, 2023
Abstract We investigate the political bias of a large language model (LLM), ChatGPT, which has become popular for retrieving factual information and generating content. Although ChatGPT assures that it is impartial, literature suggests LLMs exhibit involving race, gender, religion, orientation. Political in can have adverse electoral consequences similar to from traditional social media. Moreover, be harder detect eradicate than gender or racial bias. propose novel empirical design infer whether biases by requesting impersonate someone given side spectrum comparing these answers with its default. also dose-response, placebo, profession-politics alignment robustness tests. To reduce concerns about randomness generated text, we collect same questions 100 times, question order randomized on each round. find robust evidence presents significant systematic toward Democrats US, Lula Brazil, Labour Party UK. These results translate into real general, extend even amplify existing challenges processes posed Internet Our findings important implications policymakers, media, politics, academia stakeholders.
Язык: Английский
Процитировано
105
Electronic Markets, Год журнала: 2023, Номер 33(1)
Опубликована: Дек. 1, 2023
Abstract Recent developments in the field of artificial intelligence (AI) have enabled new paradigms machine processing, shifting from data-driven, discriminative AI tasks toward sophisticated, creative through generative AI. Leveraging deep models, is capable producing novel and realistic content across a broad spectrum (e.g., texts, images, or programming code) for various domains based on basic user prompts. In this article, we offer comprehensive overview fundamentals with its underpinning concepts prospects. We provide conceptual introduction to relevant terms techniques, outline inherent properties that constitute AI, elaborate potentials challenges. underline necessity researchers practitioners comprehend distinctive characteristics order harness potential while mitigating risks contribute principal understanding.
Язык: Английский
Процитировано
97
Global Journal of Flexible Systems Management, Год журнала: 2023, Номер 24(4), С. 659 - 689
Опубликована: Сен. 28, 2023
Язык: Английский
Процитировано
85
Опубликована: Май 11, 2024
In our era of rapid technological advancement, the research landscape for writing assistants has become increasingly fragmented across various communities. We seek to address this challenge by proposing a design space as structured way examine and explore multidimensional intelligent interactive assistants. Through large community collaboration, we five aspects assistants: task, user, technology, interaction, ecosystem. Within each aspect, define dimensions (i.e., fundamental components an aspect) codes potential options dimension) systematically reviewing 115 papers. Our aims offer researchers designers practical tool navigate, comprehend, compare possibilities assistants, aid in envisioning new
Язык: Английский
Процитировано
16
Nature Human Behaviour, Год журнала: 2024, Номер unknown
Опубликована: Окт. 28, 2024
Abstract Inspired by the increasing use of artificial intelligence (AI) to augment humans, researchers have studied human–AI systems involving different tasks, and populations. Despite such a large body work, we lack broad conceptual understanding when combinations humans AI are better than either alone. Here addressed this question conducting preregistered systematic review meta-analysis 106 experimental studies reporting 370 effect sizes. We searched an interdisciplinary set databases (the Association for Computing Machinery Digital Library, Web Science Information Systems eLibrary) published between 1 January 2020 30 June 2023. Each study was required include original human-participants experiment that evaluated performance alone, alone combinations. First, found that, on average, performed significantly worse best or (Hedges’ g = −0.23; 95% confidence interval, −0.39 −0.07). Second, losses in tasks involved making decisions greater gains creating content. Finally, outperformed combination, but losses. Limitations evidence assessed here possible publication bias variations designs analysed. Overall, these findings highlight heterogeneity effects collaboration point promising avenues improving systems.
Язык: Английский
Процитировано
15
SSRN Electronic Journal, Год журнала: 2023, Номер unknown
Опубликована: Янв. 1, 2023
A standing issue is how to measure bias in Large Language Models (LLMs) like ChatGPT. We devise a novel method of sampling, bootstrapping, and impersonation that addresses concerns about the inherent randomness LLMs test if it can capture political Our results indicate that, by default, ChatGPT aligned with Democrats US. Placebo tests our are due bias, not noise or spurious relationships. Robustness show findings valid also for Brazil UK, different professions, numerical scales questionnaires.
Язык: Английский
Процитировано
27
Transactions of the Association for Computational Linguistics, Год журнала: 2023, Номер 11, С. 1643 - 1668
Опубликована: Янв. 1, 2023
Abstract Natural language generation has witnessed significant advancements due to the training of large models on vast internet-scale datasets. Despite these advancements, there exists a critical challenge: These can inadvertently generate content that is toxic, inaccurate, and unhelpful, existing automatic evaluation metrics often fall short identifying shortcomings. As become more capable, human feedback an invaluable signal for evaluating improving models. This survey aims provide overview recent research leveraged improve natural generation. First, we introduce taxonomy distilled from categorize organize varied forms feedback. Next, discuss how be described by its format objective, cover two approaches proposed use (either or decoding): directly using We also datasets human-feedback data collection, concerns surrounding collection. Finally, nascent field AI feedback, which uses make judgments based set principles minimize need intervention. release website this at feedback-gap-survey.info.
Язык: Английский
Процитировано
25
Creativity and Cognition, Год журнала: 2024, Номер unknown, С. 413 - 425
Опубликована: Июнь 22, 2024
Large language models (LLMs) are now being used in a wide variety of contexts, including as creativity support tools (CSTs) intended to help their users come up with new ideas. But do LLMs actually user creativity? We hypothesized that the use an LLM CST might make LLM's feel more creative, and even broaden range ideas suggested by each individual user, but also homogenize different users. conducted 36-participant comparative study found, accordance homogenization hypothesis, tended produce less semantically distinct ChatGPT than alternative CST. Additionally, generated greater number detailed ideas, felt responsible for they generated. discuss potential implications these findings users, designers, developers LLM-based CSTs.
Язык: Английский
Процитировано
13
AI and Ethics, Год журнала: 2024, Номер unknown
Опубликована: Май 7, 2024
Abstract Our interdisciplinary study examines the effectiveness of US law in addressing complex challenges posed by generative AI systems to fundamental human values, including physical and mental well-being, privacy, autonomy, diversity, equity. Through analysis diverse hypothetical scenarios developed collaboration with experts, we identified significant shortcomings ambiguities within existing legal protections. Constitutional civil rights currently struggles hold companies responsible for AI-assisted discriminatory outputs. Moreover, even without considering liability shield provided Section 230, laws may not effectively remedy unintentional intangible harms caused systems. Demonstrating causal links claims such as defamation or product proves exceptionally difficult due intricate opaque nature these To address unique evolving risks AI, propose a “Responsible Legal Framework” that adapts recognize new threats utilizes multi-pronged approach. This framework would enshrine values frameworks, establish comprehensive safety guidelines, implement models tailored complexities human-AI interactions. By proactively mitigating unforeseen like health impacts privacy breaches, this aims create landscape capable navigating exciting yet precarious future brought forth technologies.
Язык: Английский
Процитировано
11