Searcharxiv⌕ Search

arXiv subjects

Ali Unlu

Publications and source records attributed to Ali Unlu.

4 recordsLinked to original sources
arXiv2025

The Evolving Media Discourse on ChatGPT and Higher Education

This research full paper examines how news media have been instrumental in creating specific narratives about generative AI applications, especially ChatGPT, in higher education, and how these narratives have changed over time. The introduction of emerging technologies in higher education is driven not only by their technological affordances but also by the narratives built around their perceived value, risks, and possibilities. Therefore, understanding how news media narratives contribute to sociotechnical imaginaries - the imagined futures of technology use that institutions and educators inherit - is important for evaluating ChatGPT's role in teaching and learning, including engineering education. Through temporal and sentiment analyses of 198 U.S. news articles from November 2022 to October 2024, we traced the evolving narratives surrounding generative AI and the use of ChatGPT in higher education. We found that the media discourse largely centered on institutional responses, with policy changes and teaching practices showing the most consistent presence and positive sentiment over time. Conversely, coverage of topics such as human-centered learning, the job market, and skill development appeared more sporadic, with initially uncertain portrayals gradually shifting toward cautious optimism. Media sentiment toward ChatGPT's role in college admissions remained predominantly negative. Our findings suggest that media narratives prioritize institutional responses to generative AI over the long-term, broader ethical, social, and labor-related implications. This imbalance is especially relevant to engineering and computing education, where students must be prepared not only to use AI tools but also to critically evaluate the broader sociotechnical consequences of AI as future designers of AI-enabled technologies.

cs.CY↗
arXiv2025

The Narrative Construction of Generative AI Efficacy by the Media: A Case Study of the Role of ChatGPT in Higher Education

The societal role of technology, including artificial intelligence (AI), is often shaped by sociocultural narratives. This study examines how U.S. news media construct narratives about the efficacy of generative AI (GenAI), using ChatGPT in higher education as a case study. Grounded in Agenda Setting Theory, we analyzed 198 articles published between November 2022 and October 2024, employing LDA topic modeling and sentiment analysis. Our findings identify six key topics in the media discourse, with sentiment analysis revealing generally positive portrayals of ChatGPT's integration into higher education through policy, curriculum, teaching practices, collaborative decision-making, skill development, and human-centered learning. In contrast, media narratives express more negative sentiment regarding their impact on entry-level jobs and college admissions. This research highlights how media coverage can influence public perceptions of GenAI in education and provides actionable insights for policymakers, educators, and AI developers navigating its adoption and representation in public discourse.

cs.CY↗
arXiv2021

Variational Laplace for Bayesian neural networks

We develop variational Laplace for Bayesian neural networks (BNNs) which exploits a local approximation of the curvature of the likelihood to estimate the ELBO without the need for stochastic sampling of the neural-network weights. The Variational Laplace objective is simple to evaluate, as it is (in essence) the log-likelihood, plus weight-decay, plus a squared-gradient regularizer. Variational Laplace gave better test performance and expected calibration errors than maximum a-posteriori inference and standard sampling-based variational inference, despite using the same variational approximate posterior. Finally, we emphasise care needed in benchmarking standard VI as there is a risk of stopping before the variance parameters have converged. We show that early-stopping can be avoided by increasing the learning rate for the variance parameters.

stat.ML↗
arXiv2020

Variational Laplace for Bayesian neural networks

We develop variational Laplace for Bayesian neural networks (BNNs) which exploits a local approximation of the curvature of the likelihood to estimate the ELBO without the need for stochastic sampling of the neural-network weights. The Variational Laplace objective is simple to evaluate, as it is (in essence) the log-likelihood, plus weight-decay, plus a squared-gradient regularizer. Variational Laplace gave better test performance and expected calibration errors than maximum a-posteriori inference and standard sampling-based variational inference, despite using the same variational approximate posterior. Finally, we emphasise care needed in benchmarking standard VI as there is a risk of stopping before the variance parameters have converged. We show that early-stopping can be avoided by increasing the learning rate for the variance parameters.

stat.ML↗