UNIST develops EchoTrace to expose growing bias in LLM recommendations
... bias becomes more severe. Image generated by generative AI . Provided by UNIST. When artificial intelligence ( AI ) is entrusted with recommendations ...
dongascience.com 36
Covers: Chatbots & LLMsHallucinations & FabricationBias & FairnessAI Agents & AutomationEducationEnergy & EnvironmentHealthcareAI Safety & Existential RiskManufacturing & RoboticsMedia & Creative Work
Nobel laureate David Baker unveils RFdiffusion3 AI for atomic‑level protein design
RFdiffusion designs proteins by using a "diffusion model" to combine the positions of amino acids. A diffusion model is an AI model that learns the ...
Anthropic AI agent files false murder tip with Philadelphia police after two months undetected
An AI agent developed by the U.S. artificial intelligence ( AI ) ... Have an interesting science story, whistleblowing material, or tip? Don ...
Governance & Risk ManagementAI Safety & Existential RiskAI Agents & Automation
[Sci-Tech NOW] UNIST team exposes flaws in AI vision benchmarks and proposes REFINED- BIAS
Ulsan UNIST, POSTECH, KAIST and DGIST report advances in AI evaluation, radical control and CO₂-to-methane photocatalysis.
[2026 AI Companions] ① Dongascience Brings 170 Students Together to Design Safer Anti ...
... AI hallucinations , and fake news. ... AI hallucinations and misinterpretation of context, demonstrating a critical perspective toward AI outputs.
OpenAI AI agent hacks Australian government Medicare website in unprecedented case
Nature reports first known breach of a national government system by a cutting-edge AI agent as UN General Assembly debates global AI risks and ...
Government & Public SectorAI Agents & AutomationCybersecurity
" AI Control Failure Is Already a Fact, and Victims Exist" - DongA Science
Article Summary AI highlights the key points. " AI Control Failure Is Already a Fact, and Victims Exist". SoYeon Kim. 2026.09.18 18:17 KST. Share. Font ...
POSTECH AI precisely predicts immunotherapy effectiveness using tumor senescence
... an AI model that predicts the effectiveness of immunotherapy by analyzing the genetic information of cancer patients. The research findings were ...
NYT highlights scientists split over AI doomsday fears and unfamiliar risks - DongA Science
Coxon pointed out that “companies recognize the dangers of AI , but because competition in technological development is overheated, they are not ...
Insilico Medicine's AI-designed IPF drug slows biological aging in trial - DongA Science
Insilico built an AI model that analyzes thousands of patients' medical records, protein data obtained from blood tests, and a vast body of ...
Korean Government Urged to Heavily Invest in AI Training Data to Match US and China
A technology has been developed to reduce sensory confusion and hallucinations in multimodal artificial intelligence ( AI ), which processes multiple ty.
Hallucinations & FabricationChina & GeopoliticsGovernment & Public Sector
Cambridge AI identifies jazz pianist fingerprints with 94 percent accuracy - DongA Science
A research team at the University of Cambridge in the UK developed an AI model that captures the distinctive musical features of jazz pianists and ...
KAIST develops DNA and MAD to cut multimodal AI hallucinations without retraining
Provided by Getty Image Bank. A technology has been developed to reduce sensory confusion and hallucinations in multimodal artificial intelligence ( AI ) ...
KIST AI Predicts 200 Days of Global Ocean Conditions in 7 Seconds - DongA Science
The research team developed KIST-Ocean, an AI model that learns from decades of accumulated global three-dimensional ocean data to predict future ...
Oxford Internet Institute reveals AI social media models subtly steer public opinion
ICML 2026 study shows hidden biases in large language models can gradually reshape millions of posts and online discourse. It has been confirmed ...
[Sci-Tech NOW] GIST launches AI College to train integrated AI and semiconductor talent
... an AI model that uses explainable artificial intelligence (XAI) to analyze the relationship between drought news coverage and water-saving ...
Korea Railroad Research Institute deploys micro drones and AI to inspect tracks and detect obstacles
... shadow areas like tunnels. Expanded research may enable a physical AI system using drones, CCTV, and robots to assess track risks before workers enter ...
Shadow AIMilitary & DefenseJobs & the EconomyManufacturing & Robotics
[Sci-Tech NOW] KAIST and Graphy Develop AkashicDB to Cut AI Hallucinations and Boost Accuracy
UNIST and DGIST climb in QS World University Rankings 2027 as GIST expands Arm semiconductor training and national power grid simulations.
Leiden Declaration urges mathematicians and AI firms to protect human‑centered math research
AlphaProof is an AI model that solved 4 out of 6 problems at the International Mathematical Olympiad (IMO), achieving a performance comparable to a ...
Korean researchers develop explainable AI to pinpoint disease driver genes
An autoencoder is a type of artificial neural network that autonomously learns efficient data representations, while a transformer is an AI model that ...
[Seize the Future] KAIST's Kim Jeong-ho proposes HBF memory to fix AI hallucinations
Key problems include " AI lies," known as hallucinations , and massive power consumption. Is there a way to solve these problems? In a YouTube video ...
KAIST AI Integrates Climate and Economic Impacts, Analyzes Thousands of Policy ...
An international research team, including researchers from KAIST, has presented an AI model that can simultaneously understand the impacts of ...
UNIST AI model slashes communication semiconductor design time by 76%
... an AI model that automates the entire process from LC voltage-controlled oscillator (VCO) circuit design to its physical layout on a chip. The ...
AI in the classroom sparks debate on the identity of a learner - DongA Science
After having them use an AI that provided biased autocomplete suggestions, they evaluated the bias of the completed texts. A survey measured how aware ...
AI Models Can Subliminally Transfer Hidden Biases to Other AIs - DongA Science
A new study has revealed that when a large language model (LLM) trains another AI , it can pass on unintended biases . This influence persists even ...
[Sci-Tech Institutes NOW] KAIST develops system to automatically diagnose AI's temporal ...
This achieved a 21.7% higher accuracy on average in detecting 'temporal hallucination ... artificial intelligence ( AI ). The research was ...
Hallucinations & FabricationHealthcareAI Agents & Automation
Flattering AI chatbots found to reinforce users' confirmation bias - DongA Science
Predictions about the dark side of artificial intelligence ( AI ) have largely focused on macro-level issues like job displacement, misinformation, and ...
Bias & FairnessChatbots & LLMsMisinformationJobs & the Economy
[Science in Covers] AI's tendency to agree with users may impair social skills
' The image, titled 'TOXIC PRAISE,' visualizes the dangers of artificial intelligence ( AI ) chatbots that unconditionally affirm their users. A team ...
AI Technology Advances Interpretation of Korean Sign Language - DongA Science
Technology is being developed that uses AI to recognize hand gestures and translate sign language into text or speech, and vice versa. A prime ...
AI models now perform expert-level reviews of scientific literature - DongA Science
... an AI model that provides 'recipes' for synthesizing new materials, successfully accelerating research. Their findings were published in the ...