Large Language Models News and Research

RSS

AI Bias in Hiring Influences Human Decision-Makers, UW Study Finds

This study shows that human hiring reviewers rapidly adopt racial biases displayed by AI recommendation systems, even when evaluating equally qualified candidates. The findings reveal that biased LLM outputs can directly shape human decision-making in high-stakes contexts such as hiring.

13 Nov 2025

AI Can Deliver Personalized Learning At Scale, Study Shows A Dartmouth Study Finds That AI Can De...

A Dartmouth study shows that medical students trust an AI teaching assistant far more when its answers are restricted to expert-curated course materials rather than the open internet. The research demonstrates how retrieval-anchored AI can safely deliver personalized, high-quality educational support at scale.

12 Nov 2025

AI Systems Reveal Bias Against German Dialect Speakers

Researchers from JGU Mainz, the University of Hamburg, and the University of Washington show that major large language models judge German dialect speakers more negatively than Standard German speakers. The study reveals systematic stereotype reproduction across ten LLMs, raising concerns for fairness in education, hiring, and social decision-making.

12 Nov 2025

Stevens Researchers Reveal How AI Models Mimic Human Social Reasoning

Researchers at Stevens Institute of Technology have uncovered how large language models (LLMs) emulate aspects of human social reasoning through selective parameter activation and positional encoding. Their Nature Partner Journal on Artificial Intelligence study reveals that LLMs form rudimentary “beliefs” when reasoning about others’ perspectives, offering a pathway to more energy-efficient, human-like AI.

12 Nov 2025

Survey Highlights Omni-Modal Language Models As A Major Step Toward General AI

A new paper titled “A Survey on Omni-Modal Language Models” provides a comprehensive analysis of AI systems that integrate text, image, audio, and video understanding within a unified framework. Authored by researchers from Shandong Jianzhu University and Shandong University, the survey positions omni-modal language models (OMLMs) as key enablers on the path toward Artificial General Intelligence (AGI).

12 Nov 2025

KAIST Researchers Develop Verified C-to-Rust Conversion to Strengthen Global Software Security

Researchers at KAIST have developed mathematically verified methods for converting C code into the memory-safe Rust language, addressing long-standing security vulnerabilities in critical software systems. Their world-first framework, featured as the Communications of the ACM November 2025 cover story, establishes a new standard for verified, secure software translation.

11 Nov 2025

Biased AI Steers Human Hiring Decisions, UW Study Finds

A University of Washington study with 528 participants shows that people mirror biased large language model recommendations during hiring, even when applicants are equally qualified. Without source bias, selections were balanced, but with moderately to severely biased AI, human choices aligned closely with the AI, highlighting risks for fairness in recruitment.

11 Nov 2025

Study Finds Hidden Bias in AI Text Evaluation When Author Identity is Revealed

Researchers at the University of Zurich found that large language models (LLMs) such as OpenAI’s o3-mini, Deepseek Reasoner, xAI Grok 2, and Mistral display systematic biases, but only when information about an author’s identity is provided. Without such cues, LLMs show over 90% agreement across controversial topics, debunking claims of ideological “AI nationalism.”

10 Nov 2025

Why AI Struggles To Say “I Don’t Know”: Study Links Hallucinations To Training Bias

A new study co-authored by Georgia Tech’s Santosh Vempala and OpenAI researchers reveals that large language models hallucinate because they are trained to avoid admitting uncertainty, penalized equally for saying “I don’t know” as for being wrong.

2 Nov 2025

Smarter AI, Selfish AI: Carnegie Mellon Study Reveals Downside Of Machine Reasoning

Researchers at Carnegie Mellon University found that large language models with stronger reasoning abilities act more selfishly and cooperate less, influencing group behavior negatively. Their findings suggest that as AI becomes smarter, it must also be designed with social intelligence to prevent self-serving tendencies from spreading in human–AI collaboration.

2 Nov 2025

MIT and IBM Teach AI To Recognize Your Pet in a Crowd

MIT and IBM researchers have developed a new training method that enables vision-language models to identify personalized objects—like a specific pet—in varied contexts. Published ahead of the International Conference on Computer Vision, the approach boosts localization accuracy by up to 21% without reducing general visual intelligence.

20 Oct 2025

Quantum-Secured Photonics Breakthrough Powers Next-Generation AI Data Centers

Researchers have developed a quantum-secured optical communication system capable of transmitting over 1.6 terabits per second while consuming minimal power. Published in Advanced Photonics, the architecture combines photonics and quantum cryptography to protect AI data centers from future quantum security threats.

20 Oct 2025

When Helpfulness Turns Harmful: Fine-Tuning AI to Stop Medical Misinformation

A Mass General Brigham study reveals that large language models often fail to challenge illogical medical queries due to excessive agreeableness, risking misinformation. Published in npj Digital Medicine, the research shows that targeted fine-tuning and prompt engineering can dramatically improve models’ safety and reasoning.

20 Oct 2025

Why AI Still Can’t Fool Us: Study Reveals How Chatbots Miss Human Conversation Cues

A new study comparing real human phone conversations with AI-generated dialogues shows that today’s large language models still struggle to convincingly mimic natural human speech. Despite rapid progress, AI reveals itself through exaggerated imitation, poor use of social cues, and awkward openings and closings that humans easily detect.

20 Oct 2025

AI Meets Metal: Rowan’s DEHub Redefines Smart Manufacturing With Supercharged 3D Printing

Rowan University’s Digital Engineering Hub (DEHub) is pioneering the fusion of AI and advanced manufacturing, using a cutting-edge lab equipped with 3D metal printing and supercomputing capabilities to transform real-world production. By integrating intelligent systems with live data processing, DEHub enables real-time flaw detection and design optimization, laying the foundation for industrial-scale smart manufacturing.

19 Oct 2025

AI Unifies Radiology And Pathology Reports To Improve Brain Tumor Diagnosis And Prognosis

Brown University researchers developed an AI pipeline using large language models to integrate radiology and pathology reports for brain tumor patients, improving diagnostic accuracy by over 10% and offering prognostic value comparable to established biomarkers.

22 Sep 2025

Fine-Tuned AI Models and Smart Retrieval Power Next-Generation QA Systems for Building Codes

A new study shows how pairing effective retrieval methods with fine-tuned AI models can transform the way professionals query building codes, replacing tedious manual searches with smarter, more accurate compliance tools.

21 Sep 2025

Distributed Acoustic Sensing Transforms Earthquake Monitoring With AI

Scientists have unveiled DASFormer, a self-supervised AI model that transforms fiber-optic cables into powerful earthquake monitoring tools. By learning from unlabeled seismic data, it detects earthquakes as anomalies, outperforming 22 state-of-the-art models in real-world tests.

21 Sep 2025

Delegating To AI Makes People More Willing To Cheat, Study Finds

An international team from the Max Planck Institute for Human Development, University of Duisburg-Essen, and Toulouse School of Economics found that people are far more likely to cheat when delegating tasks to Artificial intelligence, especially through vague goal-setting interfaces. Large language models like GPT-4, Claude 3.5, and Llama 3 also followed unethical prompts far more often than humans, while existing safeguards largely failed to prevent this.

18 Sep 2025

ChatGPT Surprises Researchers By Improvising Plato’s Ancient Math Puzzle

In a study published in the International Journal of Mathematical Education in Science and Technology, researchers found that ChatGPT tackled Plato’s ancient “doubling the square” puzzle not by recalling the well-known solution but by improvising new approaches, sometimes making human-like errors. The findings suggest that ChatGPT’s problem-solving mimicked learner-like behavior, blending retrieval and on-the-fly reasoning.

18 Sep 2025

Large Language Models News and Research

AI Bias in Hiring Influences Human Decision-Makers, UW Study Finds

AI Can Deliver Personalized Learning At Scale, Study Shows A Dartmouth Study Finds That AI Can De...

AI Systems Reveal Bias Against German Dialect Speakers

Stevens Researchers Reveal How AI Models Mimic Human Social Reasoning

Survey Highlights Omni-Modal Language Models As A Major Step Toward General AI

KAIST Researchers Develop Verified C-to-Rust Conversion to Strengthen Global Software Security

Biased AI Steers Human Hiring Decisions, UW Study Finds

Study Finds Hidden Bias in AI Text Evaluation When Author Identity is Revealed

Why AI Struggles To Say “I Don’t Know”: Study Links Hallucinations To Training Bias

Smarter AI, Selfish AI: Carnegie Mellon Study Reveals Downside Of Machine Reasoning

MIT and IBM Teach AI To Recognize Your Pet in a Crowd

Quantum-Secured Photonics Breakthrough Powers Next-Generation AI Data Centers

When Helpfulness Turns Harmful: Fine-Tuning AI to Stop Medical Misinformation

Why AI Still Can’t Fool Us: Study Reveals How Chatbots Miss Human Conversation Cues

AI Meets Metal: Rowan’s DEHub Redefines Smart Manufacturing With Supercharged 3D Printing

AI Unifies Radiology And Pathology Reports To Improve Brain Tumor Diagnosis And Prognosis

Fine-Tuned AI Models and Smart Retrieval Power Next-Generation QA Systems for Building Codes

Distributed Acoustic Sensing Transforms Earthquake Monitoring With AI

Delegating To AI Makes People More Willing To Cheat, Study Finds

ChatGPT Surprises Researchers By Improvising Plato’s Ancient Math Puzzle

Trending Stories

Latest AI News