Large Language Models News and Research

RSS

AI Moderation Systems Show Striking Inconsistencies In Detecting Hate Speech

This paper by Yphtach Lelkes and Neil Fasching is the first large-scale comparative analysis of seven major AI content moderation systems, evaluating their consistency in detecting hate speech across 1.3 million synthetic sentences. The findings reveal substantial inconsistencies, with some systems flagging identical content as harmful while others deem it acceptable, especially for certain demographic groups.

14 Sep 2025

AI Can Read Emissions Reports But Can You Trust What It Says?

Researchers created a gold-standard dataset to benchmark AI systems that extract greenhouse gas emissions data from corporate sustainability reports. The study reveals widespread data quality issues, urging caution in using LLMs without verification.

11 Sep 2025

Researchers Warn: AI Could Automate Every Stage of Ransomware

Researchers at NYU Tandon have developed "Ransomware 3.0," a proof-of-concept showing how criminals could use large language models to autonomously execute ransomware attacks. The AI system carried out every phase of an attack in simulations, highlighting future threats and the urgent need for defensive countermeasures.

10 Sep 2025

AI System Translates and Edits Human Thoughts Into Images Using Brain Scans

Researchers have developed DreamConnect, a dual-stream AI system that interprets and edits visual mental imagery directly from fMRI signals using natural language prompts. This novel framework allows users to reshape imagined scenes, bridging brain activity and interactive visualization.

3 Sep 2025

EnvGPT Transforms Environmental Research By Matching Bigger AI Models With Compact Precision

Researchers developed EnvGPT, a fine-tuned language model trained on environmental science data, to address gaps left by general-purpose LLMs. It outperformed similar-sized models on benchmarks and neared the performance of much larger AI systems.

1 Sep 2025

AI Gives Video Game Characters Real Personalities, Study Shows

SMU researchers found GPT-4 can maintain consistent personality traits in video game NPCs with nearly 74% accuracy, far surpassing earlier models. The breakthrough could bring lifelike, emotionally complex characters into mainstream gaming.

27 Aug 2025

ChatGPT Buzzwords Spill Into Everyday Speech, Study Finds

A Florida State University study finds that ChatGPT and other large language models are influencing not just writing but also unscripted spoken English. Words like “delve,” “intricate,” and “garner” are showing marked increases in everyday conversations since the chatbot’s release in 2022.

27 Aug 2025

AI Chatbots Handle Suicide Extremes But Struggle With Grey Areas

A RAND study found that ChatGPT, Claude, and Gemini generally respond appropriately to very-high-risk and very-low-risk suicide questions but show inconsistency with intermediate-risk queries. While ChatGPT and Claude sometimes answered directly, Gemini was more variable, highlighting the need for refinement in mental health contexts.

27 Aug 2025

AI Framework Speeds Up Relay Catalysis Design With Transparent, Literature-Linked Pathways

Researchers from Xiamen University and the University of Edinburgh developed an AI-assisted framework that combines large language models with a catalysis knowledge graph (Cat-KG) to design relay catalysis pathways. The system rapidly finds, explains, and verifies multi-step reaction routes with links to original literature.

26 Aug 2025

Generative AI Flags Hidden Bird Flu Risks In Patient Records With Speed And Accuracy

Researchers at the University of Maryland School of Medicine used a generative AI large language model to scan emergency department notes and flag high-risk patients with potential H5N1 bird flu exposure. The tool identified hidden occupational and animal-contact risks quickly, cheaply, and with strong accuracy.

25 Aug 2025

Pseudolaw And AI Converge To Threaten Australia’s Justice System

The rise of pseudolaw in Australia is straining courts, police, and councils by spreading false legal theories that mimic the law’s language but lack substance. Now intersecting with generative AI, this movement poses a growing threat to democracy and judicial efficiency.

25 Aug 2025

Symbols And Scaling Unite To Unlock The Next Frontier Of Artificial Intelligence

Researchers from Peking University argue that scaling alone cannot drive breakthroughs in data-scarce frontier domains. They propose combining LLM intuition with human-designed symbolic systems, illustrated by AlphaGeometry’s success in solving Olympiad-level problems.

21 Aug 2025

AI Transforms Scientific Workflows With PegasusAI To Boost Discovery

Researchers from USC, UT, and partner universities are developing PegasusAI, a $5 million NSF-funded project that integrates AI into scientific workflow management, making data processing more intelligent, adaptive, and accessible. The system aims to democratize advanced computing across disciplines, from space science to public health.

19 Aug 2025

AI Chatbots Spread Medical Misinformation but a Simple Prompt Fix Cuts Errors in Half

A new study from the Icahn School of Medicine at Mount Sinai shows that AI chatbots frequently repeat and elaborate on false medical information embedded in user prompts. However, a simple warning line added to the input cut these hallucinations nearly in half, offering a promising safety solution.

14 Aug 2025

Malicious Chatbots Trick Users Into Oversharing Personal Data, Study Finds

Malicious AI chatbots can manipulate users into revealing up to 12.5 times more personal information by mimicking empathy and emotional reciprocity, a new study by King’s College London reveals. The research highlights how easily large language models can be weaponized for covert data extraction, with users largely unaware of the privacy risks.

13 Aug 2025

AI Tools Play Distinct Roles in Creativity: Study Maps How LLMs, T2I, and T2-3D Boost Design Innovation

This study systematically compares how different generative AI tools—LLMs, text-to-image (T2I), and text-to-3D (T2-3D) models—support distinct stages of the creative process. The findings reveal a practical framework for aligning each AI type with ideation, visualization, or prototyping tasks.

17 Jun 2025

AI’s Blind Spot: Machines Fail to Capture the Full Bloom of Human Experience

A new study finds that artificial intelligence, even with advanced training, cannot represent human concepts grounded in sensory and motor experiences, such as “flower,” with the same richness as humans do. AI language models excel at abstract terms but fall short when concepts require multisensory and embodied understanding.

5 Jun 2025

AI Chatbots Struggle With Mental Health Drug Side Effects, Georgia Tech Study Finds

Researchers at Georgia Tech developed a new framework to evaluate how well AI chatbots detect and respond to adverse psychiatric medication reactions in user conversations. The study found that while LLMs mimic human empathy, they often struggle to provide clinically accurate, actionable advice for mental health side effects.

3 Jun 2025

VizTrust Reveals How Users Build And Lose Trust In AI Chatbots In Real Time

Researchers at Binghamton University have developed VizTrust, a novel analytics tool that tracks and visualizes the dynamic nature of user trust during real-time interactions with AI chatbots. The system uses machine learning to analyze emotional and behavioral cues, offering actionable insights for improving conversational AI.

3 Jun 2025

Aaron Courville Champions Collaborative AI To Solve Society’s Biggest Challenges

Aaron Courville, incoming director of IVADO and co-founder of Mila, champions an optimistic and pragmatic vision for artificial intelligence, emphasizing its potential to solve societal challenges. Rather than fearing AI, Courville advocates for responsible, collaborative development to maximize its benefits across fields from climate action to public policy.

28 May 2025

Large Language Models News and Research

AI Moderation Systems Show Striking Inconsistencies In Detecting Hate Speech

AI Can Read Emissions Reports But Can You Trust What It Says?

Researchers Warn: AI Could Automate Every Stage of Ransomware

AI System Translates and Edits Human Thoughts Into Images Using Brain Scans

EnvGPT Transforms Environmental Research By Matching Bigger AI Models With Compact Precision

AI Gives Video Game Characters Real Personalities, Study Shows

ChatGPT Buzzwords Spill Into Everyday Speech, Study Finds

AI Chatbots Handle Suicide Extremes But Struggle With Grey Areas

AI Framework Speeds Up Relay Catalysis Design With Transparent, Literature-Linked Pathways

Generative AI Flags Hidden Bird Flu Risks In Patient Records With Speed And Accuracy

Pseudolaw And AI Converge To Threaten Australia’s Justice System

Symbols And Scaling Unite To Unlock The Next Frontier Of Artificial Intelligence

AI Transforms Scientific Workflows With PegasusAI To Boost Discovery

AI Chatbots Spread Medical Misinformation but a Simple Prompt Fix Cuts Errors in Half

Malicious Chatbots Trick Users Into Oversharing Personal Data, Study Finds

AI Tools Play Distinct Roles in Creativity: Study Maps How LLMs, T2I, and T2-3D Boost Design Innovation

AI’s Blind Spot: Machines Fail to Capture the Full Bloom of Human Experience

AI Chatbots Struggle With Mental Health Drug Side Effects, Georgia Tech Study Finds

VizTrust Reveals How Users Build And Lose Trust In AI Chatbots In Real Time

Aaron Courville Champions Collaborative AI To Solve Society’s Biggest Challenges

Trending Stories

Latest AI News