936 stories tagged Research · page 21 of 39
Researchers developed PhyDrawGen, an AI system that generates physics diagrams from text while respecting physical laws. Unlike previous tools, it doesn't make common mistakes about forces or geometry. This could help students and teachers create precise diagrams for learning and teaching physics.
Scientists found that AI models often predict realistic-looking but physically impossible outcomes. They propose a new approach that focuses on understanding real-world physics to make AI more reliable.
Scientists studied how AI agents update their own tools and strategies. They found that not all AI models benefit equally from these self-improvements, even if they're good at their original tasks.
Researchers introduced PReMISE, a framework that uses pairwise human-preference data to discover policy-level rubric sets and audit LLM judges, ensuring AI scores align with human preferences and avoiding misleadingly polished but factually incorrect responses.
Researchers developed a method called AdaCoM to help AI agents manage information overload during long tasks. Unlike prior approaches that require retraining the agent itself, AdaCoM adapts context management strategies to each agent, making it practical for closed-source systems.
Researchers developed a new AI system to simulate how hospitals react to policy changes. This could help design better healthcare systems by predicting real-world impacts.
Researchers introduced DecomposeR, a new AI framework that improves how large language models plan and execute complex research tasks. By representing research plans as typed directed acyclic graphs (DAGs), DecomposeR enables better credit assignment for planning and execution, potentially leading to more structured and accurate long-form answers.
Researchers have created a new benchmark to test how well AI models make clinical decisions. This could help improve AI tools that assist doctors in diagnosing and treating patients.
Researchers introduced MAVEN (Modular Agentic Verification and Execution Network), a lightweight symbolic reasoning scaffold that helps AI agents use tools more reliably across different tasks. This could make AI assistants better at handling complex, multi-step problems.
Researchers found that AI models often prioritize globally dominant narratives over local cultural knowledge. They created a dataset to study this issue in Bengali, showing how AI can misrepresent cultural contexts.
Researchers propose a new way for AI to automatically gather and prepare its own specialized training data. This could make AI models better at niche tasks without human help. The approach, called 'Autonomous Agentic Data Engineering,' lets AI act as its own data engineer, potentially speeding up customization for fields like medicine or law.
Researchers at UC San Diego have developed an AI model that can predict how tumors will respond to treatments by analyzing their genetic mutations. This could lead to more personalized and effective cancer treatments in the future.
Researchers have developed VFEAgent, an AI system that automates Finite Element Analysis (FEA) from images and text descriptions. This could make advanced engineering tools accessible to non-experts, speeding up design processes.
Scientists have created StoryMI, a multi-AI system that can generate realistic therapeutic dialogues. This tool could help train therapists and provide new ways to practice motivational interviewing techniques.
Researchers created a new AI model called the Cognitive Categorical Transformer (CCT) that mimics how the human brain categorizes information. It outperforms a standard model by 12% on a language test, showing promise for smarter, more context-aware AI assistants.
Researchers found that the tone of your questions can change how well AI models like ChatGPT answer. Polite or neutral tones often get better results than aggressive or overly casual ones. This matters because it shows how small changes in how we talk to AI can make a big difference in the answers we get.
A new study tested AI-generated reviews for academic papers and found they don't always match human opinions. The research also reveals that authors can "game" the system by using AI to revise their papers before submission, raising serious questions about fairness and reliability in scientific publishing.
The Austrian Academy of Sciences, in collaboration with Mistral AI and Reply, is developing an AI model called Apollo to read and translate ancient Greek papyri. This project could revolutionize how we study historical texts that were previously unreadable.
AI models can't reliably tell cause from effect, even with fine-tuning. Researchers prove this is a fundamental limitation and propose a new method to overcome it.
A new study found that AI models often avoid answering ethical questions directly, showing a tendency to be non-committal. This 'omissive bias' could have real-world consequences when people rely on AI for advice.
A new study reveals a simple method to bypass safety filters in AI image generators. The technique, called PAST2HARM, uses past tense prompts to trick models into creating harmful content.
AI models often verify facts better than they generate them, creating a 'factual gap'. This research explores why this happens and how it affects our trust in AI. New research reveals AI models often verify facts better than they generate them, creating a 'factual gap'. This research explores why this happens and how it affects our trust in AI.
Researchers have developed a new AI system that can detect human values in text, both explicit and implicit. This could help make AI decisions more aligned with ethical and moral considerations.
Researchers have introduced two advanced AI models designed for long-term coding tasks. These models, Laguna M.1 and XS.2, represent a significant leap in AI capabilities for software development.