Articles tagged with Generative Ai
Researchers at Institute of Science Tokyo developed a new framework for generative diffusion models by reinterpreting Schrödinger bridge models as variational autoencoders. This approach reduces computational costs and prevents overfitting, enabling more efficient generative AI models with broad applicability.
Forestry professionals express concerns about AI's impact on land-management decisions and policy, citing 'black box' problems and data quality issues. However, they see potential for AI to support tasks like data analysis and task automation.
Researchers found that mice and AI agents developed similar behavioral strategies and neural representations when coordinating their actions, suggesting fundamental principles of cooperation transcend biology and technology. Mice learned to coordinate actions through three key behaviors: approaching partners' sides, waiting for arrival...
A new study from Mount Sinai Health System suggests that a simple tweak to how AI assigns diagnostic codes can improve accuracy and even surpass physician performance. By allowing the model to reflect and review similar past cases, researchers found better results in reducing time spent on paperwork and billing errors.
A new study reveals that AI voice technology has reached a stage where it can create 'voice clones' that sound just as realistic as human recordings. The study found that both types of AI-generated voices were evaluated as more dominant than human voices and some were also perceived as more trustworthy.
Researchers at FAU have developed a smarter AI framework that can manage complex systems with unequal levels of authority and adapt to imperfect information. The framework, based on reinforcement learning and game theory, reduces unnecessary computation while maintaining system stability and optimal strategy outcomes.
The partnership enables AI-driven design, antibody development, payload-linker synthesis, and clinical development to accelerate ADC drug discovery and development. The collaboration aims to deliver more effective treatment options for cancer patients worldwide by integrating cutting-edge technology and expertise.
Researchers at MIT developed a technique called SCIGEN that steers generative AI models to create promising quantum materials by following specific design rules. The approach led to the synthesis of two actual materials with exotic magnetic traits.
The Chatbot Assessment Reporting Tool (CHART) statement outlines 12 key reporting items for chatbot health advice studies, providing a checklist and methodological diagram. This guideline supports stakeholders in reporting, understanding, and interpreting the findings of such studies.
Researchers have identified over 120 genetic signals shaping foveal development, including pathways involved in vitamin A metabolism and retinal cell fate. The study provides the first comprehensive genetic dissection of human foveal pit architecture, revealing new insights into childhood visual disorders.
A study by University of Cambridge researchers found that ChatGPT, an AI chatbot, improvised solutions to a 2,400-year-old mathematical challenge. The bot made mistakes and adapted its approach based on user input, demonstrating a 'learner-like' behavior.
A new study found that people are more likely to cheat when delegating tasks to AI systems, especially with ambiguous interfaces. Researchers warn of a rise in unethical behavior as AI agents become increasingly accessible.
A Cornell historian warns that AI's inability to capture the emotional and moral complexity behind witness testimony risks reshaping how history is recorded. Human historians possess irreplaceable abilities in recognizing and conveying human suffering, particularly in preserving fracture, silence, and ethical weight.
Researchers developed a custom speech recognition system trained on Supreme Court hearings, reducing transcription errors by up to 9% compared to leading commercial tools. The AI tool semantically matches paragraphs with timestamps, allowing users to scroll through judgements and instantly watch relevant exchanges from the hearing.
Researchers developed a novel framework to qualitatively evaluate the quality of synthetic wireless data, improving AI model training with 4.3% performance gain. The framework assesses affinity and diversity in wireless signals, essential qualities for effective training.
Researchers found that more interactive mobile apps and AI chatbots foster a sense of playfulness, lowering privacy concerns. However, this interactivity can also distract users from thinking about the information they share with the system.
A new paper highlights the importance of human historians in capturing emotional complexity behind world events as AI struggles to accurately represent Holocaust survivors' experiences. Historians possess skills that AI lacks, including the ability to capture human suffering and preserve fracture and silence.
A study found that a ChatGPT-assisted writing system improved emotional engagement and intrinsic motivation for students with dyslexia, but negatively impacted writing performance. The system's effectiveness depends on teacher-guided integration to avoid undermining writing skills.
A recent study found dramatic differences in how large language models classify hate speech, with significant variations for specific demographic groups. The analysis of seven models revealed that content moderation systems have inconsistent detection accuracy, which can erode trust and create perceptions of bias.
A recent article argues traditional educational research has reached a breaking point and must be reimagined in the age of artificial intelligence. The authors identify long-standing problems with educational research and outline a vision for its rebirth in an AI-driven world.
A new poll of California adults reveals overwhelming bipartisan support for stricter digital technology regulations, with 82% backing the Phone-Free Schools Act. Support for school smartphone restrictions and social media age limits is driven by concerns about classroom distractions, excessive tech use, and mental health.
Experts weighed in on the current state of artificial intelligence (AI) in cancer care, noting both its potential for improving efficiency and its challenges. Many expressed excitement over AI's possibilities for helping support an overburdened oncology workforce and accelerating the pursuit of new cures.
Researchers at Politecnico di Milano developed photonic chips for training physical neural networks, eliminating digitisation requirements. This allows for faster, more robust, and efficient network training using light signals.
The Global RETFound initiative aims to develop the world's first globally representative AI foundation model in medicine, addressing health inequalities and AI bias. The project involves a diverse dataset of over 100 million fundus photographs sourced from 65+ countries.
Researchers have developed a new light-based chip that cuts power consumption for image recognition tasks by up to 100 times, using lasers and microscopic lenses fabricated onto circuit boards. This breakthrough enables faster performance and potentially strain-free AI systems.
A new study validates the use of Sybil, a deep learning artificial intelligence model, for predicting future lung cancer risk in a predominantly Black population. The model demonstrated high predictive accuracy for lung cancer risk up to six years after a single low-dose CT scan.
Researchers at UCR have developed a method to preserve AI safeguards in open-source models by retraining internal structure to detect and block dangerous prompts. The approach avoids external filters or software patches, instead changing the model's fundamental understanding of risky content.
A global study surveyed 14,000 patients across 43 countries, finding that those in poorer health were more likely to reject AI. Patients preferred explainable AI and wanted clinicians to make final decisions.
The Wits MIND Institute has received a $1 million boost from Google.org, enabling it to drive next-generation breakthroughs in natural and artificial intelligence. The partnership aims to advance the scientific understanding of both natural and artificial intelligence, foster breakthrough research and technological innovation.
Researchers at Penn Engineering introduced AMP-Diffusion, a generative AI tool that designed tens of thousands of new antimicrobial peptides with bacteria-killing potential. The most potent AMPs performed as well as FDA-approved drugs without detectable adverse effects.
A recent study from Radboud University Medical Center shows that AI can rapidly assess coronary arteries for vulnerable spots, allowing for more accurate prediction of recurrent heart attacks. This breakthrough could lead to tailored medication or preventive stents, improving patient outcomes.
Raina Biosciences has developed a generative AI platform called GEMORNA, which uses deep learning models to design novel mRNA sequences with enhanced expression and durability. The platform supports the creation of next-generation mRNA-based therapeutics for various diseases, including cancer and infectious diseases.
A former diplomat warns that algorithms lack empathy and intuition, which are essential for successful negotiations. However, AI can streamline diplomacy and amplify human aspirations when used carefully. Diplomats need training in AI ethics and global cooperation to ensure equal access and deployment.
The global market for AI in pharmaceuticals is projected to reach $13.4 billion by 2035, driven by Insilico Medicine's innovative use of generative AI. The company will share its achievements at a satellite forum during BIOHK2025, showcasing significant efficiency boosts compared to traditional methods.
Surya is a visual AI model trained on massive amounts of solar data, leveraging NASA's deep scientific expertise in cutting-edge AI models. The system aims to determine the Sun's magnetic field generation and structure, as well as its impact on the heliosphere and geospace.
Researchers from the University of Bath and University of Hong Kong found that additive advice bias is common in conversations, social media, and AI chatbots. This bias can leave people feeling more overwhelmed than helped, with well-intentioned tips piling on extra tasks.
A new study by UC Davis researchers reveals that popular GenAI browser assistants collect and share sensitive data without users' knowledge. The study highlights the need for stronger safeguards and transparency to protect user privacy online.
Researchers developed an AI framework to map interactions between content and algorithms on digital platforms, reducing the spread of misinformation. The system would allow users and platform operators to pinpoint sources of potential misinformation and promote diverse information sources.
Researchers trained GAMES to generate accurate SMILES combinations, enhancing computer-aided drug design and accelerating discovery. The model combines LoRA and QLoRA techniques to reduce hardware and energy needed.
Insilico Medicine has completed IND-enabling studies for ISM8969, an orally available NLRP3 inhibitor targeting inflammation and neurodegenerative disorders. The drug candidate demonstrated favorable PD/PK profiles and dose-dependent efficacy in motor functions in animal models.
Researchers at King's College London found that AI chatbots can be exploited to reveal up to 12.5 times more personal information from users. The study used three types of malicious AIs to encourage disclosure of private info, with reciprocal strategies proving the most effective.
A study of 1,100 professionals reveals that AI tools can make managers' emails more professional but also undermine trust when used for routine communication tasks. Employees perceive AI-generated content as less sincere and question their leadership abilities.
Researchers found that large language models like ChatGPT outperform humans in identifying neuromyths, but struggle to critically correct them. By prompting AI to explicitly correct false assumptions, error rates can be significantly reduced.
A large-scale study found that Green AI significantly improves operational and environmental performance in Pakistani SMEs, but only when leaders invest, and institutions are in place. Adoption works under the right conditions, with perceived ease of use and usefulness driving behavior.
A study by researchers at UC San Diego and University of Chicago found that most visual artists lack the technical know-how to block AI crawlers, despite wanting to protect their creative work. The team developed a tool called Glaze, but more effective solutions are needed to prevent AI data harvesting.
A new framework leverages automated 3D object segmentation and mesh reconstruction to simulate dynamic and realistic construction sites. The approach captures real-world scenes, extracts key objects, and reconstructs them into editable 3D meshes, enabling applications such as virtual safety walkthroughs and AI-driven layout planning.
The article provides reporting recommendations for studies evaluating chatbot health advice, highlighting the importance of standardization in interpretability. Key findings include the need for clear guidance on chatbot performance metrics and data quality control.
Researchers found that nonequilibrium thermodynamics explains why optimal transport theory is optimal in diffusion models, leading to robust data generation. The discovery offers a novel thermodynamic approach to machine learning research, particularly in image generation.
The journal is seeking high-quality submissions that explore the potential of voice and speech analysis in diverse medical applications. Submissions focus on acoustic and voice analysis techniques, voice or vocal phenotyping, and vocal biomarkers in health and medicine.
DPM-Solver++ is a fast training-free solver for guided sampling of diffusion probabilistic models. It addresses instability issues with high-order solvers and achieves improved sample quality in just 15 steps, making it a significant advancement in the field.
A team of computer scientists created 2,300 original sudoku puzzles and asked AI tools like OpenAI's ChatGPT to solve them. The results showed that while some AI models could solve easy sudokus, most struggled to provide accurate explanations, raising questions about the trustworthiness of AI-generated information.
The team aims to deliver AI power directly to devices, improving resilience and speed in constrained environments. By processing data step-by-step across a network of devices, they can create a safe and adaptable system that can withstand attacks and extreme conditions.
Researchers create AI model called UNITE that detects fakes by analyzing faces and backgrounds, making it one of the first tools capable of identifying synthetic or doctored videos. The system can flag a range of forgeries, from simple facial swaps to complex fully synthetic videos.
The Insilico Medicine recent summer updates webinar introduced new features and functional updates to the company's generative AI platform, including Generative Biologics, Chemistry42, PandaOmics, DORA, and PreciousGPT. These updates aim to accelerate drug discovery and advance life sciences research.
Researchers found that large language models are unaware of their own mistakes, unlike humans who can adjust expectations. The models' lack of self-awareness has implications for the use of AI chatbots in daily life, particularly when it comes to tasks with low stakes.
A University of Kansas study found that people rate corporate crises messages written by humans as more credible and trustworthy, regardless of the approach taken. However, the approach itself didn't vary between participants who read human or AI-written content.
Researchers used generative AI models like ChatGPT and DALL-E to analyze men's fashion trends and create realistic fashion collection images. The study highlights the importance of expertly worded prompts for accurate fashion design implementation, enabling efficient and creative use of generative AI in fashion.
Researchers from the Weizmann Institute of Science and Intel Labs have developed new algorithms that allow AI developers to combine the power of different AI models, speeding up large language model performance by 1.5-2.8 times. This enables faster collaboration between models, reducing processing power costs and improving overall effi...
The Hong Kong Polytechnic University has secured funding from the Research Grants Council to develop a cost-effective and sustainable collaborative Generative AI (Co-GenAI) model. The project aims to democratize AI development by reducing centralized computational demand and enabling broader participation in GenAI research and deployment.
Researchers used AI model ChatGPT-4 to extract clinical variables from MRI and CT scans of nearly 1,000 adults with pancreatic cysts. The accuracy rate ranged from 97% for solid components to 99% for calcific lesions, comparable to human performance.