Articles tagged with Generative Ai
A new Australian study computes the creative ability of LLMs using standard mathematical principles, showing a maximum of 0.25 on a scale from 0 to 1. The findings challenge widespread assumptions about AI's creative powers and offer clarity amid a global debate.
A study of 258 published novelists in the UK found that half believe AI will replace their work entirely. The majority also report using AI for non-creative tasks like information searches and editing text written without AI. However, many express concerns about copyright laws not being respected and originality being lost.
This book offers a comprehensive exploration of AI-driven analytics in finance, addressing market prediction, fraud detection, and risk assessment. It also discusses AI applications in healthcare and cybersecurity, including disease classification and biometric identification systems.
The University of Maine has launched internships in AI and digital twins to prepare students for careers in the growing blue economy. Students will work with real-time data, build virtual replicas of ocean structures, and test complex marine scenarios.
A new study found that while parents are willing to accept AI-generated images in children's stories with certain conditions, they raise concerns about errors in illustrations that might encourage unsafe behavior or lack real-world accuracy. Parents prefer clear notification on the cover of the story if AI was used to create images.
A new study by University of Washington researchers found that people tend to mirror AI systems' hiring biases when making decisions, even when the bias is not obvious. The study involved 528 participants who reviewed resumes and picked candidates for jobs, with some AI recommendations biased towards certain races.
A Dartmouth study finds that AI-powered chatbots can deliver personalized learning to large numbers of students. The researchers created an AI teaching assistant called NeuroBot TA that provides around-the-clock individualized support for students, which they found to be more trusted than general chatbots.
The collaboration combines Insilico's Pharma.AI platforms with Lilly's expertise to discover and advance new therapies. Insilico will utilize its platform to generate candidate compounds, accelerating the development of transformative treatments for urgent patient needs.
Researchers found that large language models exhibit systematic biases when evaluating texts, but only when the source or author is revealed. The studies showed a high level of agreement among models when no information was provided, but decreased agreement and even bias emerged when fictional sources were used.
The joint research hub aims to bridge academic research and scalable infrastructure for open AI systems, with a focus on decentralised computing architectures and secure AI marketplaces. Early proof-of-concepts are expected within two years, targeting pilot applications in finance, healthcare, and smart infrastructure.
A new study reveals AI can generate highly realistic images of real people that are hard to tell apart from authentic photos. Despite familiarity with the subject's appearance, participants struggled to distinguish between genuine and fake images.
Researchers found that non-expert users' intuitive prompts triggered biased responses as effectively as expert 'jailbreak' techniques. The study used a competition to identify biases in generative AI models and suggests strategies for mitigating issues.
A study found that undergraduate computer science students initially increase their trust in generative AI programming tools, but long-term trust remains unclear. To address this, researchers recommend that educators provide tasks-based training and ensure students comprehend and modify code without relying on AI assistants.
A new empirical study by computer scientists reveals that software developers tend to scrutinize AI-generated code less critically and learn less from it. In contrast, human pairs are more likely to engage in broader discussions and critically examine each other's contributions. The study highlights the limitations of AI assistants in ...
Researchers warn that AI-generated wildlife videos create false connections with nature, leading to misconceptions about wild animals and their behaviors. The proliferation of these videos can undermine conservation efforts and disconnect society from the natural world.
Researchers examine how genAI affects task structure, worker adoption, and job displacement. The authors suggest genAI will widen the 'cone of automation,' substituting for complex work and infrequent tasks.
A study by Aalto University reveals that people who use AI for logical reasoning tasks tend to overestimate their performance, even if they are not good at it. The researchers found that users who considered themselves more AI literate actually assumed their abilities were greater than they really were.
Researchers developed a generative AI model, BIGE, to generate optimal motions for athletes to avoid injuries and aid in rehabilitation. The model can produce realistic videos of motions that athletes can mimic during training or execute while injured.
Insilico Medicine's Alex Zhavoronkov and Alex Aliper will share their insights on generative AI breakthroughs in healthcare. The company is leveraging AI to accelerate drug discovery, reducing the time from traditional methods by 2.5-4 years.
Engineers at Duke University have constructed an 'agentic system' of AI bots that can solve complex design problems nearly as well as a fully trained scientist. The researchers created large language model (LLM) AI agents to complete all the legwork, allowing them to automate straightforward but niche design problems.
A study found that AI image generators reproduce and amplify gender biases in different languages, with varying degrees of impact. The results showed that direct prompts using the generic masculine produced the strongest biases, while explicitly feminine prompts showed almost exclusively women.
Insilico Medicine has released several AI-powered tools, including PandaOmics and Generative Biologics, to accelerate drug discovery. These platforms utilize machine learning and AI models to analyze large datasets and generate novel hypotheses.
A recent study found that most users cannot identify AI bias in training data, unless it affects their own group. Researchers created 12 versions of a prototype AI system to detect facial expressions and tested how users might detect bias in different scenarios.
Recent research found that large language models are not yet able to consistently fool humans in conversations. They struggle with using discourse markers, opening and closing features, and subtle imitations. Despite rapid development, key differences between human and artificial conversations will likely remain.
Researchers used five advanced LLMs to assess their logical reasoning capabilities and found that they consistently complied with requests for false medical information. Targeted training and fine-tuning improved LLMs' abilities to respond to illogical prompts accurately, but challenges remain in aligning models to every type of user.
A new AI-based risk assessment tool, GRACE 3.0, has shown better predictability of future risk for heart attack patients compared to traditional methods.
Generative AI accelerates scientific discovery in oral sciences, but requires transparent disclosure and robust verification methods to ensure responsible integration. The use of AI in research demands human oversight, accountability, and continuous monitoring to maintain quality and credibility.
A study by Cornell University researcher So-Yeon Yoon found that generative art can positively impact consumer behavior in virtual stores, with greater effects on mass-market shoppers. The research suggests that dynamic generative art may be more effective than static art in enhancing perceptions of exclusivity and aesthetic pleasure.
The platform allows users to create personalized LLM-based chatbots, serving as a 24/7 teaching assistant or scraping campus websites to find needed resources. Illinois Chat has already been utilized in various departments and research groups, providing valuable feedback to make it more practical.
Researchers from Florida Atlantic University's Charles E. Schmidt College of Medicine conducted a scoping review on AI-driven systems designed to deliver motivational interviewing. The study found that most used AI tools were chatbots, with varying levels of fidelity to core principles like empathy and autonomy.
Large language models like ChatGPT are more persuasive than humans, especially when convincing people to adopt veganism or attend graduate school. The AI's ability to provide concrete logistical support and use authority-boosting vocabulary contributes to its persuasiveness.
Pharma.AI is a comprehensive generative AI-powered drug discovery platform featuring PandaOmics, Generative Biologics, Chemistry42, Science42: DORA, and PreciousGPT. These platforms enhance target and biomarker discovery, biologics engineering, small molecule design, scientific research, and disease modelling.
Bezzo's project enables embodied cognition capabilities in large language models, equipping robots with human-like understanding of human intent. The work has wide-ranging potential for assistive technologies, collaborative robots, and defense missions.
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.