The survey found that more than half of Americans have used generative AI services, with concerns about misinformation and manipulation dominating public attitudes. Despite these concerns, the study highlights opportunities for growth and development as AI continues to evolve.
Researchers create SciAgents framework to autonomously generate and evaluate promising research hypotheses in biologically inspired materials. The framework uses graph reasoning methods to organize relationships between scientific concepts, mimicking biological systems.
A study by Washington State University found that ChatGPT struggles with nuanced financial tasks, even when compared to human professionals. The AI model performed well on broad concepts but showed significant inaccuracy on specialized topics such as determining clients' insurance coverage and tax status.
A recent study from UBC researchers found that AI chatbots contain biases that can shape environmental discourse in unhelpful ways. The bots amplified existing societal biases and leaned heavily on past experience to propose solutions, largely steering clear of bold responses.
Researchers at the University of Washington developed a new method called variational preference learning, which predicts users' preferences and tailors its outputs accordingly. This approach overcomes the limitations of traditional RLHF, which can lead to averaging user preferences and resulting in incorrect outputs for all users.
Leading AI chatbots exhibit mild cognitive impairment in tests for early dementia detection, performing poorly in visuospatial skills and executive tasks. This finding challenges the assumption that AI will replace human doctors and highlights a significant area of weakness that could impede their use in clinical settings.
A new study found that large language models can provide useful physical exam recommendations and instructions based on patient symptoms. The researchers used GPT-4 to recommend physical exam instructions and evaluated its performance by three attending physicians, scoring at least 80% of the possible points.
A new study by UCL researchers found that AI systems amplify human biases, leading to a snowball effect where small initial biases increase the risk of human error. The researchers demonstrated real-world consequences, including overestimating white men's likelihood of holding high-status jobs and underestimating women's performance.
The Keck School of Medicine of USC has joined the Ryght Research Network to leverage AI tools and optimize cancer clinical trials. The collaboration aims to offer more clinical trials to patients, build new AI tools, and speed up the development of medical treatments.
The Leopoldina discussion paper emphasizes the need for a more nuanced understanding of generative AI, addressing concerns around bias, transparency, and data protection. It highlights the importance of developing strategies to counteract bias and ensure accountability in AI development.
The study uses generative AI to identify and optimize potent and selective HPK1 inhibitors for immunotherapy. The results show a relatively balanced candidate compound with adequate in vitro ADME, in vivo PK properties, good oral bioavailability, and robust in vivo efficacy in various cancer models.
Researchers found that large language models adjust their responses to appear more desirable when given the Big Five personality test. This 'social desirability bias' suggests that LLMs can emulate human-like preferences for certain personalities.
A study by the University of Copenhagen found that many students fear using plagiarism detection software, leading to counterproductive behavior and misdirected learning. The researchers recommend clearer guidelines, instruction on academic writing practices, and institutions' responsibility in explaining software limitations.
A new study finds that large language models are prone to social identity biases, favoring their perceived ingroup while expressing negativity toward outgroups. However, the researchers discovered that carefully selecting the training data can reduce these biases, suggesting promising directions for improving AI development and training.
A team of researchers proposes creating an AI virtual cell to model human biomolecules, cells, and tissues. The virtual cell would allow scientists to experiment in silico, accelerating personalized medicine and disease research.
Insilico Medicine has received its first clinical milestone payment of $10 million from Exelixis for XL309, a selective USP1 inhibitor discovered with the company's AI platform. The drug is being developed for advanced solid tumors and has shown efficacy in preclinical studies.
A new AI model developed at the University of Georgia predicts nearby traffic movements and incorporates innovative features for planning safe vehicle movements. This approach helps reduce crashes and near-misses by consolidating two steps: predicting surrounding traffic movements and planning a self-driving car's motion.
Insilico Medicine has nominated ISM8969, an orally available NLRP3 inhibitor, as a potential treatment for various inflammatory diseases. The compound has shown promising results in preclinical evaluation studies, including balanced druggability and efficacy against inflammation.
Researchers at Tokyo University of Science have developed a new method called black-box forgetting, which enables selective removal of unnecessary information from large pre-trained AI models. This approach enhances model efficiency and improves privacy by reducing computational resources and information leakage.
A pioneering AI model has been developed to understand the genetic 'language' of plants, allowing for precise predictions about RNA functions and identification of functional patterns. This breakthrough has significant implications for crop improvement and the next generation of AI-based gene design.
Insilico Medicine has successfully integrated cutting-edge generative AI models into its platform, achieving significant increases in accuracy and speed. The company's AI-powered solutions have shown promising results in preclinical candidates and Phase II clinical trials.
Evo, a generative AI model, uses patterns in microbial genomes to write new genetic code, expanding the length of sequences models can process and improving resolution. Researchers use Evo to understand microbial and viral genomes, fashion new proteins, and reprogram microbes for remarkable tasks.
Researchers develop a simple fix to an existing technique, enabling the generation of sharp, high-quality 3D shapes that rival top model-generated 2D images. The new approach improves upon previous methods by avoiding costly retraining and complex postprocessing.
A state-of-the-art flood monitoring dashboard is being developed in Central Appalachia to test its reusability and replicability. The project uses generative artificial intelligence to provide communities with the information and tools needed to build resilience in the face of climate change.
The book examines AI's current advances, hurdles, and potential, emphasizing the need for science to maintain core norms and values. Experts advocate for human accountability and responsibility when using AI in research, highlighting the importance of transparent disclosure and attribution.
Researchers found that teenagers use generative artificial intelligence for emotional support, social interactions, and academic purposes. However, they also reported significant misconceptions about generative AI among parents and children. Parents had little to no understanding of their children's use of GAI, and they viewed AI as a ...
Most medical large language models offer little better than guesses when it comes to interpreting medical information. ChatGPT-4 achieved the best performance, but its accuracy remained insufficient for some medical code questions. The study highlights the need for broader clinical language in models and caution in their widespread use.
Insilico Medicine has developed an innovative platform using generative AI to discover novel molecules. The company has nominated over 20 preclinical candidates and received IND clearance for 10 molecules.
Researchers developed an AI tool called BrainBench to test large language models' ability to predict neuroscience study outcomes. The results showed that LLMs surpassed human experts with an average accuracy of 81%, highlighting their potential as powerful tools for accelerating research.
Recent Nobel Prizes in physics and chemistry have recognized the convergence of AI with physics and chemistry, emphasizing the need for interdisciplinary research. Researchers advocate for nurturing AI-enabled polymaths to bridge the gap between theoretical advancements and practical applications.
Alex Zhavoronkov, Insilico Medicine's founder and CEO, was named Highly Cited Researchers of 2024. He has published over 40 peer-reviewed research papers, with a total H-index of 64, and led the development of Insilico's lead drug pipeline, ISM001-055.
A study found that ChatGPT provides higher fatality numbers when asked in Arabic compared to questions in Hebrew, highlighting the impact of user language on information dissemination. The researchers believe this has profound social implications, as it can shape perceptions of conflict and fuel biases.
A new AI tool generates realistic satellite images of future flooding, which can help communities visualize and prepare for approaching storms. The method combines a generative artificial intelligence model with a physics-based flood model, producing more accurate and realistic images than an AI-only approach.
A novel no-code prototype, Auto-DSM, generates design structure matrices (DSMs) using large language models. It improves productivity and accuracy compared to traditional methods, with speeds of up to four minutes per DSM generation.
Insilico Medicine has been selected by Fortune as one of the top AI innovators, leveraging its generative AI platform to develop novel drugs and treatments. The company's lead drug pipeline, ISM001-055, has shown promising results in Phase IIa trials, offering new hope for patients with idiopathic pulmonary fibrosis.
Experts discuss the potential of AI to enhance healthcare, including its use in diagnosing diseases and educating patients. However, concerns about the accuracy of chatbot responses and the need for verified sources highlight the perils of relying on AI in medicine.
The GenAI4ED project aims to revolutionize secondary education by developing a digital platform that assesses and selects GenAI-based educational software. The platform will provide personalized recommendations based on predefined criteria, taking into account both technical effectiveness and educational impact.
Researchers developed BiomedParse, an AI medical image analysis model that works across nine types of medical images to predict systemic diseases. This enables non-specialists to analyze and identify subtle variations in medical images, leading to improved diagnosis.
New algorithm CALDERA enables leaner LLMs with improved performance, reducing costs, energy consumption and privacy risks. This allows organizations to adapt models to their specific needs without sharing sensitive data, enhancing user privacy.
Researchers found that membership inference attacks on large language models (LLMs) are not effective in measuring information exposure risks. The common method used to test LLM leaks suffers from ambiguity due to the fluidity of language, making it difficult to define a representative set of non-member candidates.
Pharma.AI Week will showcase the latest advancements in Insilico's generative AI platform, including PandaOmics and Science42, with expert speakers and hands-on demos. The event aims to empower researchers and scientists with tools for faster and more accurate discoveries.
The Healthcare AI Challenge enables healthcare professionals to evaluate and compare AI solutions for their effectiveness in various clinical situations. By crowdsourcing input from experts, the challenge aims to create continuous and reliable expert evaluations of AI technologies in medicine.
The Phase IIa trial of ISM001-055 showed dose-dependent improvement in forced vital capacity (FVC) and percent predicted FVC, suggesting potential to slow or reverse disease progression. The treatment also demonstrated a favorable pharmacokinetics profile with minimal adverse events.
Insilico Medicine CEO Alex Zhavoronkov to discuss AI business potential and economic growth at Fortune Global Forum 2024. The company has developed a generative AI-powered platform that utilizes deep learning techniques for novel target discovery and molecular structure generation.
Despite their ability to provide turn-by-turn driving directions in New York City with near-perfect accuracy, generative AI models do not form a coherent internal map of the city. The researchers developed new metrics to test a transformer's world model and found that only one generated a coherent world model for Othello moves.
Insilico Medicine has entered into a revolving loan facility of up to US$100 million with HSBC, enabling the company's global expansion and AI-driven innovation in biotechnology. The credit line will support Insilico's proprietary novel drug discovery pipeline and its end-to-end diversified AI platform, Pharma.AI.
A new generative AI tool called Theseus is being developed to reduce time and labor in computational modeling, suggesting and running simulations to accelerate research and discovery. The AI assistant aims to enhance the work of computational scientists by providing scientific intuition, checking code for bugs and making suggestions.
A new study from Virginia Tech shows large language models can assess human-made environments using street-view images, similar to traditional methods. LLM-based performance offers a more accessible tool for users in small to medium-sized cities, making it easier to manage smart urban infrastructure.
Researchers found significant bias in AI-powered hiring systems, with LLMs favoring white-associated names 85% of the time, female-associated names only 11% of the time, and never Black male-associated names. The study highlights the need for intersectional analysis to ensure fairness in AI systems.
Researchers found that AI doctors who recall patients' social information, with privacy control offered, increase patient satisfaction. Human doctors don't require social or medical information to establish a close relationship with patients.
The collaboration utilized Insilico's generative biology AI platform, Pharma.AI, to identify a novel lead for treating oncology diseases. The joint R&D team addressed the druggability challenge of highly novel targets using novel scaffolds generated by Chemistry42.
A new study by University of Technology Sydney researchers challenges the Infinite Monkey Theorem, finding it unlikely a monkey can type out Shakespeare's works before the universe ends. With around a 5% chance, even an increased number of monkeys would not produce the Bard's entire works.
Researchers at UC Berkeley found that humans create more diverse and nuanced stories than generative AI tools like ChatGPT. While AI shows improvement, it still struggles with complexity and clichés.
MIT researchers developed a versatile technique that combines diverse data from various sources into a shared language for generative AI models. This approach outperformed traditional techniques by 20% in simulation and real-world experiments.
A study of 50 U.S.-licensed physicians found that GPT-4 did not significantly improve clinical reasoning compared to conventional resources. The integration of GPT-4 as a diagnostic aid alongside clinicians showed promising results but required further exploration to understand its potential benefits.
A new study by Penn State researchers suggests that making AI training data diversity information available can shape users' expectations of algorithmic fairness and trust. Displaying racial diversity cues in AI interfaces can enhance users' perceptions of algorithmic fairness and trust, according to the study's findings.
A recent study reveals that when people are told a story was written by AI, they rate it poorly, suggesting a distrust of AI-generated art. Researchers believe this could hinder the use of AI in public health campaigns to encourage healthy behaviors.
Researchers at University of Zurich used AI to analyze antibiotic resistance using GPT-4 model, creating EUCAST-GPT-expert system for accurate interpretation of antimicrobial resistance mechanisms. The AI system performed well in detecting certain types of resistance but had limitations, while human experts were more accurate but slower.
A new study reveals that AI-driven chatbots may perpetuate racial and ethnic biases in pain assessment, leading to further inequalities in healthcare. Researchers found that Black patients were consistently underassessed for their pain compared to white patients, regardless of whether the rater was human or AI.
A study from Lehigh University and Seattle University found that making patients aware of biases in human healthcare decisions increases receptiveness to AI recommendations. By highlighting the limitations of human judgment, healthcare providers can create a more balanced relationship between patients and emerging technologies.