1 Chinese Institute for Molecular and Cellular Therapeutics, Chinese Institutes for Medical Research, Beijing, China.
2School of Basic Medical Sciences, Capital Medical University, Beijing, China.
Address for correspondence. Chuan-Yuan Li, PhD, Chinese Institutes for Medical Research, Beijing, China 100069. Email: [email protected].
Data from the Large Hadron Collider has revealed evidence of a long-sought particle process that serves as a guide into the realm of the smallest scales ever probed
Ready to see how bispecific antibody therapy is reshaping the future of marginal zone lymphoma?
Weight regain is only part of the story. Scientists are discovering how stopping GLP-1 drugs may affect the heart, blood sugar, and other protective benefits that go well beyond the scale.
Researchers at the University of Sydney have identified a fundamental mathematical reason why artificial intelligence systems can learn different behaviors even when trained under almost identical conditions, a discovery that could help improve the reliability and trustworthiness of AI technologies.
The study by Professor Pulin Gong from the School of Physics and his Ph.D. student Andrew Ly has been published in Nature Machine Intelligence.
As AI systems become increasingly embedded in health care, business, research and everyday life, ensuring they behave consistently and predictably has become a major challenge. The new research suggests that some forms of AI unpredictability are not due to engineering flaws or inconsistencies in computer hardware, but arise from the underlying mathematics of how neural networks learn.
(Cell Reports 24, 630–641; July 17, 2018)
As published, blots showing the PD168393-sensitive, EGF-dependent phosphorylation of the CrkII-based Picchu-X biosensor in HCC1954 cells (Figure 1A) were inadvertently duplicated from an experiment showing the Mo-IPQA-sensitive, EGF-dependent phosphorylation of the CrkII-based Picchu-X biosensor in HCC1954 cells. This latter experiment was published as Figure 1B in PMID: 28166195.
Mo-IPQA is a PD168393 derivative, and figures for both papers were prepared concurrently, leading to the inadvertent duplication.
“A modern chatbot writes warm, clean, grammatically perfect text, often better than the average person in a hurry. If you are scanning for typos or robotic phrasing, you are using a 2022 test on a 2026 system,” he said.
“What has not been solved is consistency, specificity, and backbone.”
Instead, you need to think about speed and sentence shape. “Chatbots answer almost instantly, and every answer is roughly the same length and structure, neatly organised, often with a tidy summary at the end,” said Martynenko.
We’ve never had a better understanding of how cities move, but turning that knowledge into smoother, cleaner and more equitable trips remains a challenge. The main obstacle to sustainable mobility may not be technological, but rather political and organizational.
Apps track buses in real time, sensors count cars and pedestrians and algorithms predict traffic jams. Advances in data-driven analytics even promise to let bus networks adjust service frequency to demand or to coordinate deliveries with fewer vehicles. And yet, getting across a city is often still a struggle: breakdowns, congestion, double-parked delivery vehicles blocking buses and cyclists, night shifts that end after the subway has stopped running.
Improving urban mobility without increasing emissions or deepening inequalities in access means making better use of existing data rather than building more infrastructure. One thing is becoming clear: mobility data is still underutilized. It can only transform how we get around if it’s shareable, representative and actionable.
Researchers have created a molecular “cage” that keeps protein-degrading drug molecules inactive until they are exposed to low-energy red light. The technology, called RuPHOTACs, controls when and where these molecules become active, which could help scientists develop treatments that destroy specific proteins while limiting unwanted effects elsewhere.
The paper is published in the journal Nature Communications.
Large language models (LLMs), the technology that powers tools like Gemini and ChatGPT, are more nuanced than ever. But that doesn’t mean they aren’t biased. Some models are more prejudiced than others, a Rutgers University researcher has found.
In findings published in Nature Machine Intelligence, Calvin K. Lai, an associate professor of psychology at Rutgers-New Brunswick, reports that although LLM outputs are increasingly aligned with human values, the processes by which they reason are often rooted in deep-seated stereotypes.
“What we found across most models is that they consistently reproduced human implicit biases during their reasoning process,” said Lai, a member of the scientific advisory board at Project Implicit, a nonprofit organization for research and education on hidden biases. “No one wants models that are spewing out stuff that is racist or sexist.”
