The Mathematics Search Engine

8+ min ago   (804+ words) Researchers found that, for reliable question answering using retrieval-augmented generation, sentence-based document chunking offers the best balance of cost and performance, rivalling more complex methods up to 5,000 tokens and avoiding unnecessary expense from overlapping chunks. https://quantumzeitgeist.com/wp-content/uploads/Image_fx-2026-01-23T095944.679.jpg Researchers are increasingly focused…...

26+ min ago   (1064+ words) There is no strong reason to expect these simple techniques to elicit superhuman knowledge from very powerful base models, e.g. because these techniques may fail in real applications where some consistent and salient human beliefs are wrong. We'll explore more challenging…...

38+ min ago   (779+ words) Researchers have developed LocBAM, a new technique for analysing 3D medical images that improves the accuracy of identifying anatomical structures by explicitly considering the spatial location of image sections, even when limited data is available. https://quantumzeitgeist.com/wp-content/uploads/Image_fx-2026-01-23T095855.037.jpg Researchers are tackling a critical…...

43+ min ago   (518+ words) We have incredibly wonderful tools through generative AI that could potentially take creativity and innovation to new levels. Yet, despite all the technology, we're not seeing significant changes. A group of researchers think they know why. The key to AI-boosted…...

45+ min ago   (1095+ words) Researchers have demonstrated that a new collaborative machine learning approach, utilising data from multiple hospitals without sharing it, can achieve brain tumour detection accuracy equivalent to analysing all data in one place, with analysis revealing the model prioritises specific MRI…...

45+ min ago   (705+ words) Researchers have developed a new artificial intelligence-powered system capable of rapidly and accurately designing large-scale metasurfaces , materials engineered to control light , by learning from smaller designs and ensuring they meet specific electromagnetic requirements. https://quantumzeitgeist.com/wp-content/uploads/Image_fx-2026-01-23T095559.212.jpg Researchers are tackling the longstanding problem…...

47+ min ago   (716+ words) Researchers have discovered a universal principle underlying intelligence , both biological and artificial , involving the iterative refinement of information through remapping and navigating internal representations, suggesting a pathway towards building truly adaptive AI systems. https://quantumzeitgeist.com/wp-content/uploads/Image_fx-96-7.jpg Scientists are increasingly exploring the fundamental…...

47+ min ago   (404+ words) The standard AI privacy playbook protects data after it leaves your environment. Here's how to protect it before it ever does. If you have spent any time researching AI and data protection, you have encountered these five steps in some…...

47+ min ago   (778+ words) Researchers have developed a rapid and cost-effective method to detect when large language models are likely to "hallucinate" incorrect information by analysing internal thought processes, allowing uncertain answers to be flagged or verified before being presented. https://quantumzeitgeist.com/wp-content/uploads/Image_fx-65-15.jpg Scientists are tackling…...

51+ min ago   (718+ words) Researchers have developed a new graph-based system, Transition-Aware Graph Network, which accurately predicts online shopper preferences by analysing sequences of actions , such as clicks and purchases , with significantly improved speed and efficiency compared to existing methods. https://quantumzeitgeist.com/wp-content/uploads/Image_fx-2026-01-23T095709.803.jpg Researchers are tackling…...