Google Research has released TurboQuant, a new compression algorithm designed to reduce the memory demands of large language models. The company says it can shrink a model’s key-value cache by at least six times and speed up a core processing step called attention computation by up to eight times, all without retraining the model or …
The development of large language models has undergone fundamental changes in 2025, marked by new training methods and surprising capabilities that reveal a fundamentally different form of intelligence than expected. AI researcher Andrej Karpathy writes on his blog about six major shifts that defined the year. The most significant change involves a new training technique …
Two major tech companies are tackling one of artificial intelligence’s most persistent problems: getting AI systems to stop making things up or hiding their mistakes. OpenAI and Amazon have each developed distinct approaches to make large language models more honest and reliable. OpenAI’s thruth serum OpenAI researchers introduced a technique called “confessions” that functions like …
Researchers at the University of Zurich and Cologne University are creating large language models trained solely on historical documents up to specific dates. The team, led by Daniel Göttlich, Dominik Loibner, Guohui Jiang, and Hans-Joachim Voth, describes their project on GitHub. The models, called Ranke-4B, contain four billion parameters and are trained on 80 billion …
Researchers at New York University have created a new AI architecture that generates higher-quality images more efficiently. Ben Dickson reports for VentureBeat that the model, called Representation Autoencoders or RAE, improves an AI’s understanding of image content, which leads to better results. The new method challenges common practices in building diffusion models, the technology behind …
Researchers at Google have introduced a new AI paradigm called “Nested Learning” to address a major weakness in current large language models (LLMs). Ben Dickson reports for VentureBeat that this approach could enable AI systems to learn and update their knowledge continuously after their initial training. Today’s LLMs are largely static. Their knowledge is limited …
Researchers have developed a method called Verbalized Sampling that uses a single sentence to make generative AI models produce more diverse and creative responses. The technique works on large language models like GPT-4 and Claude without requiring any retraining. Carl Franzen reports for VentureBeat that this method addresses the common problem of AI models giving …
Researchers at the Massachusetts Institute of Technology (MIT) have developed a technique that allows large language models to improve on their own. Carl Franzen reports for VentureBeat that the method, called SEAL, enables AI to autonomously generate its own training data. Instead of relying on fixed external datasets, models using the SEAL framework create instructions …
Researchers at Florida State University have found that buzzwords commonly used by AI are appearing more frequently in unscripted human conversations, McKenzie Harris reports for Florida State University News. The study analyzed 22.1 million words of spoken language, revealing a measurable increase in the use of words such as “delve,” “intricate,” and “underscore” after the …
Chinese researchers have created a system named MemOS, designed to provide artificial intelligence with a persistent, human-like memory. According to a report by Michael Nuñez in VentureBeat, the technology addresses a fundamental limitation that causes AI models to forget information between user interactions. Current AI assistants often cannot recall past conversations, a problem the researchers …