Innovation

Meta Unveils Four New Generations of Custom AI Chips to Reduce Nvidia Reliance Meta has announced a significant expansion of its custom silicon efforts with four new generations of its in-house Meta Training and Inference Accelerator (MTIA) chips. The new chips—MTIA 300, 400, 450, and 500—are engineered to manage the massive AI workloads for Meta’s services, including ranking, recommendation systems, and generative AI. This strategic move aims to reduce the company’s reliance on external chip suppliers like Nvidia while improving performance and efficiency for its specific needs. The MTIA 300 is already in production, with subsequent generations planned for a rapid six-month release cadence. The more advanced MTIA 450 and 500 models are specifically optimized for generative AI inference and are slated for mass deployment in 2027. ...

AI Creativity Now Matches Average Humans, But Lags Behind Top Innovators, Study Reveals A landmark study from the Université de Montréal, co-authored by AI pioneer Yoshua Bengio, reveals that leading AI models now match or exceed the average human’s creative abilities in linguistic tasks. Published in Scientific Reports, the research is the largest comparative study of its kind, testing large language models like GPT-4, Claude, and Gemini against 100,000 people. While models like GPT-4 have achieved a significant milestone by reaching average human creativity, the study underscores that the most creative individuals still significantly outperform the most advanced AI systems. ...

OpenAI Unveils GPT-5.2, A Powerful AI Model for Professional Workflows OpenAI has officially launched GPT-5.2, its most advanced AI model series yet, engineered to enhance professional workflows and tackle complex, multi-step projects. The new series—comprising GPT-5.2 Instant, Thinking, and Pro—delivers significant improvements in coding, spreadsheet and presentation creation, image perception, and understanding long contexts. According to OpenAI, GPT-5.2 Thinking shows a 30 percent reduction in errors compared to its predecessor, GPT-5.1, making it more reliable for tasks like research, writing, and analysis. The model also demonstrates improved performance in mathematical problem-solving and abstract reasoning benchmarks. The rollout of GPT-5.2 will begin with paid ChatGPT users, with the API now available to all developers, marking a strategic move to solidify OpenAI’s leadership in the competitive enterprise AI landscape. ...

AI Digital Twin Breakthrough: DT-GPT Predicts Patient Health Futures Researchers at the University of Melbourne have developed a groundbreaking AI tool, DT-GPT, that creates virtual digital twins of patients to forecast their individual health paths. The model was trained on three datasets comprising thousands of electronic patient health records for conditions like Alzheimer’s disease and non-small cell lung cancer. In testing, DT-GPT outperformed 14 other leading machine learning models in predictive accuracy. This technology could potentially simulate clinical trial outcomes, accelerating drug development and reducing costs. It may also enable medical professionals to anticipate patient health deterioration for earlier intervention and predict adverse medication side effects for more personalized treatment plans. The research has been published in NPJ Digital Medicine. ...