Insights into the global AI industry — major companies, partnerships, funding, and technological impact across sectors.
BitVLA introduces a compact 1-bit vision-language-action model that cuts memory use to 1.4GB and runs on low-cost hardware without cloud dependency.
Liquid AI releases a 350M-parameter model built for agentic tasks, delivering consistent benchmark scores while fitting in under 500MB quantized.
A new robotics framework enables faster and safer learning through simulated environments, significantly improving success rates in complex real-world manipulation tasks.
A new study finds that AI trained on its own medical data rapidly loses reliability and diversity. The findings raise serious concerns about large-scale AI deployment in healthcare.
Meituan introduced LongCat-AudioDiT, a diffusion-based TTS model operating in waveform latent space. The system improves voice cloning and multilingual audio generation by eliminating intermediate representations like mel-spectrograms.
Meta has released the Efficient Universal Perception Encoder, a vision backbone for edge devices. The model unifies multiple vision tasks using multi-teacher distillation.
Microsoft Research and CUHK have built an autonomous system that generates research ideas, runs experiments, and writes scientific papers - with quality approaching top medical AI conference standards.
Optimizing the system around an AI model - not the model itself - can deliver measurable performance gains, according to new Stanford research.
Trace2Skill is a new framework that improves AI performance by converting execution traces into reusable skills. The method avoids model parameter updates entirely while enhancing task outcomes across a range of agent-based systems.
China's first fully automated humanoid robot production line in Foshan can produce over 10,000 units annually - with robots building robots around the clock.
Kuaishou's Kling Team has introduced HyDRA, a hybrid memory system that keeps AI-generated subjects consistent even when they leave the frame entirely.
NVIDIA has released Privasis, a synthetic dataset designed to train AI models for text sanitization. The dataset includes over 54 million annotated privacy attributes spanning medical, financial, and legal documents.
Microsoft has open-sourced VibeVoice, a voice AI system capable of processing long-form audio and real-time speech tasks. The model supports structured transcription, multilingual capabilities, and low-latency voice generation.
Huawei and Shanghai Jiao Tong University introduced HyperOffload, a compiler-assisted framework that reduces LLM peak device memory usage by up to 26%. The solution rethinks how data moves across AI hardware - maintaining performance while solving one of the field's core infrastructure challenges.
Humanoid robots are being piloted in hospitals to assist with routine operations. Early deployments focus on reducing workload through automation of non-clinical tasks.
Global data center capacity is projected to nearly triple by 2030, driven primarily by AI workloads, while power infrastructure constraints are emerging as a critical limiting factor.
NVIDIA and academic researchers introduced EGGROLL, a new method for training AI models without gradient-based backpropagation. The approach uses low-rank evolution strategies to improve efficiency and scalability for large models.
A new token-dropping technique accelerates large vision-language models at inference time - no retraining, no architectural changes, and minimal performance loss.
A new JEPA-based AI framework simplifies world model training, achieving planning speeds up to 48 times faster with minimal compute resources.
A new AI training approach improves traffic signal systems, reducing average waiting times by over 10% and addressing key limitations in traditional models.
We use cookies to improve your experience on our site and to show you relevant advertising. To find our more, read our privacy policy and cookie policy
Victoria Bazir
Saad Ullah
Peter Smith
Eseandre Mordi
Usman Salis
Marina Lyubimova
