Google Releases Large Language Model Gemini Pro, Bard Gets Smarter


Google Gemini Pro/Ultra subscribers can now experience the Veo 3.1 video model, featuring the new 'Ingredients to Video' function: supports uploading three reference images at once, extracting character, scene, and style features respectively, and generating an 8-second 1080p video. The generated content includes an embedded SynthID invisible watermark, supporting text input on web and mobile devices for one-click generation. The system ensures character consistency across frames and consistent lighting, with demonstration cases showing that three self-portraits + cyber city background + oil painting style image can
Apple and The Ohio State University jointly launched the FS-DFM model, which can generate long text comparable to traditional models after only 8 iterations, achieving a writing speed improvement of up to 128 times, breaking through the efficiency bottleneck of long text generation. The model uses discrete flow matching technology, different from self-regressive models like ChatGPT that generate text character by character.
Alibaba launched Qwen3-Max-Preview, a trillion-parameter language model, setting a new AI benchmark. Available via Qwen Chat and Alibaba Cloud API, it outperforms predecessors in knowledge, dialogue, tasks, and execution.....
NVIDIA launches Jet-Nemotron language models (200M & 400M params), achieving 53.6x faster generation than SOTA with equal/higher accuracy via 'post-neural architecture search' that modifies pre-trained models.....
Factorio, a complex video game centered around building and resource management, has emerged as a novel tool for researchers to evaluate artificial intelligence capabilities. The game allows for testing the abilities of language models in planning and constructing complex systems while managing multiple resources and production chains. To this end, a research team developed a system called the "Factorio Learning Environment" (FLE), offering two distinct testing modes. The "Experiment Mode" contains 24 structured challenges with specific goals and limited resources, with tasks ranging from simple two-machine setups...