NerdNews

Nvidia announces DGX Spark and DGX Station, 'personal AI supercomputers' powered by the Grace Blackwell platform, allowing developers to run AI models locally. The systems will be manufactured by Asus, Dell, HP, and Lenovo, with prices expected to start around $3,000.

Nvidia announced Rubin Ultra and Feynman AI chips for 2027 and 2028. Rubin Ultra will provide 15 exaflops of FP4 inference compute, while Feynman's details are scarce. These chips will power robots and billions of AI agents, with Nvidia's CEO envisioning 'AI factories' producing tokens instead of physical objects.

Google DeepMind's Gemini Robotics integrates vision, language, and action to make robots smarter and more capable. It features embodied reasoning, enabling robots to understand and react to their environment in a human-like way, and is being developed for use in various settings, including homes and offices.

Nvidia has announced the availability of its Isaac GR00T N1 foundation model, designed to expedite the development of humanoid robots. The model utilizes a dual-system architecture and can be customized for specific needs. It has been demonstrated on 1X's NEO Gamma humanoid robot, performing autonomous tasks. The model is now available for download, and several companies, including Boston Dynamics and Agility Robotics, have had early access to it.

Meta CEO Mark Zuckerberg announced that the company's Llama models have reached 1 billion downloads, a 53% increase from 650 million in December 2024. Llama is an open AI model family that powers Meta's AI assistant across various platforms, including Facebook, Instagram, and WhatsApp. Despite facing setbacks, including an AI copyright lawsuit and data privacy concerns, Llama has achieved widespread success, with companies like Spotify, AT&T, and DoorDash using the models in production.

Google's Gemini 2.0 Flash AI model enables image editing through conversational dialogue, allowing users to add, remove, or modify objects, scenery, and lighting. The model is trained on a large dataset of images and text, and can generate images with varying levels of success. While the output quality is not perfect, the technology has the potential to revolutionize image editing and media creation.

A study claims Amazon uses algorithms to discourage unionization among warehouse workers, citing examples of 'algorithmic slack-cutting' and surveillance. Amazon denies the allegations, questioning the study's sample size and methodology.

The European Union's law enforcement agency, Europol, has warned that artificial intelligence (AI) is turbocharging organized crime, making it more efficient and sophisticated. The agency's report highlights the use of AI in cybercrime, including AI-driven attacks, deepfakes, and voice cloning, which are being used to deceive victims, impersonate individuals, and commit fraud. The report also notes that states are using criminals as contractors to carry out cyberattacks, making it difficult to distinguish between state-sponsored and traditional cybercrime.

NVIDIA CEO Jensen Huang's GTC 2025 keynote covered the company's AI plans, including robots, in-car technology, and next-generation graphics architectures Blackwell Ultra and Vera Rubin. The keynote also featured the announcement of the DGX Spark desktop AI supercomputer and the GR00T N1 general foundation model for robots.

Stability AI's new model, Stable Virtual Camera, generates 'novel views' of a scene from one or more images, allowing for dynamic camera paths and realistic depth and perspective. The model can produce videos in various aspect ratios and is available for research use under a noncommercial license.

Google is transitioning to Gemini and removing popular Assistant features, including photo sharing, interpreter mode, and daily updates. Gemini will replace Assistant later this year, but may lack direct replacements for some features.

Nvidia has announced the DGX Spark and DGX Station, two new 'personal AI supercomputers' powered by the company's Grace Blackwell platform. The DGX Spark is a small, $3,000 desktop computer, while the DGX Station is a larger desktop counterpart with no announced price. Both are designed for users to work on large AI models with or without a connection to a datacenter.

Nvidia announced two 'personal AI supercomputers', DGX Spark and DGX Station, powered by the Grace Blackwell chip platform, allowing users to prototype, fine-tune, and run AI models at the edge. DGX Spark delivers up to 1,000 trillion operations per second, while DGX Station features Nvidia's GB300 Grace Blackwell Ultra Desktop Superchip with 784GB of memory.

Nvidia's Groot N1 is a generalist model trained on synthetic and real data, featuring a dual-system architecture for 'thinking fast and slow'. It's available in open source and aims to enable humanoid robots to perceive, reason, and plan actions. Groot N1 is an evolution of Nvidia's Project Groot and broadens the focus to humanoid robots in various form factors.

Graphite, an AI-powered code review platform, has raised $52 million in a Series B round led by Accel. The platform uses AI to flag errors and possible oversights in code, and also suggests code changes and generates possible fixes for code failures. With this funding, Graphite plans to invest in product development and grow its team.

Nvidia's GTC 2025 conference is expected to feature announcements on AI, GPUs, and quantum computing. CEO Jensen Huang will give a keynote address on Tuesday, focusing on AI and accelerating computing technologies. The company may reveal new GPU iterations, including Blackwell Ultra and Rubin, and discuss its approach to quantum computing.

Nvidia unveiled its next-generation CPU and GPU architectures, Vera and Rubin, which will be used in its rack-scale compute architecture. The Vera CPU will feature 88 custom-designed Arm cores, while the Rubin GPU will have two reticle-limited dies capable of up to 50 petaFLOPS at FP4 precision. The company plans to release the Rubin Ultra in late 2027, which will double the number of GPU dies and HBM modules, and deliver 15 exaFLOPS of FP4 inference performance and 5 exaFLOPS of FP8 for training.

Dell, HPE, and Lenovo have announced new hardware powered by Nvidia's GB300 Superchip, including a 20-petaFLOPS desktop and servers with high-speed interconnects. The new systems are designed for large-scale AI workloads, scientific research, and other compute-intensive applications.

Nvidia announced new GPUs, including Vera Rubin, Blackwell Ultra, and Feynman, with significant performance upgrades and custom CPUs, targeting AI inferencing and training tasks.

Over 300 malicious Android apps were downloaded 60 million times from Google Play, acting as adware or attempting to steal credentials and credit card information. The apps, part of the 'Vapor' campaign, were able to bypass Google's security reviews by downloading malware functionality post-installation.