Google Gemini Robotics 1.5, Intel, AMD AI Servers

The artificial intelligence landscape is rapidly evolving with new developments in robot capabilities, infrastructure, and global regulation. Google DeepMind has introduced Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, advanced AI models designed to enhance robot understanding and interaction with the physical world, enabling tasks like object detection, movement planning, and even web searches for information. These models are now accessible to developers and select partners, aiming to move robots beyond single instructions towards genuine problem-solving. Meanwhile, the United States is implementing AI export controls, a move to regulate technology by restricting exports of AI chips to "countries of concern" while allowing shipments to allies, aiming to prevent adversaries from accessing advanced AI. In parallel, there's a push for a national AI policy framework in the U.S., with experts urging Congress to preempt conflicting state regulations that could stifle innovation and increase costs for AI companies. On the infrastructure front, Huawei has unveiled SuperPoD, an architecture capable of linking thousands of AI chips into a single logical computer, utilizing its UnifiedBus 2.0 protocol to manage large-scale AI computing. MiTAC Computing is also addressing the demand for AI infrastructure with scalable server solutions, offering models like the G4520G6 for inference and the G8825Z5 for training, supporting Intel and AMD processors alongside NVIDIA GPUs. Beyond these core developments, AI is finding diverse applications: a French startup, Nolej, is making classrooms interactive by transforming materials into quizzes and exercises, while a new app called Neon is paying users to record calls for AI voice assistant training, though raising privacy concerns. In Africa, the International Telecommunication Union (ITU), Google, and will.i.am have launched an initiative to provide AI and robotics training to students in underserved communities, with Google contributing $1 million and AI expertise. However, amidst this rapid progress, some AI researchers are voicing serious concerns about potential existential risks posed by increasingly advanced AI, urging caution and a slowdown in development.

Key Takeaways

• Google DeepMind has launched Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 to enhance robot capabilities in understanding environments and performing complex tasks.
• The U.S. has implemented AI export controls to regulate technology, restricting chip exports to "countries of concern" while allowing them to allies.
• Experts are urging Congress to establish a national AI policy framework to preempt conflicting state regulations and foster innovation.
• Huawei has introduced SuperPoD, an AI infrastructure architecture designed to connect thousands of AI chips into a single logical computer.
• MiTAC Computing offers scalable server solutions, including models supporting Intel CPUs with NVIDIA GPUs for inference and AMD EPYC processors for training.
• A French startup, Nolej, has developed an AI tool to convert classroom materials into interactive learning experiences for teachers.
• An app named Neon is paying users to record outgoing calls for AI voice assistant training, raising privacy and consent questions.
• The ITU, Google, and will.i.am have launched an initiative to provide AI and robotics training to students in underserved African communities, with Google contributing $1 million.
• Some AI researchers are warning of potential existential risks from advanced AI and are advocating for caution and a slowdown in development.
• Exabeam and DataBahn are partnering to enhance AI security data pipelines, aiming to improve threat detection and reduce operational costs.

DeepMind's new AI helps robots understand and act

Google DeepMind has released Gemini Robotics-ER 1.5, a new AI model designed to power robots. This advanced model allows robots to detect objects, plan movements like grabbing a mug, and even search the internet for information, such as local recycling laws. Developers can now access this model, which aims to enhance robot capabilities in the physical world.

Google DeepMind AI models help robots learn and search web

Google DeepMind has launched new AI models, Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, to help robots perform complex, multi-step tasks. These models enable robots to understand their surroundings, use web searches for information, and even learn from each other. This advancement allows robots to move beyond single instructions to genuine problem-solving in physical tasks, with Gemini Robotics-ER 1.5 available to developers and Gemini Robotics 1.5 for select partners.

Google DeepMind adds advanced thinking to robot AI models

Google DeepMind has introduced two new AI models, Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, to help developers create smarter robots. These models add advanced thinking capabilities, allowing robots to understand their environment and plan complex tasks. Gemini Robotics-ER 1.5, a vision-language-action model, turns visual data into actions, while Gemini Robotics 1.5, a vision-language model, creates multi-step plans. Gemini Robotics-ER 1.5 is now available to developers, with Gemini Robotics 1.5 accessible to select partners.

US implements AI export controls to manage technology

The United States has established export controls to regulate artificial intelligence (AI) technology, becoming one of the first countries to do so. This rule allows U.S. companies to export AI chips to allies but restricts exports to "countries of concern." The Department of Commerce finalized this rule in January 2025 to help U.S. companies safely export AI technology while preventing adversaries from accessing advanced AI capabilities. This marks a shift towards a more government-led approach to regulating technology.

Expert urges national AI framework to boost innovation

Adam Thierer, a Senior Fellow at the R Street Institute, testified before the House Judiciary Committee, urging Congress to create a national AI policy framework. He argued that a unified federal approach is needed to preempt conflicting state regulations that stifle innovation and increase costs for AI companies. Thierer recommended specific actions, including preempting state rules on frontier AI labs and algorithmic bias, and establishing a working group for federal and state AI policy. He emphasized that existing state laws can still address local concerns.

Huawei unveils SuperPoD to link thousands of AI chips

Huawei has introduced SuperPoD, a new AI infrastructure architecture designed to connect thousands of AI chips into a single, powerful logical computer. This system uses the UnifiedBus 2.0 protocol to overcome challenges in large-scale AI computing, such as communication reliability and bandwidth. The Atlas 950 SuperPoD, a key implementation, features 8,192 Ascend 950DT chips and offers massive processing power. Huawei is also opening up UnifiedBus 2.0 specifications to foster an ecosystem for AI infrastructure development.

French AI app Nolej makes classes interactive

A French startup called Nolej has developed an AI tool that transforms classroom materials into interactive learning experiences in seconds. Teachers can use it to convert documents, slides, or videos into quizzes, review sheets, or adaptive exercises, saving significant preparation time. The platform also includes a progress tracker and a smart tutor chatbot. Recognized globally, Nolej aims to support teachers and enhance student engagement and learning effectiveness.

Exabeam and DataBahn partner to enhance AI security data

Exabeam and DataBahn have formed a strategic partnership to improve security data pipelines using AI. This collaboration integrates DataBahn's AI data fabric with Exabeam's security operations platform, giving security teams better control over data routing, ingestion, and retention. The partnership aims to reduce operational costs, improve threat detection accuracy with AI-driven behavioral analytics, and support compliance by allowing flexible data retention. This integration helps security teams focus on critical data for faster and more effective threat detection.

MiTAC offers scalable servers for AI training and inference

MiTAC Computing is providing flexible and scalable server solutions to meet the growing demand for AI infrastructure. Their new servers, the G4520G6 and G8825Z5, are designed for demanding AI model training, inference, and data analytics. The G4520G6 supports diverse AI inference workloads with Intel CPUs and NVIDIA GPUs, while the G8825Z5 is built for extreme AI training using AMD EPYC processors and GPUs. MiTAC offers customized solutions from individual servers to full cluster integrations.

ITU, Google, will.i.am launch AI training for African students

The International Telecommunication Union (ITU), Google, and will.i.am have launched an initiative to provide AI and robotics training to students in underserved communities across Africa. This program aims to boost AI literacy and equip young people, especially girls and underrepresented groups, with essential digital skills. Google is contributing $1 million and AI expertise to the initiative, which will deliver hands-on training, robotics kits, and localized curricula to schools in countries like Ghana, Kenya, and Nigeria.

App pays users to record calls for AI training

A new app called Neon is gaining popularity by paying users to record their outgoing phone calls for AI training purposes. The app, available on iOS and Android, offers up to $30 daily or per-minute rates for call recordings, which are then anonymized and used to train AI voice assistants. While the iOS version is a top free app, the Android version has faced issues. Concerns have been raised about privacy and consent, especially regarding state laws requiring two-party consent for recordings.

AI researchers warn of existential risks

Some artificial intelligence researchers are expressing serious concerns about the rapid advancement of AI, warning of potential existential risks to humanity. As companies heavily invest in AI development, these researchers urge caution and a slowdown in progress. They highlight the possibility of AI becoming superhuman and posing a threat, a concept once confined to science fiction but now considered a closer reality by some experts.

Sources

• DeepMind Releases New Robot AI Models
• Google DeepMind’s new AI models can search the web to help robots complete tasks
• Google DeepMind Adds Agentic Capabilities to AI Models for Robots
• The United States Regulates Artificial Intelligence with Export Controls
• Testimony in Review: Adam Thierer’s September 2025 Testimony Before the House Judiciary’s Subcommittee on Courts, Intellectual Property, Artificial Intelligence, and the Internet
• Inside Huawei's plan to make thousands of AI chips think like one computer
• This AI turns any class into an interactive tool… in seconds - Futura-Sciences
• Exabeam, DataBahn Partner to Transform Security Data Pipelines with AI Integration
• How MiTAC Helps Organizations Scale for Both AI Training and Inference
• ITU, Google, and will.i.am Launch AI and Robotics Training Initiative for African Students
• This Free App Pays You to Record Your Calls for AI Training, and It's a Hit
• They warned about AI before it was cool. They're still worried : Consider This from NPR