By the TensorMax editorial team
· Drawing from sources across the AI industry
Today's top story
model release
Google has announced the General Availability of Pathways, a distributed system for training large models, including Gemini, on Google Cloud Platform.
Why it matters. The General Availability of Pathways on Google Cloud Platform marks a significant milestone in large-model training, as it addresses the coordination challenge that arises when training jobs stop being uniform programs. With Pathways, Google aims to provide a distributed system that can train its own large models, including Gemini, by introducing a single client that describes the whole computation as a graph, and a distributed runtime that handles the transfers, dependencies, and ordering. This development is crucial, as it can help improve the efficiency and scalability of large-model training, with Google citing that hardware failures interrupted Gemini 2.5 training multiple times per hour, highlighting the need for a more robust system.
Google has announced the General Availability of Pathways, a distributed system for training large models, including Gemini, on Google Cloud Platform. Pathways is designed to address the coordination challenge that arises when training jobs stop being uniform programs, such as in pipeline parallelism or RL post-training for an LLM. The system introduces a single client that describes the whole computation as a graph, and a distributed runtime that handles the transfers, dependencies, and ordering. This allows for more efficient and scalable training, as it can recover from failed hosts and remap work onto healthy chips, instead of relaunching the whole job. Pathways also enables elastic training, where the runtime can adjust the number of devices used for a given computation, and compilation can start before physical chips are provisioned. The system is available on Google Cloud and can be used in conjunction with other frameworks, such as Ray, to orchestrate the cluster and pipeline. With Pathways, model building and execution are separated, and users can reason about the whole cluster instead of one slice of it. The system has been designed to work with JAX programs on TPUs and provides a range of features, including asynchronous dispatch, parallel asynchronous dispatch, and a central scheduler. Google has also provided a range of resources for users to learn more about Pathways, including documentation, papers, and GitHub repositories.
More from today
product launch
$1B
Why it matters. The release of Grand Theft Auto VI is a strategic milestone, with a development cost of over $1 billion and the potential to become the biggest video game release of all time. The game's predecessor, Grand Theft Auto V, has sold more than 230 million copies, and the new installment is expected to have a significant impact on the gaming industry. With its lavish variety of gameplay and detailed simulations, GTA VI is poised to set a new standard for the industry, and its success will be closely watched by competitors and investors.
product launch
$3.5M
Why it matters. The launch of Apple's Vision Pro, a $3,499 spatial computer, marks a significant strategic stake for the company as it attempts to create a new platform and redefine the future of computing. With over 2.5 billion active Apple devices and $416.2 billion in net sales in fiscal 2025, Apple is leveraging its enormous ecosystem to introduce innovative products that can potentially disrupt the market. The success of Vision Pro will be crucial in determining the direction of Apple's future growth and its ability to maintain its position as a leader in the technology industry.
model release
Why it matters. The strategic stake of Anthropic's AI systems dramatically accelerating AI research is that capabilities progress has doubled in the past year, with the company's internal measurement, AECI, showing a twofold increase in progress between October 2025 and October 2026. This acceleration, likely driven by AI automation, has significant implications for the AI industry and research timelines, potentially leading to much shorter timelines than previously projected.
model release
Why it matters. The release of Google's Gemini 4 Argon, a general-purpose physical-reasoning model, has significant implications for the robotics and AI industries, with the model scoring 87% on the Embodied Reasoning Question Answering ERQA benchmark and demonstrating improved performance in tasks such as robot control and simulation, including a notable rise from 56% to 75% success in a real setup after fine-tuning with domain-randomized MuJoCo episodes.
product launch
Why it matters. The development of a public certificate authority for post-quantum web security using Merkle Tree Certificates by Cloudflare has significant implications for the future of web security, with the company reporting a 9% faster median performance in a Chrome experiment. This improvement is crucial as post-quantum signatures can be approximately 40 times larger than classical signatures, increasing bandwidth and storage requirements. Cloudflare's efforts to upgrade web security before quantum computers can break it are part of a larger coordination challenge in the history of the Internet.
research paper
$109.2B
Why it matters. The strategic stake of Apple's services business growth to $109.2 billion is significant, as it now makes up a substantial part of the company's revenue alongside the iPhone. With over 2.5 billion active Apple devices and $416.2 billion in net sales in fiscal 2025, Apple's services business has become a core component of the company's ecosystem, providing a huge business built around what people do after they buy an Apple device. This growth is a testament to the company's successful expansion into new areas, including the App Store, advertising, cloud services, and more.
Catch up quick
-
Researchers demonstrate that private LLM copies can agree on a secret key over a public channel using shared preferences.
-
Researchers from Hong kyu Lee, Kai Shu, and Li Xiong propose a training-free un-alignment method for compositional safety in large vision-language models, called CMaG, which reduces compositional safety scores by up to
-
Researchers from multiple institutions introduce BINDIntent, a framework to improve the robustness of smaller large language model (sLLM) agents to instruction paraphrases, achieving 17.5-49.1 percentage point success率im
-
Nvidia CEO Jensen Huang faces investigation over alleged affair with Chinese spy, potentially impacting his leadership role
-
Anthropic agrees to pay $1.5 billion in copyright settlement
-
Nvidia CEO Jensen Huang had an affair with a Chinese spy, Alice Huang, who was an executive at Singapore-based data center company Megaspeed, a major purchaser of Nvidia AI hardware.
-
Nvidia's market cap falls by $300 billion after CEO Jensen Huang's affair with Chinese spy is revealed
-
UBS agrees to buy Credit Suisse for $3.66 billion in a state-brokered rescue
-
Hackers who stole $6.3 billion from Binance used memecoins to launder the proceeds
-
US lawmakers launch investigation into Nvidia CEO Jensen Huang's relationship with Chinese spy Alice Huang
-
Democrats to introduce legislation requiring independent safety testing of advanced AI models and holding developers liable for harmful outputs
-
AI companies sold or priced $300B+ of stock in 2026, with ~$100B more on tap
-
Hedge funds sold global technology stocks at a record pace due to concerns over AI data-center buildout
-
Copper prices fall on fears of AI backlash after US midterm elections
-
IDC predicts global telecom and pay-TV spending to reach $1.6 trillion by 2026, driven by inflation and high energy costs
Also on the desk
What landed, what bounced.Hit reply with one story from today that landed for you and one that bounced. Two sentences is plenty.
|
|