AWS Machine Learning BlogServiceNow configures OBO token exchange through Amazon Bedrock AgentCore Gateway, allowing AI agents to call the ServiceNow REST API while maintaining the identity of the calling user.
Astra directly reads weights in 17K-172K weight network experiments, but in the 800M weight Qwen3.5 mathematical ability improvement experiment, it requires combining Python programming and model execution, not just reading weights; Astra modifying 0.016% weights increased Qwen3.5-0.8B's MATH-500 from 7.9% to 48.1% and tripled IFEval.
404 Media6565World of Warcraft: Forever Released on Wednesday, Aiming to Recapture the Wonder of Players First Exploring the Game World in 2004. But by 2026, the Game-related Subreddit had been Flooded with AI Extensions Spreading Developer Achievements. Due to the Extensive Commercial Abuse of Warcraft Logs' API, Management Needs to Be Strengthened to Maintain Service Terms and Scalability.
O'Reilly Radar2222Claude Security reveals increased AI security risks, open-source model competition accelerates, Kimi 3.1 Pro becomes a leader and gets an accelerated version by Moonshot AI; Meta releases Muse series and OpenAI Decisions API.
Bilibili creator 'Yusang 10297' successfully installed iOS 16 on Huawei MateBook E Go 2023, achieving an ultra-large screen iPhone experience. The process was completed using Alibaba Tongyi Qwen3.8-Flash (Next) local model, and currently there is no driver, which will be gradually fixed and open-sourced.
China Telecom AI Research Institute (TeleAI) has launched the 'TeleAI Privacy Camera' for the public safety field. It achieves security alerts by pushing text descriptions through AI analysis of the environment, while protecting user privacy. This product does not store or transmit original video footage and can identify violations such as fighting, smoking, and falling. Developed by Professor Li Xuelong's team, it is based on AI Flow technology, using a edge-cloud collaborative reasoning architecture and a homologous model to complete privacy information de-sensitization processing at the moment of collection.
YC's CEO Gary Tan spends $100,000 to $200,000 annually on AI agents for completing work, reading emails, and private notes, even including psychological counseling records from 20 years ago. He publicly shares Claude Code skills and believes AI agents should be as flexible as humans in assisting tasks, rather than just focusing on repeatability and safety. Tan points out that the current usage costs are too high, expecting the costs of AI agents to decrease and achieve personal control.
Cai Xinxin, co-founder of Alibaba Group, said at the 2026 World Internet Conference that open-source can reduce AI R&D costs and promote AI popularization, especially important for developing countries. Currently, Alibaba's Qwen has open-sourced over 470 models, with a global cumulative download volume exceeding 4 billion and derivative models exceeding 330,000.
TechCrunch · AI6565Google launches its experimental AI app Dreambeans in India, marking its first release outside the US, hoping to bring personalized visual stories generated from user digital activities to a broader audience with the world's second-largest internet market.
Ars Technica · AI6565Apple plans to launch the HomeView Ultra home robot at a price of 1000 US dollars in 2027, supporting FaceTime calls and automatic light switching. This AI-powered robot prototype first appeared in 2018, with the code name J595, but its functions are still insufficient compared to existing smart home devices.
The Verge · AISame story · 2 outlets6565
MIT News · AI3838Regina Barzilay and Jim Collins win the 2026 Prizewinner in Science Research Excellence AI Award for their pioneering work in AI-assisted antibiotic discovery, integrating deep learning, bioengineering, and experimental validation to identify novel antibacterial candidates to combat drug-resistant pathogens.
OpenAI NewsSame story · 2 outlets3838OpenAI's Grid Protectors Program will collaborate with utilities to help future computing campuses support grid reliability and protect consumer electricity costs.
SenseTime has released the multimodal intelligent agent model SenseNova 6.8 Flash, which is available for a limited time free and achieves first or second place in office automation, multi-step collaboration, and data analysis benchmarks, while tying with GPT 6 Luna for first place in long-term memory evaluations. This model can complete dynamic webpages, dynamic posters, SVG animations, and dynamic PPT designs, and has proactive interaction capabilities, able to remember and remind to update old materials.
Google Search is celebrating the release of the Zelda: Ocarina of Time remake with an easter egg, entering zelda in the search will display the Master Sword, clicking it will redirect to the 1998 version of the Google interface. This interface only retains text search results and the 'I'm Feeling Lucky' button, and cannot browse other pages, one of Google's classic game easter eggs.
O'Reilly Radar6565Claude Code provides three loop trigger methods: goal, time, and active, as well as a manual loop mode. Task iteration is achieved through commands like /goal, /loop, /schedule. For goal loops, success criteria must be defined, such as an Lighthouse score of 90 or above.
Ai2 BlogNextcloud, used by German federal government and Spanish CSIC and other institutions as an open-source collaboration platform, requires AI models to meet transparency and accountability requirements of public organizations; Olmo 3 has received the company's highest ethical AI rating: Green.
Vision Technology's integrated 800V high-voltage DC power supply system is commercialized in the Wulanchabu Xinghe base, becoming the world's first similar system to operate in a large AI data center. The system consists of a Tianshu controller, solid-state transformer, DC energy storage, and DC/DC, supporting a green energy direct connection ratio of over 80%.
Sakana AI3838Jürgen Schmidhuber, professor at the Sakana AI Symposium, presented a speech reviewing the many ideas that contributed to modern deep learning and generative AI from the 1980s to the 1990s, and introduced the LSTM paper proposed in 1990. Sakana AI announced the development of a pre-trained model, planning to release a small language model named 'Sawara Small' in spring 2027.
TechCrunch · AI3535OpenAI plans to conduct public relations activities after the US midterm elections on November 2, 2026, and may rebrand its data centers as 'AI Computing Parks'. Chris Lehane, OpenAI's global affairs chief, said the move aims to influence the future of AI, including setting AI 'speed limits', helping the public understand how AI empowers daily life, and promoting the US's role in global AI standards. Lehane also believes the discussions around data centers are 'healthy and positive'.
Qt has released the zero-code embedded interface design tool Clate Beta 0.6, supporting visual editing and Claude Agent-driven AI generation, exporting QML code for integration with Qt 6.12, and does not currently support component reuse and data binding
vLLM releasesvLLM v0.33.0 released, including 927 commits and support from 389 contributors, new 501B MoE Reflection Beam-1.0 and its DFlash2 renderer.
New Chip Voyage releases the X7 Ultra chip, describing its performance for the first time using 'large model computing power'. The NPU computing power reaches 700 TOPS with a utilization of 98.7%, the quantized model output similarity to the FP32 benchmark model exceeds 99.99%, memory bandwidth consumption is reduced by 80%, and energy efficiency is improved by 100%. This chip is designed specifically for mass production of intelligent driving large models, featuring a super-large core NPU architecture and a dedicated computing engine, paired with AI toolchain optimization to support kilowire-level LiDAR processing and camera latency jitter optimization. It is equipped with an Arm v9 automotive-grade CPU and meets ASIL-B/ASIL-D safety standards.
vLLM releasesvLLM v0.33.0 released, adding 927 commits, 389 contributors, introducing 10 new models such as Reflection Beam-1.0 (501B MoE, 23B active) and Yandex AliceAI-Foundation-80B-A3B
Xiaomi Smart Doorbell 4 Pro Dual-Camera Edition is now on sale, with a launch price of 439 yuan. It features a 5MP + 2MP dual-camera design, with the main camera recording visitors and the sub-camera covering the ground; equipped with a 940nm infrared night vision light, with a night vision range of 5 meters; built-in 9800mAh battery, providing approximately 172 days of battery life under normal use; supports AI-powered recognition of visitors, packages, and pets, and pushes reminders.
OpenRouter Blog6262OpenRouter has released a resident AI agent named Dots through its Agent SDK, allowing users to customize models, tools, and running costs.
Simon Willison6565Drew Breunig embedded the NES 6502 CPU emulator into the Qwen3.5-4B model weights, allowing users to run NES ROMs like Super Mario Bros, but the speed is extremely slow. The project was built using Opus 5.5 prompts and is based on the NES CPU validation test set released 15 years ago by Christopher Pow.
The Verge · AISame story · 3 outlets6565Google's Project Suncatcher satellite successfully launched its on-board Tensor Processing Units (TPUs), enabling the on-board Gemma 4 model to create haikus about space based on prompts.
How the accounts differ:TechCrunch states the satellite creates poetry; The Verge specifies haiku about space.
Epoch AI6565Epoch AI analyzed 47,000 preprints on bioRxiv from January 2020 to September 2026 using GPT-5.6 Sol, finding that 15.5% admitted to using LLM, almost the same as the proportion of those who admitted to using professional AI tools (15.7%); OpenAI Decisions API, based on GPT-6 Luna, categorizes LLM usage into four major categories: research, code computation, writing, and literature review.
TechCrunch · AISame story · 3 outlets6060Google Suncatcher satellite, carrying four Tensor Processing Units (TPUs), has been launched by SpaceX and is now operational, successfully sending prompts to Google DeepMind's Gemma 4 large language model to create poems.
How the accounts differ:TechCrunch states the satellite creates poetry; The Verge specifies haiku about space.
AWS Machine Learning Blog3838Trakkr Research tracked and analyzed five AI assistants: ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. It was found that 40% of the statistics they cited could be traced back to the original research, 8% of the data paths showed changes, and the remaining 52% could not be verified for sources or had no research basis.
Google Developers Blog1212DevFest is the largest community-led developer event organized by Google, with over 800 events held from October 1, 2026, to December 31, 2026, covering 115 countries. Nearly a million developers will learn how to build, secure, and scale applications. The event delves into Google's latest AI technologies and offers comprehensive practical courses on Google Cloud, Android, and Web, organized by local communities and open to all developers.
