Baseten Labs6565Research indicates that on-policy self-distillation (OPSD) may still be valuable in specific scenarios, but more research resources should not be invested. OPSD trains through the outputs generated by the student model itself with context prompts, but research shows that this intensive supervision may introduce bias, especially for trained models, making it less effective.
TSMC is considering expanding its business scope, entering advanced memory chip manufacturing to gain a larger share in the rapidly expanding AI hardware supply chain. TSMC has recently started planning to produce an advanced memory chip named '3D DRAM' at NVIDIA's request. TSMC is evaluating two paths: one is to build a dedicated new factory, and the other is to acquire an existing wafer factory and transform it into a memory production facility.
Planet AI app launched, users can naturally search, observe, and analyze changes in Earth's surface over ten years; currently free to use, AI conclusions are based on viewable raw satellite imagery; supports three core capabilities: 'Search Earth', 'Explore Changes', 'In-depth Analysis', and has been used in ecological conservation, land management, etc.
Anthropic releases research stating that Claude AI agents can improve long-term task performance through a 'dreaming' mechanism, reflecting on historical records, correcting errors, and building knowledge structures during idle time. Claude Opus 5.5 and Sonnet 5.5, after using memory in CL-Bench database question-answering tasks, achieve average scores of 0.838 and 0.833 respectively, outperforming the memoryless baseline of 0.655 and 0.644.
The Verge · AI6666China reportedly recruited Megaspeed executive Alice Huang to spy on Nvidia CEO Jensen Huang, attempting to obtain US AI chip information and possibly influence Nvidia's supply of the most advanced chips to China. Alice Huang has close ties with Nvidia executives, and Megaspeed is a customer of Nvidia. Three committees in the US House of Representatives have demanded that Huang attend a closed-door meeting regarding his 'interactions' with Alice Huang.
Amp introduces the Spaces feature, where the main Agent can automatically generate and coordinate multiple parallel Agent threads to complete projects. Each Space allows teams to plan, execute, and track tasks within a single chat, with the main Agent responsible for goal decomposition and progress organization, while developers only need to define goals and review results.
Cursor launches the 'Real-time Voice Chat with Agents' feature, supporting developers in two-way voice conversations with coding agents to describe code modification requirements, inquire about task progress, etc., while the agents complete coding work in the background. This feature is used in Cursor's Agents Window, IDE, web version, and iOS app, driven by Grok Voice for speech recognition and synthesis, offering multiple voice options like Ara, Eve, and supporting switching during calls.
Databricks Blog3838OpenSharing is an open protocol developed based on Delta Lake and Iceberg, used for directly reading customer tables without copying data. It supports incremental pull of change data, real-time access to the latest data, and smooth backfill of historical data, while reducing costs for customers and consumers in data management, storage, and access control. Consumers do not need to maintain SFTP or API connections; customers can share tables with authorization and can track access records through system tables.
Databricks Blog2222Databricks exposes shared Delta tables as real-time data sources through OpenSharing, allowing recipients to read incremental changes using Structured Streaming without local file tracking.
GitHub Blog · AI & ML3838GitHub Copilot supports cross-device session continuity on canvases, allowing users to execute tasks in cloud sandboxes, isolate repository editing through the Copilot app, enhance efficiency with storage instructions and shared memory project knowledge, and optionally match work requirements with the Copilot Fusion model.
NVIDIA Technical Blog6565FLUX 3 Image Klein achieves a 3.5x speedup on NVIDIA Blackwell GPU. The model parameter size is 175B, scoring 95.3 on MMLU evaluation, supporting 8k context length, and providing closed-source services via API.
Anthropic Research3434Anthropic proposes that AI building scientific models requires three steps: modeling, constraining, and explaining, where the constraining model can obtain probability distributions rather than single solutions with Bayesian reasoning; AI has made significant progress in constraining models, but if only the model is given, it lacks understanding of the real system.
Anthropic ResearchFeatured6262Anthropic has published a research paper, breaking down Claude's intelligent agent memory into components such as memory notes, offline dreaming organization, and memory hierarchy improvement, and explaining that memory refers to context that is preserved across sessions and can be read again, excluding weight updates, context learning within a single session, and compaction.
Why it matters: Anthropic has disclosed the hierarchical design of agent memory and the dreaming mechanism, providing comparable experimental results using CL-Bench, Pokemon, and internal document corpora.
OpenRouter Blog3232Sentry migrated its internal AI usage to OpenRouter in July 2026, clearing about 500 Anthropic and OpenAI API keys within three weeks, and closed direct access in August, achieving a 93% reduction in cost limits while maintaining model access freedom.
Codex releases00GitHub Copilot 0.169.0-alpha.2 released, this AI coding assistant is used for tasks, projects, and automation work, providing code review, quality analysis, and security vulnerability detection features.
IBM Research Blog3838IBM has deployed the IBM Quantum System Two system, equipped with the IBM Quantum Nighthawk r2 processor, at the University of Chicago's Hyde Park Laboratory, supporting the National Quantum Algorithms Center (NQAC) and the broader Illinois quantum ecosystem. The system will initially run at the University of Chicago and will eventually move to the Illinois Quantum and Microelectronics Park (IQMP), a park built with a $500 million investment by the Illinois government, which will become the long-term home for NQAC and new quantum computers.
The Verge · AI3838AI data center construction has drawn global attention due to issues such as energy consumption, environmental impact, and community opposition. OpenAI and Amazon have committed to cooperating with local communities and covering energy costs, but the U.S. Congress may pass restrictions. AMD CEO Lisa Su stated that the company prioritizes data centers, and OpenAI is gradually renaming data centers to 'compute campuses'.
Epoch AI6262Huawei is building a large AI data center in Linhe, Inner Mongolia, which is one of the largest AI computing power centers in China, but its computing power is only one-tenth of that of the largest US data center, Colossus 2. The Ascend 910C chips deployed in the center are inferior in both performance and energy efficiency to Nvidia H100, and their production capacity is far lower than Nvidia's, resulting in Chinese AI infrastructure supporting only 40% of the computing power of leading US facilities at the same power level.
Tomasz TunguzAnthropic's Claude Mythos 5 will simulate the cost of corporate network intrusion from $770 in May to $385 in July, while Zhipu's GLM-5.3 matches the attack rate at $90, a 99% decrease from the $5000-10000 cost of GPT-5.5 in April.
The DecoderSame story · 4 outlets6565
Newcomer3737Founders/CEOs of Perplexity, Applied Intuition, Flapping Airplanes, Vercel, and General Catalyst join the speaker lineup for the Cerebral Valley AI Summit; summit tickets $1,299, founder tickets $199.
MIT News · AI3838MIT's open-source model CheMeleon, pre-trained with millions of molecular data, can predict 1600 chemical descriptor properties of new molecules, outperforming other baseline and expert models in 58 test cases. The model has been adopted by several pharmaceutical companies and has performed outstandingly in the OpenADMET competition, significantly reducing the need for experimental validation of new molecules.
OpenAI News3838OpenAI has released BioBasicsBench, a benchmark containing 100 tasks, to evaluate the ability of AI agents to interpret raw experimental data and handle daily wet experimental science.
Cursor ChangelogFeaturedSame story · 2 outlets6060Cursor launches Voice Chat, allowing users to have real-time voice calls with agents, dictate modification requests, ask what the agent is doing, and receive progress updates.
Why it matters: Officially, it explains which platforms the voice call is available on and how to divide tasks with coding agents, allowing readers to judge which collaboration scenarios it suits.
The Verge · AI5858Meta Muse AI has launched the 'Space' feature, allowing users to create customized application experiences such as shopping malls and school dashboards, and can continuously adjust them through descriptions and questions. The feature supports synchronization updates with associated application backends, and users can add 'Space' to the sidebar or main tab and plan to display time. Meta is rolling out this feature to all users and plans to support space sharing and collaboration.
Anthropic ResearchFeatured7272Anthropic announced Project EVE research, allowing Claude to trade in the EVE Online simulated market and complete tasks in a four-person team in EVE Frontier.
Why it matters: Anthropic enabled Claude to trade and team up in EVE, quantifying the generational progress of the model and the weak points of agent team collaboration and destruction.
TechCrunch · AI6060Meta introduces Spaces, enabling users to have breakthrough conversations with Muse intelligent agents around specific topics. These conversations exist as 'custom experiences' in the sidebar of the interactive interface. Clicking on them allows Muse to achieve specific goals for the user, such as planning vacations or collecting recipes, and can be integrated with third-party applications to update information.
TechCrunch · AI5858Security researchers discover hundreds of Chinese AI vulnerability tools Artex exposed on the internet, potentially allowing external attackers to target innocent victims. Artex is an open-source autonomous penetration testing tool that has been used in multiple cyber attacks since its public release in July.
Amp News5858Amp launches Spaces feature, allowing users to have conversations with their teams through a single agent, which is responsible for generating and coordinating other agents, simplifying thread creation and organization. After trying Spaces for a few weeks, the Amp team believes it is the future of interacting with agents, providing a global perspective and necessary detail dives when needed. Spaces has been used in scenarios such as collaborating on new feature development, service account releases and support, and regular security reviews.
OpenAI News2222OpenAI collaborates with Hacktron to find system vulnerabilities, test connections with third-party services, and leverage cutting-edge AI to enhance defenses.
Claude app release notes3535Enterprise organizations can now track cost code expenses through project tags by attaching existing codes such as billing codes or customer case numbers to projects, allowing project expenses to be viewed by tags. This feature is currently in the Beta testing phase.
NVIDIA Technical Blog3535NVIDIA has launched the Reshard Collective technology, which enhances the speed of reinforcement learning strategy model weight reshuffling by 3.5 times through cross-GPU resource coordinated refreshing. This solution supports parallel running of inference and training in distributed GPU architectures, achieving training efficiency optimization. Reshard Collective can be used in the post-training stage of reinforcement learning, reducing latency through dynamic weight synchronization.
MarkTechPost3838Zyphra has released ZYFA (Zyphra FlashAttention on AMD), a set of FP8 and BF16 pre-filled attention kernels optimized for the AMD Instinct MI355X GPU. On real model activations, its FP8 kernel is 1.16-1.76x faster than AMD's default AITER kernel library, and the FP8 error is 62-78% lower than AITER under 128-dimensional heads.
Tesla CEO Elon Musk stated on Friday that the company will cease production of its Cybertruck at the Texas Gigafactory in mid-2027, as the vehicle has seen consistently weak sales since its launch in 2019 and has only been sold in a few markets in the United States.
AWS Machine Learning Blog4242AWS launches serverless Model Context Protocol (MCP) server on Amazon ECS Fargate to address integration challenges of LLM agents accessing corporate data lakes in AI applications. The server achieves metadata-driven, configurable data lake access without the need to write integration code for each dataset, and supports elastic scaling in multi-tenant environments.
