
Know AI. See What’s Next.
Slide 1 of 5: 姜宇辉的AI分身:知识助手与“另一个自己”之间
Latest news

RoboSpoof Tests How Visual Interference Affects Robots' Safety Refusals (opens in a new tab)
Robocurve compared original, blurred and AI-tampered images across 900 trials with five models and five tasks, reporting a sharp increase in dangerous-action attempts under controlled, prop-based conditions.

Hygon outlines agent compute architecture with separate orchestration, inference and memory nodes (opens in a new tab)
The report covers Chinese CPU/DCU platforms, modular data centers and AgentServer Flex private deployment, examining how to match resources to different agent workloads.

Tsinghua survey reviews embodied autonomous systems, data and evaluation through an agent lens (opens in a new tab)
The survey organizes research around perception, planning, execution and feedback, proposes levels of human-machine responsibility, and discusses cloud-edge-device coordination, safety and evidence for cross-task validation.

Baidu Famou 3.0 debuts, bringing continuous optimization into enterprise R&D and production (opens in a new tab)
Baidu World presents materials, chip and apparel cases that connect data, algorithm search, independent validation and human review in feedback loops. Results are those disclosed by the teams.

HKUST assigns image IDs for tool reuse and evolves training questions with ODE (opens in a new tab)
The visual-native harness lets tools reuse retrieved and cropped images; the team reports that ODE training raises an 8B model's eight-evaluation average accuracy from 24.9% to 39.0%.

Peking University Team Reviews a Decade of Deep Time-Series Forecasting Through Autocorrelation (opens in a new tab)
A TPAMI survey unifies backbone architectures and loss functions around autocorrelation, comparing dependency modeling in input and label sequences and discussing missing data, noise and distribution shift.

DeepSeek Expands Its Application APIs With Search, File Parsing and Voice Services (opens in a new tab)
The new interfaces expose web search, webpage reading, document and voice services to developers, alongside an open-source chat demo. Model tokens and tool calls incur separate charges.

Atlas opens its web app and API; APPSO tries 3D scene generation and export (opens in a new tab)
World Labs connects generation, set design, lighting and mesh export into a toolchain. APPSO describes sketch and Chongqing-style scenes, while noting that finished films or games still require further work.

Huiwen Wang's family office reportedly backs AirBuild, focused on iOS app development and submission (opens in a new tab)
AirBuild coordinates planning, design, coding and App Store review feedback through multiple agents for independent developers. Investment terms are undisclosed, and automated submission does not guarantee approval.

Yisuo Intelligence launches Sciland to connect scientific computing, experiments and feedback (opens in a new tab)
Sciland uses LAEP to link computational tasks and lab equipment, adopting system changes through replay, regression tests and human checks. Deployment and performance figures are company-reported.

Inside Silicon Valley's Startup 'Monastery' HF0 and OpenRouter's Early Days (opens in a new tab)
ACE Studio founder Guo Jing recalls his 2024 HF0 residency, the $800,000 monthly revenue peak and subsequent decline, and the focus and growth of OpenRouter's small early team.

Wujie Fangzhou releases Edge0-145B and inference engine to explore local laptop use (opens in a new tab)
The team reports sparse architecture, tiered storage and hardware-specific memory and speed tests, with coding and 3D demos. Performance depends on hardware, context and tools.

Leiphone reports Li Muqing left Tencent to become Qiming Daren co-CEO (opens in a new tab)
The report traces his roles at Yitu, Tencent Youtu and Tencent Cloud, saying he joined the education AI company after leaving Tencent in September to advance its business.

Tractian announces $130 million Series D to expand industrial equipment health monitoring (opens in a new tab)
The company combines sensor signals with AI diagnoses to recommend maintenance actions. Its team and investors seek broader industrial coverage while facing the risks of real-world operations.

a16z on B2A: agent-driven opportunities in product distribution and infrastructure (opens in a new tab)
Zhidongxi summarizes a16z's views on agent apps, product optimization, payments and identity services, alongside company-reported usage, order and deployment figures.

Ropedia releases Xperience 2 to align real scenes with first-person behavior (opens in a new tab)
Scene geometry, video and body trajectories share one coordinate system. Demos cover simulation conversion, retargeting, virtual teleoperation and trajectory generation; benefits for real-robot training need further tests.

Google adds TPU 8tx and 8i Max for confidential computing and long-context tasks (opens in a new tab)
Google describes new eighth-generation TPU variants expected in 2027. The report covers customer needs, supply constraints and external sales; its 2028 revenue figures are institutional forecasts.

RoboScience introduces cloud-based embodied agent Visics-Agent, plans December 2026 launch (opens in a new tab)
The team demonstrates voice teaching, skill accumulation and multiple end effectors, reporting tests of 454 internal task instructions. First trial applications are open; formal launch is planned for December 2026.

Study Uses Astra to Edit 65 Token-Embedding Rows, Improving Robot Instruction Understanding (opens in a new tab)
The best edit raised held-out simulation success from 86.3% to 92.0% across 765 episodes, with four regressions. Real-robot transfer remains unverified.

Ant Lingguang CTO on personal agents: goals, ecosystems and the boundaries of trust (opens in a new tab)
Silicon Valley 101 discusses Lingguang's shift from Flash Apps to ongoing goals, Alipay integration, and unresolved questions around memory, proactive reminders, permission checks and monetization.



