Today's briefing
- Industry & Policy
Google unveils TPU 8i Max and TPU 8tx: a dual-chip TPU and a confidential TPU
Industry & PolicyMarkTechPostNov 10, 2026 - Physical AI
Nvidia open-sources VFlash, making robot AI inference 5x faster
Physical AIKorean outletAI TimesNov 9, 2026 - LLM
Bytedance Seedream 5.5 offers stable image editing over up to 100 rounds without color drift
LLMThe DecoderNov 10, 2026
Open source & papers radar
- RepoOpen Source & ToolsRepository: ZhuLinsen/daily_stock_analysisStars: 68,752New stars today: 505Language: Python
daily_stock_analysis, an LLM-powered multi-market stock analysis system
daily_stock_analysis is an LLM-driven system that analyzes stocks across multiple markets. It integrates multi-source market data, real-time news, a decision dashboard, and automated notifications, and supports cost-free scheduled runs.
- RepoAgentsRepository: homeassistant-ai/ha-mcpStars: 5,853New stars today: 31Language: Python
ha-mcp enables Home Assistant configuration with Claude and ChatGPT
ha-mcp is an open-source project that configures and automates Home Assistant using Claude, ChatGPT, and other AI assistants. It lets users set up smart-home rules and automations in natural language.
- PaperPhysical AIarXiv 2611.08175Upvotes: 18
PCBRouteAgent coordinates global and local routing to achieve 40–100% clean-pass rates on real PCB boards
The authors present PCBRouteAgent, a multimodal agent that coordinates global routing strategy with local execution for automated PCB routing. On real-world test subsets, it achieves clean-pass rates of 40–100% and outperforms evaluated baselines by 13.9–36 percentage points.
- PaperAgentsarXiv 2611.07461Upvotes: 27
HEELO uses agent-native primitives to automate real-world tasks
The authors propose agent-native primitives, reusable units equipped with a resident LLM that applies resources in context and communicates with other primitives. They introduce HEELO, a framework that composes these primitives into task-specific graphs and revises them based on environmental feedback. On Terminal-Bench 4.0, the approach raises resolution from 37.3% to 55.5%, and on 46 real-world tasks, Claude Opus 5.5 with Claude Code and GPT-6.1 Sol with Codex complete 39 and 35 tasks respectively when connected to HEELO, compared to 9 each on their own.
Korea Desk
What Korea’s tech press is reporting, summarized in English.
4 stories in this issue come from Korean outlets; those outside the desk carry a KOREA mark. Originals are in Korean.
- Physical AI
Embodied AI unveils AIR, a wheeled humanoid robot for collaborative industrial work
Swiss robotics startup Embodied AI unveiled AIR, a wheeled humanoid robot designed to work alongside human operators in industrial settings. The robot concentrates heavy actuators in the torso and uses lightweight arms to reduce collision risk with workers.
iRobotNewsNov 9, 2026 - Physical AI
Legged robots move into industrial operations; industry calls for safety and certification standards
Four-legged and two-legged robots are now being deployed for actual industrial work, including equipment inspection, patrol and welding, beyond pilot programs. Industry identifies accumulating field data and establishing safety and certification standards as key priorities.
ETNewsNov 9, 2026 - Research
Physical Super Intelligence proves 18-year-old quantum conjecture using private Anthropic model
Physical Super Intelligence used a private Anthropic model with its own research system to prove the Aaronson-Ambainis conjecture in full. The quantum computing problem had remained unresolved for 18 years, and the AI system both produced a mathematical proof and passed computer verification.
AI TimesNov 9, 2026
Stories in this issue
22 storiesAnthropic buys $3.4 billion worth of natural gas turbines
Anthropic purchased billions of dollars' worth of natural gas turbines in a deal that could give it buyer's remorse.
Google ships four TPU variants in one year as it ramps up competition with Nvidia
Google expanded its TPU 8 family with two specialized variants: the TPU 8tx keeps sensitive data encrypted during processing, while the TPU 8i Max doubles compute power and memory by placing two chips on a single package. With nearly 100 external customers, Google is increasingly positioning its chips as an alternative to Nvidia's GPUs.
Meta launches Muse Ember, a small AI model built for repetitive agent tasks
Meta introduced Muse Ember, its smallest and most efficient AI model, optimized for repetitive agent tasks such as tool calls, routing, and data extraction. According to Meta, it delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna, though it falls short on actual programming tasks.
Nature Will launches 50-billion-parameter physical general foundation model Natura-1
Nature Will, a physical AI lab, announced Natura-1, a 50-billion-parameter physical general foundation model for dexterous hand control and zero-shot instruction following in advanced robotics. The model was pre-trained on over 1 million hours of egocentric human data.
Google unveils TPU 8i Max and TPU 8tx: a dual-chip TPU and a confidential TPU
Google added TPU 8i Max and TPU 8tx to its eighth-generation TPU lineup. TPU 8i Max packs two TPU 8i chips per package for long-context agents, while TPU 8tx adds hardware confidential inference. According to Bloomberg, both will be available in 2027.
Nvidia open-sources VFlash, making robot AI inference 5x faster
Nvidia unveiled VFlash, a new decoding framework that accelerates visual planning for physical AI robots and autonomous vehicles. By predicting entire motion paths at once instead of generating coordinates sequentially, VFlash speeds up robot motion planning inference by 5x while reducing prediction errors.
Nvidia helps Korea advance physical AI across industries
Nvidia is advancing Korea's physical AI ecosystem by tapping into the country's strengths in manufacturing and engineering. Korean industry leaders are applying accelerated computing, AI, and simulation to semiconductor manufacturing, robotics, and autonomous vehicles, helping bring physical AI from development into real-world production.
daily_stock_analysis, an LLM-powered multi-market stock analysis system
daily_stock_analysis is an LLM-driven system that analyzes stocks across multiple markets. It integrates multi-source market data, real-time news, a decision dashboard, and automated notifications, and supports cost-free scheduled runs.
ha-mcp enables Home Assistant configuration with Claude and ChatGPT
ha-mcp is an open-source project that configures and automates Home Assistant using Claude, ChatGPT, and other AI assistants. It lets users set up smart-home rules and automations in natural language.
PCBRouteAgent coordinates global and local routing to achieve 40–100% clean-pass rates on real PCB boards
The authors present PCBRouteAgent, a multimodal agent that coordinates global routing strategy with local execution for automated PCB routing. On real-world test subsets, it achieves clean-pass rates of 40–100% and outperforms evaluated baselines by 13.9–36 percentage points.
Bytedance Seedream 5.5 offers stable image editing over up to 100 rounds without color drift
Bytedance Seed introduced Seedream 5.5, an image model for professional production workflows. It supports stable multi-pass editing over up to 100 iterations, generates native RGBA images with transparent backgrounds, and the company says it outperforms GPT-Image-2.5 Sunburst in realism.
Baidu ERNIE Evolve beats NanoChat and NanoGPT records with automated code experimentation
Baidu's ERNIE Evolve is an automated research system that edits code, runs experiments and keeps only verified gains. It beat public records on NanoChat and NanoGPT Speedrun and sped up ERNIE-5.5-Flash inference.
Embodied AI unveils AIR, a wheeled humanoid robot for collaborative industrial work
Swiss robotics startup Embodied AI unveiled AIR, a wheeled humanoid robot designed to work alongside human operators in industrial settings. The robot concentrates heavy actuators in the torso and uses lightweight arms to reduce collision risk with workers.
Legged robots move into industrial operations; industry calls for safety and certification standards
Four-legged and two-legged robots are now being deployed for actual industrial work, including equipment inspection, patrol and welding, beyond pilot programs. Industry identifies accumulating field data and establishing safety and certification standards as key priorities.
Physical Super Intelligence proves 18-year-old quantum conjecture using private Anthropic model
Physical Super Intelligence used a private Anthropic model with its own research system to prove the Aaronson-Ambainis conjecture in full. The quantum computing problem had remained unresolved for 18 years, and the AI system both produced a mathematical proof and passed computer verification.
HEELO uses agent-native primitives to automate real-world tasks
The authors propose agent-native primitives, reusable units equipped with a resident LLM that applies resources in context and communicates with other primitives. They introduce HEELO, a framework that composes these primitives into task-specific graphs and revises them based on environmental feedback. On Terminal-Bench 4.0, the approach raises resolution from 37.3% to 55.5%, and on 46 real-world tasks, Claude Opus 5.5 with Claude Code and GPT-6.1 Sol with Codex complete 39 and 35 tasks respectively when connected to HEELO, compared to 9 each on their own.
Higher agent harness scores do not guarantee reliable, efficient and transferable improvement
The authors examine whether higher benchmark scores from agent harness evolution yield reliable, efficient and transferable improvement. They find that all sixteen evolved agents gain previously uncovered tasks while losing previously covered ones, and reliability gains typically increase deployment cost. The highest-scoring agent does not yield the largest mean improvement in any of the four test pools.
EvoWorld improves causal reasoning in generative video models through world state evolution
The authors address how generative video models can reason about world state changes when events occur, not just render plausible pixels. They introduce EvoWorld, a framework where a vision-language model acts as a world brain to observe segments, maintain an evolving world account, and reason about state transitions, while a video renderer realizes them visually. They also propose the Out-of-Sight Causal Evolution task and introduce EvoWorld-Data and EvoWorld-Bench for training and evaluation.
Reason Machines open-sources Aspen, a Git server for AI agents
Reason Machines open-sourced Aspen, a self-hosted Git server for AI agents. It follows OpenAI's Pando design and sustained 567 pushes per second on a single 8-core server in the company's own test.
0sec Labs open-sources 0, an AI security harness for code review and penetration testing
0sec Labs open-sourced 0, an AI security harness for code review, penetration testing, reverse engineering and fixes. It runs locally through a browser console, a headless CLI and an MCP server, with your own model provider.
MemoryDojo benchmarks poisoning attacks on LLM agent memory systems
The authors introduce MemoryDojo, a benchmark that traces how poisoned information flows through agent memory systems and measures its impact on subsequent agent behavior. The benchmark covers 155 multi-session trajectories across four domains and supports five different memory systems. They find that all memory systems are susceptible to poisoning, with substantial variation in downstream effects, and propose a memory-aware poison optimization process that reveals differences between architectures.
Learning high-speed sprinting and aggressive cornering on a full-scale humanoid robot
The authors present a learning framework that enables a bipedal humanoid to sprint above 7 m/s and perform high-speed cornering with up to 25 degrees trunk lean. They use causal multi-head attention to model high-frequency feedback and design a heading-direction reward that supervises low-frequency velocity direction rather than yaw rate, allowing agile turning without suppressing natural gait oscillations. The framework is validated in both simulation and real-robot deployment.