Today's briefing
- LLM
Claude Fable 5.2 completes 41 percent of real freelance jobs, up from 2.5 percent a year ago
LLMThe DecoderNov 10, 2026 - Agents
Anthropic launches Claude Meetings to turn workplace discussions into plans and tasks
AgentsTechCrunch +1 moreNov 11, 2026 - Physical AI
Ant Group open-sources LingBot-Touch, a tactile AI model for robots
Physical AIKorean outletiRobotNewsNov 10, 2026
Open source & papers radar
- RepoAgentsRepository: mattpocock/skillsStars: 320,784New stars today: 1,223Language: Shell
Skills, real engineer capabilities from an agents directory
A GitHub repository of skills for real engineers, compiled directly from the creator's .agents directory.
- PaperPhysical AIarXiv 2611.12806Upvotes: 92
LingBot-Touch learns unified tactile representation for general dexterous manipulation
The authors develop LingBot-Touch, a tactile foundation model that creates a unified representation across different tactile sensor types—piezoresistive, magnetic, and visuotactile—and integrates seamlessly into policy models from diffusion policies to vision-language-action models. Extensive experiments on simulation and real-robot deployments show substantial performance gains on complex manipulation tasks while maintaining computational efficiency.
- PaperResearcharXiv 2611.12762Upvotes: 50
Sage-Taiyi, a metacognitive architecture for scientific discovery, tops AI ideation benchmarks
The authors introduce Sage-Taiyi, a metacognition-inspired architecture that guides what to investigate next and how to investigate it by revising understanding of evidence and assumptions. On the Taiyi-Bench AI co-scientist ideation benchmark of 265 tasks from 50 papers, Sage-Taiyi achieves an overall score of 54.0 compared with 47.9 for the strongest baseline, AI Scientist-v2. Expert-guided collaborations validated proposals in chemistry, raising cobalt-catalyzed carbonylation yield from trace amounts to 84% and photocatalytic allylic amination from 13% to 69%.
- PaperResearcharXiv 2611.12886Upvotes: 37
Reframe improves agentic visual reasoning through observation-aligned distillation
The authors introduce Reframe, a post-training framework that aligns the observation state for visual agents using tool-returned information. Ground-Drop-GRPO trains specialists with evidence-aware rewards, while Observation-Gated-MOPD replaces deficient tool actions and resumes distillation from repaired observations. Extensive experiments on 14 image and video benchmarks show Reframe outperforms both Mix-RL and vanilla MOPD.
Korea Desk
What Korea’s tech press is reporting, summarized in English.
5 stories in this issue come from Korean outlets; those outside the desk carry a KOREA mark. Originals are in Korean.
- LLM
Claude Fable 5.2 designs three electronic devices that work without circuit changes
Anthropic's Claude Fable 5.2 completed the full hardware development process for three electronic devices—from circuit design through component layout, firmware writing, and manufacturability review—and produced working prototypes without requiring any circuit modifications.
AI TimesNov 10, 2026 - Physical AI
European semiconductor firms compete for physical AI market dominance
Infineon Technologies, ST Microelectronics, and NXP Semiconductor, major European semiconductor firms, are targeting the global robotics and industrial AI market as a next growth driver. The companies say their accumulated expertise in automotive and industrial semiconductors puts them in a position to compete with U.S. and Asian competitors.
iRobotNews +1 moreNov 10, 2026 - Industry & Policy
SpaceX plans $24 billion AI data center and gas plant in Texas
SpaceX is planning to build a $24 billion AI data center and gas power plant at its Starbase facility in Texas. The project, covering approximately 1.8 million square meters, will include self-contained power generation to reduce grid connection delays and support AI computing demands.
AI TimesNov 10, 2026
Stories in this issue
18 storiesClaude expands to a full workspace with plugins, spreadsheets, whiteboards, and upcoming meeting notes
Anthropic is expanding Claude with Sheets, Whiteboards, and Interactive Connectors that let partners like Adobe and Atlassian embed their apps directly inside Claude. Claude Meetings, which will join calls and take notes, is coming soon.
Anthropic targets $1.4 trillion valuation in IPO
Anthropic's IPO is expected to price between $70 and $78 per share, valuing the company at as much as $1.4 trillion, according to the New York Times. The Wall Street Journal reports that shares are expected to start trading Thursday or Friday next week. This represents a lower expectation than the previously reported $2 trillion valuation.
China blocks SoftBank's $5.4 billion acquisition of ABB Robotics
China has blocked SoftBank Group's $5.4 billion acquisition of ABB's robotics business, dealing a setback to the Japanese technology investor's plans to expand in industrial robotics and physical artificial intelligence.
Claude Fable 5.2 completes 41 percent of real freelance jobs, up from 2.5 percent a year ago
Anthropic's Fable 5.2 completes 41.3 percent of real freelance jobs at professional quality on the Remote Labor Index, more than double the next-best systems Opus 5.5 and GPT 6 Astra. A year ago the top score was 2.5 percent.
Claude Fable 5.2 designs three electronic devices that work without circuit changes
Anthropic's Claude Fable 5.2 completed the full hardware development process for three electronic devices—from circuit design through component layout, firmware writing, and manufacturability review—and produced working prototypes without requiring any circuit modifications.
Boston Dynamics reveals wheeled Atlas performing 360-degree spinning jumps
Boston Dynamics unveiled its humanoid robot Atlas fitted with wheels instead of feet, using reinforcement learning-based control to perform rapid movement, directional changes, and 360-degree rotating jumps. The experiment validates the performance of the robot's actuators and AI control model.
Skills, real engineer capabilities from an agents directory
A GitHub repository of skills for real engineers, compiled directly from the creator's .agents directory.
LingBot-Touch learns unified tactile representation for general dexterous manipulation
The authors develop LingBot-Touch, a tactile foundation model that creates a unified representation across different tactile sensor types—piezoresistive, magnetic, and visuotactile—and integrates seamlessly into policy models from diffusion policies to vision-language-action models. Extensive experiments on simulation and real-robot deployments show substantial performance gains on complex manipulation tasks while maintaining computational efficiency.
Sage-Taiyi, a metacognitive architecture for scientific discovery, tops AI ideation benchmarks
The authors introduce Sage-Taiyi, a metacognition-inspired architecture that guides what to investigate next and how to investigate it by revising understanding of evidence and assumptions. On the Taiyi-Bench AI co-scientist ideation benchmark of 265 tasks from 50 papers, Sage-Taiyi achieves an overall score of 54.0 compared with 47.9 for the strongest baseline, AI Scientist-v2. Expert-guided collaborations validated proposals in chemistry, raising cobalt-catalyzed carbonylation yield from trace amounts to 84% and photocatalytic allylic amination from 13% to 69%.
Anthropic launches Claude Meetings to turn workplace discussions into plans and tasks
Anthropic launched Claude Meetings, which drafts plans during discussions, maps timelines on whiteboards, answers questions in meeting chat, and accesses connected work apps to retrieve information such as historical figures.
Anthropic launches Claude Fable 5.2 and Mythos 5.2 with different safeguard levels
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, one model with two safeguard levels. Fable 5.2 scores 49.1% on FrontierCode 2.0, costs $8/$40 per million tokens, and includes a 1-million-token context window.
Ant Group open-sources LingBot-Touch, a tactile AI model for robots
Ant Group's embedded AI development organization Robbyant open-sourced LingBot-Touch, a robotic tactile AI model that processes data from different types of tactile sensors. The model converts information from magnetic, vision-based, and pressure-sensitive tactile sensors into a unified representation.
European semiconductor firms compete for physical AI market dominance
Infineon Technologies, ST Microelectronics, and NXP Semiconductor, major European semiconductor firms, are targeting the global robotics and industrial AI market as a next growth driver. The companies say their accumulated expertise in automotive and industrial semiconductors puts them in a position to compete with U.S. and Asian competitors.
Path Robotics raises $300 million Series E at $2.45 billion valuation
Path Robotics closed a $300 million Series E funding round at a valuation of $2.45 billion to scale physical AI for manufacturing.
SpaceX plans $24 billion AI data center and gas plant in Texas
SpaceX is planning to build a $24 billion AI data center and gas power plant at its Starbase facility in Texas. The project, covering approximately 1.8 million square meters, will include self-contained power generation to reduce grid connection delays and support AI computing demands.
Reframe improves agentic visual reasoning through observation-aligned distillation
The authors introduce Reframe, a post-training framework that aligns the observation state for visual agents using tool-returned information. Ground-Drop-GRPO trains specialists with evidence-aware rewards, while Observation-Gated-MOPD replaces deficient tool actions and resumes distillation from repaired observations. Extensive experiments on 14 image and video benchmarks show Reframe outperforms both Mix-RL and vanilla MOPD.
New York's AI safety law takes effect, adding oversight for OpenAI and Anthropic
New York is preparing to enforce its sweeping AI safety law, requiring major developers like OpenAI and Anthropic to register, disclose safety practices, and report critical incidents or face fines.
Google releases Gemini 4 Carbon, ranking first on several benchmarks at less than half the cost
Google made Gemini 4 Carbon generally available after a two-week testing phase. The model ranks first in several benchmarks, in some cases ahead of Anthropic's Fable 5.2, at less than half the per-task cost. Google is also previewing coordinated agent teams through the '/swarm' feature, which splits up complex tasks in real time.