Notes
Quick takes on AI and robotics news.
See how these notes are created-
Neural chip decodes brain networks on 200 microwattsMost closed-loop brain implants monitor activity in a single area, but psychiatric disorders rarely originate from a single spot.
-
Physical AI tackles retail returns on standard cobot armsHandling customer returns is one of the messiest operational problems in logistics because items rarely arrive in predictable orientations, shapes, or packaging.
- Anthropic moves Agent Skills to open governance
Anthropic is handing its Agent Skills standard over to the Agentic AI Foundation, moving the format for packaging procedural workflows under the Linux Foundation alongside the Model Context Protocol.
-
Enterprise AI gains stall outside the tech sectorApollo chief economist Torsten Slok reports that revenue per worker and profit margins have climbed for the largest tech companies, but remain flat across the rest of the S&P 500.
-
Quadruped feet handle contact tasks without dedicated armsMounting an articulated arm on a quadruped adds weight, drains battery life, and increases hardware cost.
- Blender MCP turns generated 3D into editable scenes
Most generative 3D tools hand artists a dense, uneditable mesh with lighting baked into the surface.
-
Humanoid models cannot learn contact physics from the webLanguage models scaled because the open internet provided trillions of tokens of text for free, but robotics has no pre-existing archive of physical work.
- OpenShell puts agent containment in the kernel
Giving an AI agent real work usually requires letting it read files, install packages, and talk to external services, but handing that agent ambient production credentials makes it an immediate security risk.
- Global forest mapping moves past binary canopy cover
Satellite monitoring often flattens ecology into a binary measure: whether an area has tree canopy or bare ground.
- Isaac Lab 3.0 decouples robot learning from single solvers
Most robot learning frameworks lock training to a single simulation engine, forcing teams to accept whatever physics approximations the platform provides.
-
Task automation models run into organizational dragAnthropic's economics team broke down the US economy into bundles of occupational tasks to model how AI changes output by 2030, publishing their technical report and interactive explorer at Anthropic.
- Testing engineering agents against real physics solvers
Industrial design and manufacturing tolerate almost no guesswork.
- Automated 3D data fusion still requires verification
Google DeepMind demonstrated Gemini 4 Argon combining three open-source datasets into an interactive 3D visualization of human anatomy.
-
Hugging Face puts TPUs one device string awayRunning a Hugging Face model on a Google TPU used to mean porting your PyTorch code to torchxla or learning JAX.
-
Claude's Perseverance route ran through JPL's daily simulatorOn December 8 and 10, 2025 (sols 1707 and 1709), NASA's Perseverance rover drove about 400 meters through a rock field along waypoints planned by Claude.
-
Amazon let AVERI grade its safety tests, not its modelsOutside reviews of AI models usually look at what the model says. In a post on Amazon Science, Rahul Gupta and Jwala Dhamala describe a different setup.
-
Mistral's Voxtral TTS copies a voice from three secondsMistral says Voxtral TTS, released March 23, 2026, can adapt to a new speaker from a reference clip as short as three seconds. It picks up accent, intonation, and even the speaker's hesitations.
-
Mistral's connectors move agent permissions into configMistral's case for Connectors in Studio, released in public preview on May 22, 2026, starts from a familiar complaint.
-
Agility's Digit 5 trades grace for uptimeJonathan Hurst spent more than a decade building legged robots that move like animals, and the humanoid his company just shipped is, in his own words, heavy, bulky, and missing the natural motion of its predecessors.
- Microsoft wants agent security to read model reasoning
Microsoft's case for securing AI agents starts from an uncomfortable premise: the agents a company deployed itself are insiders it can't fully trust.