Weekly Piece of Future #157
From Living Buildings to Lab-Grown Retinas and Brain Implants
Hey there, fellow future-addicts!
Welcome to this week's edition of Rushing Robotics! This week, the boundary between science fiction and science fact got a little thinner. From brain implants inspired by paper art to buildings that literally breathe, the pace of discovery isn't slowing — it's compounding.
🤯 Mind-Blowing
This week's Mind-Blowing section is living up to its name in every sense. A flexible brain implant floats on your neurons without tearing them. A decades-old theory about how your eyes develop was quietly dismantled in a lab. Bacteria are being turned into a building material that gets stronger with age. A full 3D object was printed faster than a camera shutter blinks. And an AI found 500 security vulnerabilities that no human had ever noticed — on its own. Each of these stories, on its own, would have been the headline of the decade a generation ago. This week, they all arrived together.
🔊 Industry Insights & Updates
The infrastructure of tomorrow is being laid down fast. Google just pushed reasoning benchmarks into new territory with Gemini 3.1 Pro. Microsoft is storing data inside glass that could outlast civilization as we know it. A 21-year-old founder just gave the robotics world its open-source moment. And NVIDIA is partnering with the Department of Energy to build nuclear reactors at AI speed. The theme connecting all of it: the tools we use to build the future are themselves being rebuilt from the ground up.
🧬 BioTech
Biology continues to be the most audacious engineering discipline on the planet. Dancing molecules are healing lab-grown spinal cords. A single 10-minute DMT infusion is holding severe depression at bay for months. And scientists in Canada are generating electricity from urine — and making it sound like the most logical thing in the world. The lab bench has never felt closer to the clinic, and this week's breakthroughs make that gap look narrower than ever.
💡 Products/Tools of the Week
This week's tools are all about collapsing the gap between idea and output. Whether you're an architect turning a 3D model into a photorealistic render in seconds, a content team wiring together 150+ AI models without writing a single line of code, or an enterprise deploying autonomous AI agents across your entire stack — the common thread is the same: less friction, more creation. These platforms aren't just saving time; they're quietly redefining what a small team can accomplish.
🎥 Video Section
This week's video picks put the humanoid robotics race front and center — from HMND's industrial ambitions to Unitree's New Year celebrations and Boston Dynamics' latest deployment. Watch them back to back and you'll notice something striking: these machines are starting to move less like robots and more like colleagues.
The future isn't some distant destination — it's being assembled in real time, piece by piece, discovery by discovery. Every week that passes, the world we're building looks less like a prediction and more like a promise. Stay hungry, stay futurish!
🤯 Mind-Blowing
For the first time, a brain implant designed to float on neural tissue was built by Chinese scientists using kirigami — the same art form behind elaborate paper cutting — and it recorded 700 neurons firing simultaneously when tested in monkeys. Developed at the Chinese Academy of Sciences, the device wraps its electrodes into coil-like spiral shapes that can stretch, compress, and flex with the brain's heartbeat-driven motion rather than tearing at surrounding tissue. A hydrogel layer coats the implant during insertion to further reduce friction and cushion the device against the brain. Fang Ying, senior researcher at the Chinese Institute for Brain Research in Beijing, said the team discovered the electrode retraction problem about four years ago and began searching for a structural solution — one that geometry, not new materials, could provide. The implications stretch from paralysis treatment to speech restoration and cognitive enhancement.
A decades-old theory about how human eyes develop sharp central vision was overturned by researchers at Johns Hopkins University. Scientists used lab-grown retinal organoids to study the foveola — a tiny pit at the center of the retina responsible for nearly 50% of all visual perception — and discovered that the blue cone cells previously thought to simply migrate away from the region actually transform into red and green cones through a two-step chemical process driven by retinoic acid and thyroid hormones during fetal development. The finding challenges a model that had stood unchallenged in the field for 30 years, and opens the door to lab-grown photoreceptors that could one day restore sight in patients with macular degeneration.
A building material that breathes, heals itself, and eats carbon dioxide was created by scientists, with results published in Nature Communications showing it stayed alive and active for over 400 days. Researchers engineered a 3D-printable hydrogel packed with living cyanobacteria — the same ancient organisms that first oxygenated Earth's atmosphere — and watched it progressively mineralize from the inside out, growing stronger the longer it lived. The bacteria fix CO₂ through photosynthesis while simultaneously triggering calcium carbonate deposits that reinforce the material's own structure, making it the rare building product that improves with age. With no energy input, no toxic byproducts, and nothing but light and nutrients to keep it running, the team behind the study believes it could fundamentally change how the construction industry thinks about carbon.
A complete 3D object was printed in just 0.6 seconds by researchers at Tsinghua University in China, shattering previous speed records while maintaining a resolution finer than one-fifth the diameter of a human hair. The team abandoned traditional layer-by-layer printing entirely, instead developing a technique called DISH — Dynamically Integrated Synthesized Holographic light fields — which uses holographic light projected from multiple angles to solidify an entire 3D structure inside a resin chamber almost instantaneously. Their system reached a printing speed of 333 millimeters per second while holding a feature resolution of just 12 micrometers. The approach works whether the resin is stationary or flowing through a channel, opening possibilities for continuous mass production of complex components.
Over 500 previously unknown high-severity security vulnerabilities were discovered in open-source software by Anthropic's Claude Opus 4.6 with minimal human direction, in findings reported exclusively by Axios. Anthropic's frontier red team gave the model access to Python and standard vulnerability tools — debuggers and fuzzers — but provided no specific instructions or expert knowledge, letting Claude work autonomously. The model identified zero-day flaws ranging from system-crashing bugs to memory corruption vulnerabilities in widely used libraries including GhostScript, OpenSC, and CGIF, all of which were confirmed by Anthropic team members or external security experts. Logan Graham, leader of Anthropic's frontier red team, told Axios the models are already exceptionally proficient at this and will only improve — and that he would not be surprised if AI-driven bug hunting becomes the primary method for securing open-source software in the future.
🔊 Industry Insights & Updates
A major reasoning upgrade to its flagship AI model was released by Google with Gemini 3.1 Pro, arriving just months after Gemini 3 launched in November and bringing record-breaking scores on some of the hardest benchmarks in the field. On Humanity's Last Exam — a test of advanced domain expertise considered one of the toughest AI evaluations in existence — Gemini 3.1 Pro scored 44.4 percent, beating Gemini 3 Pro's 37.5 percent and surpassing OpenAI's GPT 5.2, which scored 34.5 percent. On the ARC-AGI-2 benchmark, which probes genuinely novel reasoning challenges, the new model scored 771 percent higher than its predecessor. Google has kept API pricing flat — $2 per million input tokens — and is rolling the model out across AI Studio, Vertex AI, the Gemini app, and NotebookLM.
Conventional storage is losing a war against time — and the answer Microsoft found was to stop using materials that time can destroy. Project Silica, developed at Microsoft Research Cambridge encodes data permanently inside sheets of borosilicate glass using a femtosecond laser that fires four simultaneous beams, burning microscopic voxels through multiple layers of the material at 65 million bits per second — fitting 4.84 terabytes, the equivalent of roughly two million printed books, into a plate the size of a coaster just 2 millimeters thick. Because glass is chemically inert, it withstands temperatures up to 290°C, complete water submersion, and electromagnetic pulses that would wipe every hard drive and magnetic tape on the premises, requiring no power or maintenance for up to 10,000 years. Melissa Terras, a digital heritage specialist at the University of Edinburgh, welcomed the medium but warned that for data to remain truly accessible millennia from now, the instructions and equipment needed to read the glass must be preserved alongside it — a reminder that the technology is only as durable as the institutions that maintain it.
Humanoid robotics just got its Linux moment. RoboParty, a Beijing-based startup founded in April 2025 by 21-year-old Huang Yi, released its Roboto Origin robot as fully open-source in January 2026 — making it the first complete bipedal humanoid in the world to share its entire technology stack publicly, from structural hardware and supplier lists to motor control software and Sim2Real gap solutions. Built in just 120 days from April to August 2025, the 1.25-meter robot walks at 3 meters per second using a proprietary anthropomorphic gait algorithm and was designed specifically to stress-test full-stack humanoid development with a small team. With $10 million in backing from Xiaomi-affiliated MPCi and a GitHub community already crossing 1,000 stars, founder Huang Yi said his ultimate goal is a world where the best robotic solutions emerge from open collaboration.
A partnership between the Idaho National Laboratory and NVIDIA was launched to use artificial intelligence to cut nuclear reactor build times in half and slash operational costs by over 50 percent. Named Prometheus, the initiative sits inside the Department of Energy's broader Genesis program and deploys generative AI models, digital twins, and agent-based workflows to automate the engineering tasks that currently stretch reactor development across decades. NVIDIA will provide GPU-accelerated infrastructure while INL contributes historical nuclear data, laboratory findings, and experimental reactor operations to train the digital twin models. INL director John Wagner said the collaboration could fundamentally transform how quickly advanced nuclear energy comes online — an urgent need as AI data centers drive electricity demand to record levels.
🧬 BioTech
A lab-grown human spinal cord was healed after injury by researchers at Northwestern University using a therapy called “dancing molecules”. Scientists led by Samuel I. Stupp — the inventor of the therapy and a pioneer in regenerative nanomedicine — grew the most advanced spinal cord organoids ever created from stem cells, incorporated the central nervous system’s own immune cells for the first time, then simulated lacerations and contusions before applying the treatment. The dancing molecules, injected as a liquid that gels into a nanofiber scaffold on contact, rapidly move and vibrate to connect with the brain’s constantly shifting cellular receptors — and the difference was immediate: treated organoids sprouted long neurite extensions while glial scar tissue significantly shrank. The therapy already reversed paralysis in mice in 2021 and recently earned an FDA Orphan Drug Designation, bringing it one critical step closer to human trials.
A single session with one of the most powerful hallucinogenic substances on Earth wiped out severe depression for up to six months in a clinical trial. Scientists at Imperial College London administered DMT — dimethyltryptamine, the psychedelic compound famous for inducing visions of other dimensions and encounters with alien entities — intravenously to 17 adults with moderate to severe major depressive disorder, while the other 17 received a placebo. Those who received the drug showed dramatically greater improvement than the placebo group, with antidepressant effects lasting three to six months from a single 10-minute infusion. Principal investigator Dr. Erritzoe called the result exciting precisely because of its simplicity — one session, one drug, one course of psychological support, and months of relief.
Electricity was generated from human urine by researchers at McGill University in Canada, with findings showing that higher urine concentrations produced greater power output from microbial fuel cells. Professor Vijaya Raghavan and his bioresource engineering team built four dual-chamber microbial fuel cells and tested mixtures of wastewater and urine at concentrations of 20%, 50%, and 70% over two weeks, tracking energy output, pollutant removal, and shifts in the microbial communities inside each cell. The system works by allowing bacteria to oxidize organic compounds in the urine at the anode, releasing electrons that travel through a circuit to generate usable electricity. Raghavan suggested the technology could one day power rural communities or disaster-affected regions while simultaneously treating waste — no grid, no infrastructure, no fuel required.
💡Products/tools of the week
Speed and consistency define what Eler built for architects and designers. The AI rendering platform takes SketchUp or GLB models and produces photorealistic images in seconds by reading geometry directly, skipping prompt engineering altogether. One batch render locks in materials, lighting, and detail across every angle — cutting both the cost and the wait of traditional visualization workflows.
CNAPS Studio, launched in early 2026 and backed by NAVER D2SF and Bluepoint Partners, gives nontechnical users a drag-and-drop canvas to connect 150+ pre-trained text, image, video, and data AI models into production-ready workflows. CEO Innfarn Yoo and the CNAPS team built intelligence mapping at its core — the system automatically determines which models to connect and how, balancing output quality against cost. For e-commerce and content teams, that means rapid prototyping without trial-and-error integration cycles or dedicated engineering resources.
Agentic AI just got an operating system. SimplAI, built by Datafuse Technology, gives enterprises a unified platform to build and deploy autonomous AI agents using drag-and-drop builders, RAG-backed knowledge bases, and multimodal voice interfaces — all without writing code. The platform orchestrates models across OpenAI, Anthropic, Bedrock, and Mistral, routing intelligently to balance performance and cost, while 300+ connectors ground every agent in live enterprise data. LLMOps tooling for tracing, evaluation, and automated QA lets teams catch failures before they reach production, making SimplAI a full-stack solution for industries like insurance, finance, and e-commerce where accuracy and compliance are non-negotiable.
Vaethat is the AI-powered render enhancer that architects, 3D artists, and marketers use to upscale, denoise, and refine architectural renders without rerendering or manual editing. The platform's three presets — Precision, Detail, and Creativity — cover the full range of post-production intent in a single click, while its archviz-trained AI engine ensures geometry, lighting, and material choices survive the enhancement process intact. Batch project processing and a simple upload-and-export flow make Vaethat a direct replacement for the time-intensive Photoshop work or costly render farm reruns that typically end a visualization project.
