Deep Tech Dispatch | #16
Rare earth recycling | Folding proteins with quantum | Robots thinking ahead | Making robots feel | Kinetic launch gun | Space elevators and much more
🙋🏻♂️ A Short Preface
♨️ Recycling rare earth elements with microwaves
👨🔬 Folding proteins with quantum computers
🤖 Robots thinking ahead
🫳🏻 Giving robots human-like sense of touch with eFlesh
⛏️ Urban mining in Europe
🌊 China extracts rubidium from brine
🚀 Honda enters the space race
☢️ A policy update to test novel nuclear reactors
💥 Kinetic space launch prototype
🔋 Deep geothermal energy on track to power the world
🧠 Your Brain on ChatGPT
🪐 Space elevators on Ceres
Reading time: 15 minutes.
🧑🏻🔬 Science & Research News
Recycling rare earth elements with microwaves
There is no edition of this newsletter without touching minerals. This time around researchers have developed a new recycling method that uses microwaves to efficiently recover valuable and strategic metals from electronic waste. The process successfully extracts tantalum and manganese from old capacitors and is significantly cheaper and faster than other traditional methods that involve for example industrial furnaces.
Tantalum is a critical "conflict mineral" that is essential for nearly all modern electronics, from smartphones to laptops. Its supply chain is volatile and fraught with ethical concerns. By creating an effective way to recycle it from the millions of tons of e-waste generated each year, this method could create a more stable and cost-effective supply chain for the electronics industry. It is worth noting here that there are little to no sources for tantalum as well as manganese within Europe nor the US - with the exception of a small resource pool in Ukraine. The western world is heavily dependent on supply from Africa, South America and Australia.
The process itself works through a so called ‘carbothermal reduction’: end-of-life capacitors are crushed into powder and mixed with carbon, heated in a standard microwave oven (yes, you can kind of build this yourself if you want) resulting in intense heat spots as the carbon absorbs the microwaves which then triggers a chemical reaction. The carbon strips away oxygen from the metal powder and the result is high-purity tantalum carbide and manganese-rich slag.
The project is currently funded by DARPA and the US military and has proven to be working on a lab scale, with industrialisation of the process being the next milestone. You can find the article here and the paper here.
Folding proteins with quantum computers
A collaboration between the quantum computing companies Kipu Quantum and IonQ has achieved a new milestone in computational biology, successfully using a trapped-ion quantum computer to solve the most complex protein folding problems ever tackled on quantum hardware.
A protein's function is dictated by its intricate 3D shape, and when it misfolds, it can lead to a loss of its normal function or the acquisition of a toxic function, like neurodegenerative diseases. While AI tools like AlphaFold have made huge strides, predicting how a protein will fold from its amino acid sequence is an astronomically complex optimisation problem with a near endless solution space. Modelling this complex optimisation problem on a quantum computer opens a new door for drug discovery, allowing scientists to design novel drugs that target specific protein shapes with greater precision.
The experiment itself was run on IonQ’s 36-qubit trapped-ion quantum computer and a specialised algorithm developed by Kipu Quantum called BF-DCQO (Bias-Field Digitized Counterdiabatic Quantum Optimization). Here’s the simple version of how it works: Folding proteins can be represented as an optimisation problem that searches for the lowest energy state, i.e. a stable ‘valley’ that represents the sought after protein. Mapping this complex energy map onto a quantum computer makes use of quantum’s inherent advantage to efficiently find these low optima. The different states of the qubits thereby represent different possible folds of the protein. The algorithm itself then guides the quantum computer towards its lowest energy state, revealing the optimally folded protein.
The team successfully solved folding problems for three biologically important peptides with up to 12 amino acids—a new record for this type of demonstration on quantum hardware. It slowly starts to feel that quantum actually provides real value. You can find the article here and the paper here.
Robots thinking ahead
Researchers from MIT and NVIDIA have developed a new algorithm called cuTAMP that allows robots to solve complex long-horizon manipulation and planning problems within seconds, making robots not only faster but also making the human-robot interaction feel much more natural and seamless.
Planning long-horizon robot manipulation is a challenging task that requires discrete decisions about which objects to interact with and continuous decisions how to interact with them. This dual challenge within Task and Motion Planning (TAMP) of not only figuring out a high-level sequence of actions (the "task") but also the precise, low-level joint movements and gripper orientations required to execute them (the "motion") is computationally demanding in terms of runtime but also solution quality.
This new algorithm cuTAMP leverages GPU parallelism to explore all to potential solutions simultaneously, accelerating a robot’s ability to ‘think ahead’ - a task that previously took minutes or would’ve even be impossible. This works by intelligently sampling and narrowing down the potential solution space to highly likely actions and parallel optimisation of the potential plans through GPUs. All in all this tiny but mighty algorithm will speed-up a lot of applications within robotics and also - again - will make robots more human. You can find the news article here and the paper here.
Giving robots human-like sense of touch with eFlesh
A new open-source platform called eFlesh is aiming to give robots a human-like sense of touch. Developed by researchers at New York University, eFlesh is a low-cost, highly customisable magnetic tactile sensor that can be fabricated by anyone with access to a 3D printer.
This novel type of sensor gives robots a fine-grained sense of touch, critical for operating effectively in unstructured environments like homes and offices. With eFlesh, robots could handle delicate objects with greater precision, detect slip, and perform complex manipulation tasks that are currently impossible - just head over to their website, the demonstrated tasks are really impressive.
The principle behind eFlesh is quite simple: the 3D printed microstructure determines the sensor’s mechanical properties, like its stiffness and resulting sensitivity. Small pouches for off-the-shelf magnets are placed within the 3D structure and a simple magnetometer circuit board allows a readout of the sensor, detecting how the magnetic field changes through deformation and thus movement of the magnets. This simple, yet powerful sensor can improve all kinds of robotics applications ranging from manufacturing and warehouse logistics to in-home assistance and even robotic surgery. You can find website of the project here and the paper here.
🔬 Technology Frontiers
Urban mining in Europe
The four-year, EU-funded project iBot4CRMs, aims to tackle Europe's heavy reliance on imported materials and build a more resilient domestic supply chain by capturing critical raw materials (CRMs) from electronic waste piling up within Europe. Currently, the vast majority of critical minerals like neodymium, cobalt, copper, and gold get imported from a few concentrated regions.
Why "urban mining" is so fascinating is that the concentration of these critical materials in e-waste is often much higher than in natural ores—a ton of old mobile phones can contain 80 times more gold than a ton of ore from a mine. By creating an efficient way to recover these materials, the project can provide a stable and cheap sources of these materials.
The tech behind the project is a combination of robotics and computer vision. Waste streams get sorted, dismantled and stripped of their valuable parts. This precise approach is a lot more efficient than manual sorting or bulk shredding of e-waste.
The project brings together 18 partners from across Europe, including technology providers and recycling companies, and will be tested in four real-world pilot trials. You can find more details in the article here.
China extracts rubidium from brine
Adding another chapter to the ongoing "rare metals war," the Chinese Qinghai Institute of Salt Lakes has successfully extracted ultra-pure rubidium directly from salt lake brine for the first time at industrial scale.
Rubidium is a critical metal used in a host of high-tech and military applications, from ultra-precise atomic clocks and GPS systems to night vision equipment and aerospace components. Despite holding large reserves, China has been heavily dependent on imports (mostly from Canada) because its domestic sources are low-grade and difficult to process. This breakthrough significantly boosts its potential for self-sufficiency in a mineral that the US has officially listed as critical.
The developed process itself involves washing, leaching, enrichment and purification of a naturally occurring mixture of potassium and rubidium in brine within Qinghai's Qarhan Salt Lake. It is worth noting here that because of the low concentration of rubidium and its chemical similarity to potassium it was previously not possible to make use of that brine. You can find the news article here.
Honda enters the space race
With an unexpected move, Honda enters the global space race. With their entirely in-house developed reusable rocket, Honda successfully launched and landed their rocket after reaching an altitude of nearly 300 meters.
The recent test in Hokkaido, Japan, was a textbook "hop test," similar to the "Grasshopper" tests SpaceX performed in its early days. The rocket itself is thereby relatively small measuring 6.3 meters and is intended to launch small satellites for communications and Earth observation into orbit. The reusability of the rocket is thereby key to drive down cost longterm.
While unexpected, I think this is a clever move of an industrial giant that has all the necessary ingredients to enter the space industry, from combustion engines to advanced materials. Maybe we can repurpose some of the German automotive industry for space? One can dream. You can find more details in the articles from Honda and Forbes.
A policy update to test novel nuclear reactors
While not strictly a technology breakthrough, I thought this is quite fitting nonetheless. In a major policy shift designed to slash red tape, the U.S. Department of Energy has launched a new pilot program to allow private companies to build and test advanced nuclear reactors outside of national laboratories. This move is designed to dramatically accelerate the development of next-generation nuclear technologies by unleashing private sector innovation and funding.
For decades, testing new reactor concepts was a slow, complex process largely confined to government-run facilities. This new pathway effectively creates a "fast lane" for nuclear startups, allowing them to build, test, and iterate on their designs much more quickly and on their own turf. It's a clear signal that the U.S. is serious about reclaiming a leadership role in nuclear energy, aiming to get new, safer, and more efficient reactors faster to market. You can find the article here.
Kinetic space launch prototype
Longshot Space has unveiled their newest prototype to shoot payloads into orbit: a giant gas-powered cannon. The company is betting on a "kinetic launch" system that would use compressed gas to accelerate a 100 kg projectile to hypersonic speeds, flinging it into space before a small, second-stage rocket kicks in to achieve a stable orbit.
While rockets have become reusable, the marginal cost of each launch is still significant. Longshot’s approach aims to dramatically slash that cost by doing most of the hard work - getting out of the thickest part of the atmosphere - on the ground with electricity and compressed gas. They project a potential cost of just $150 per kg to orbit, a price point that could fundamentally change the economics of space transportation and make regular, high-volume launches much more feasible.
At an investor day last week, the team showed off a 70-ft long prototype that accelerates payloads to just past Mach 4. Now, they are building a 180-ft version that will pave the way for an 1800-ft gun suited for testing military hypersonics at speeds above Mach 5. Weapons researchers today might pay $6M to $8M to put materials or components in that environment, according to Grace, who said his company could do it for $150,000.
Deep geothermal energy on track to power the world
Quaise Energy has successfully demonstrated its novel rock-vaporising drill technology on a full-scale oil rig, a major step forward in its quest to unlock nearly limitless geothermal energy from deep within the Earth. The company aims to drill deeper than ever before using high-power microwaves, tapping into "superhot" geothermal resources deep within the Earth’s crust.
Traditional geothermal power has been limited to specific geographic locations, but by drilling miles deep, Quaise could make clean, baseload geothermal energy widespread available. But because of the required depth, drilling techniques have been limited in what they could achieve. Quaise gets around this by using a gyrotron to generate powerful millimeter waves (a type of microwave) to literally melt and vaporise the rock in its path.
In their recent field test near Houston, the team successfully integrated their system with a traditional oil rig and used it to deepen a hole in a granite column, proving their technology works outside the lab. With this major milestone achieved, Quaise is now focused on scaling up its gyrotron to a commercially relevant one-megawatt system within the next two years. You can find more details in their press release here.
💡 Articles & Ideas
Your Brain on ChatGPT
Relying on ChatGPT (or any other LLM) instead of your own brain significantly decreases your brain connectivity and thus cognitive abilities. I think everyone kind of knew intuitively that this was the case but thanks to researchers from MIT we now have it black on white. To be fair though, their experimental setup focusses on essay writing in an educational context and their results are not generalisable across everything, but it gives a good indication to what happens when you offload everything intellectually challenging into an LLM: your brain receives less stimulation and training. The longterm effects? Nobody knows. My intuition: people forget how to think for themselves. A new generation growing up on a good mix of TikTok and LLMs - what could go wrong? You can find the research here.
As the educational impact of LLM use only begins to settle with the general population, in this study we demonstrate the pressing matter of a likely decrease in learning skills based on the results of our study. The use of LLM had a measurable impact on participants, and while the benefits were initially apparent, as we demonstrated over the course of 4 months, the LLM group's participants performed worse than their counterparts in the Brain-only group at all levels: neural, linguistic, scoring
🤔 Odd Read of the Week
Space elevators on Ceres
Yes, the wonderful topic of building elevators from planets into outer space. For those of you that never heard of space elevators:
Almost every part of a rocket is destroyed during the launch and re-entry into the Earth's atmosphere. This makes spaceflight really expensive. Rocket delivery of even a single kilogram into orbit costs tens of thousands of dollars. But what if we could just place our payloads directly into orbit, and didn't need a rocket at all?
This is the idea of a space elevator, first envisioned by the Russian rocket scientist Konstantin Tsiolkovsky in 1895. Tsiolkovsky suggested building a tower all the way up to geostationary orbit, this is the point where a satellite appears to hang motionless in the sky above the Earth. If you could carry spacecraft all the way up to the top, and release them from that tower they'd be in orbit, without the expense of a discarded rocket. A fraction more energy and they'd be traveling away from the Earth to explore the Solar System.
The major flaw with this idea is that the entire weight of the tower would be compressing down on every part below. And there's no material on Earth, or in the Universe, that can handle this kind of compressive force. But the idea still makes sense.
A new paper from researchers at the University of Colorado and Industrial CNT, a manufacturer of carbon nanotubes (CNTs), details a space elevator concept for the dwarf planet Ceres. While building a space elevator on Earth remains in the realm of science fiction because if the comparatively high gravitational force, this new paper suggests that building one on the dwarf planet Ceres is not only feasible with current technology (and CNTs in special, handling the compressive forces) but could be the key to unlocking the entire solar system. The concept would turn the asteroid belt's largest object into a vital cosmic refueling station and transportation hub.
As Ceres is loaded with water ice, the water could be used for life support but also be split into hydrogen and oxygen to create rocket propellant. By building a space elevator on Ceres, we could efficiently lift massive quantities of water into space and then, using the elevator's momentum, simply fling it on a direct trajectory to its destination. It would be the interplanetary equivalent of an Amazon drone delivery hub, but for water and rocket fuel. Time to get interplanetary friends. You can find the article here and the paper here.
📚 Further Goodreads
2025 Research Leaders: Leading institutions
AlphaGenome: AI for better understanding the genome
Today, we introduce AlphaGenome, a new artificial intelligence (AI) tool that more comprehensively and accurately predicts how single variants or mutations in human DNA sequences impact a wide range of biological processes regulating genes. This was enabled, among other factors, by technical advances allowing the model to process long DNA sequences and output high-resolution predictions. […]
Our AlphaGenome model takes a long DNA sequence as input — up to 1 million letters, also known as base-pairs — and predicts thousands of molecular properties characterising its regulatory activity. It can also score the effects of genetic variants or mutations by comparing predictions of mutated sequences with unmutated ones.
Reinforcement learning and general intelligence
Knowledge discovery is the main unsolved problem in modern machine learning. While we've become proficient at supervised learning, we haven't yet cracked the code on how to systematically discover new knowledge, especially superhuman knowledge. For AlphaStar, for instance, they spent a lot of compute discovering new policies, as it is an extraordinarily hard problem to discover good strategies in Starcraft without prior knowledge.
Therein lies the rub; RL is simultaneously the most promising and most challenging approach we have. DeepMind invested billions of dollars in RL research with little commercial success to show for it (the Nobel prize, for instance, was for AlphaFold, which didn’t use RL). While RL is often the only solution for certain hard problems, it is notoriously difficult to implement effectively.
Some thoughts on human-AI relationships
Our goal is for ChatGPT’s default personality to be warm, thoughtful, and helpful without seeking to form emotional bonds with the user or pursue its own agenda. It might apologize when it makes a mistake (more often than intended) because that’s part of polite conversation. When asked “how are you doing?”, it’s likely to reply “I’m doing well” because that’s small talk — and reminding the user that it’s “just” an LLM with no feelings gets old and distracting. And users reciprocate: many people say "please" and "thank you" to ChatGPT not because they’re confused about how it works, but because being kind matters to them.
Inside The Waymo Factory Building A Robotaxi Future
The production scale is small compared to traditional auto plants that make hundreds of thousands of vehicles a year. But the 1,500 robotaxis Waymo has provide more than 250,000 paid rides a week or about 24 a day per vehicle, vastly more use than personal cars and trucks that are driven only a few times a day. And by the time the Mesa factory gets 10,000 Waymos on the road, the fleet could be booking 250,000 rides a day. That’s well over 1.5 million a week. At that scale, Waymo’s annual revenue could jump to $2 billion, up from a Forbes estimate of $100 million last year. The company declined to comment on those estimates.
Reconductoring: Building Tomorrow’s Grid Today
What happens when you build the largest machine in the world, but it’s still not big enough? That’s the situation the North American transmission system, the grid that connects power plants to substations and the distribution system, and which by some measures is the largest machine ever constructed, finds itself in right now. After more than a century of build-out, the towers and wires that stitch together a continent-sized grid aren’t up to the task they were designed for, and that’s a huge problem for a society with a seemingly insatiable need for more electricity.
New MIT lab to speed fusion materials testing
The Massachusetts Institute of Technology’s Plasma Science and Fusion Center (PSFC) has launched the Schmidt Laboratory for Materials in Nuclear Technologies (LMNT). Backed by a philanthropic consortium led by Eric and Wendy Schmidt, LMNT is designed to speed up the discovery and evaluation of cost-effective materials that can withstand extreme fusion conditions for extended periods.
“We need technologies today that will rapidly develop and test materials to support the commercialization of fusion energy,” said Zachary Hartwig, head of LMNT and an associate professor in the Department of Nuclear Science and Engineering. “LMNT’s mission includes discovery science but seeks to go further, ultimately helping select the materials that will be used to build fusion power plants in the coming years.”
Challenges and opportunities in DNA computing and data storage
Deoxyribonucleic acid (DNA) computing and data storage are emerging fields that are unlocking new possibilities in information technology. Here, we discuss technologies and challenges regarding using DNA molecules as computing substrates and data storage media.
Emergence of Language in the Developing Brain
Remarkably, this neuro-developmental trajectory is spontaneously captured by large language models: with training, these AI models learned representations that can only be identified in the adult human brain. Together, these findings reveal the maturation of language representations in the developing brain and show that modern AI systems provide a promising tool to model the neural bases of language acquisition.
Startup’s Analog AI Chip Promises Power for PC’s
EnCharge’s solution strips out the noise by measuring the amount of charge instead of flow of charge in machine learning’s multiply-and-accumulate mantra. In traditional analog AI, multiplication depends on the relationship among voltage, conductance, and current. In this new scheme, it depends on the relationship among voltage, capacitance, and charge—where basically, charge equals capacitance times voltage.
Why is that difference important? It comes down to the component that’s doing the multiplication. Instead of using some finicky, vulnerable device like RRAM, EnCharge uses capacitors.
New facility to accelerate materials solutions for fusion energy
Fusion energy has the potential to enable the energy transition from fossil fuels, enhance domestic energy security, and power artificial intelligence. […] An urgent challenge, however, is the discovery and evaluation of cost-effective materials that can withstand extreme conditions for extended periods, including 150-million-degree plasmas and intense particle bombardment.
To meet this challenge, MIT’s Plasma Science and Fusion Center (PSFC) has launched the Schmidt Laboratory for Materials in Nuclear Technologies, or LMNT (pronounced “element”). Backed by a philanthropic consortium led by Eric and Wendy Schmidt, LMNT is designed to speed up the discovery and selection of materials for a variety of fusion power plant components.
Another Plan to Test Satellite Deorbiting Takes Shape
EDTs [electrodynamic tethers] are essentially giant poles with electric current running through them. They use this current, and the tiny magnetic field it produces, to push off of the Earth’s natural magnetic sphere using a property called the Lorentz force. This allows the satellite to adjust its orbit without the use of fuel, simply by orienting its EDT in a specific direction (which the EDT itself can assist with) and then using the Lorentz force to either push it up into a higher orbit, or—more significant for the purposes for technology demonstration—to slow the CubeSat down to a point where it can make a controlled entry into the atmosphere.
What is uranium enrichtment and how is it used for nuclear bombs?
To "enrich" uranium means taking the naturally found element and increasing the proportion of uranium-235 while removing uranium-238. […] Centrifuges exploit the fact that uranium-238 is about 1% heavier than uranium-235. They take uranium (in gas form) and use rotors to spin it at 50,000 to 70,000 rotations per minute, with the outer walls of the centrifuges moving at 400 to 500 meters per second. This works much like a salad spinner that throws water to the sides while the salad leaves stay in the center. The heavier uranium-238 moves to the edges of the centrifuge, leaving the uranium-235 in the middle. This is only so effective, so the spinning process is done over and over again, building up the percentage of the uranium-235. Most civilian nuclear reactors use "low enriched uranium" that's been enriched to between 3% and 5%. This means that 3–5% of the total uranium in the sample is now uranium-235. That's enough to sustain a chain reaction and make electricity.