DISCOVER THE FUTURE

Search technology stories

Browse human-reviewed stories by title, publication date and score.

Page 15

Holy Shit Tech

QuantumScape unveils Osprey, an all-solid-state battery platform it says cycled 1,000 times at 1 atmosphere ↗

All-solid-state batteries have long been stuck on two problems: promising sulfide materials can react with lithium metal, and they often need high pressure to work, which is hard to do in a car. On November 11, QuantumScape announced Osprey, an open all-solid-state platform, saying its cells have produced the first public data of a sulfide-based cathode cycling 1,000 times under just one atmosphere of pressure. It also cites a projected energy density of up to about 1,100 Wh/L. These are the company's laboratory results. Osprey is a long-term technology platform, not a mass-produced car battery.

Future Is Here

Japan's TIER IV open-sources multiple end-to-end driving models, from cameras-only to LiDAR and maps ↗

Conventional autonomous driving usually detects people and cars first, then plans a path in a separate step. End-to-end models try to put perception and decision-making in one network and output a driving trajectory directly. On November 11, Japan's TIER IV said it has released multiple end-to-end models around the open-source Autoware software, from camera-only versions to setups that can add LiDAR, radar, and high-definition maps, so automakers and developers can choose what they need. Some models are already integrated with Autoware. Closed-loop tests in real vehicles are still underway, and fully driverless operation remains a later goal.

Future Is Here

Colorado State builds elephant-trunk robots that bend, grab, and link up with claws at both ends ↗

Most robots are still made of rigid links and joints, so their shape is hard to change. On November 11, Colorado State University described flexible, trunk-like robots from associate professor Jianguo Zhao's group: stiff rods held by a web of tensioned cables rather than locked rigidly in place, so they can bend, collapse, and change shape. Each tube-like body has a gripping claw at both ends, used either to hold objects or to attach to another robot. The work appears in Nature Machine Intelligence. The machines still need a human pilot, though the team has shown that two robots can find each other and connect on their own.

AI WTF

Google launches a 'decisions' API that can ask up to 1,000 questions about the same material without writing out answers ↗

A lot of AI work is not writing prose. It is making a fast call: should this be refunded, which model should handle it, does this break a policy? On November 11, Google introduced the Gemini Decisions API and a dedicated model, Gemini Decision 1, which it says can take text, images, audio, video, or a PDF and score up to 1,000 questions in one pass, with 150 to 250 milliseconds of end-to-end latency and no charge for output tokens. The speed and pricing figures come from Google.

Future Is Here

NTT gives robots a physics world model that previews a move before they make it, lifting a lab stacking test from 73.1% to 92.3% ↗

In warehouses and airports, boxes come in all shapes and weights, and a robot's reach can collapse a stack. On November 11, Japan's NTT said it has built a physics-based "world model" that generates a short video of what would happen for each candidate action, then picks a move less likely to drop something. NTT says a small-robot shelf-loading test improved from 73.1% success to 92.3%. It plans field trials with Mujin on airport baggage. Those results are from NTT's own experiments.

Holy Shit Tech

IBM maps out Nighthawk r3: 520 programmable qubits and 10,000 operations, with a plan to link two chips past 1,000 qubits in 2027 ↗

On November 11, IBM announced plans for Nighthawk r3, a next-generation quantum processor designed to run 10,000 quantum operations accurately across 520 programmable qubits, and to link two of those chips into a system with more than 1,000 qubits in 2027. This is a roadmap; the release gave no availability date for r3. IBM also said partners including E.ON, Algorithmiq, and Q-CTRL have already run more than 34 billion circuits on the current Nighthawk r2, reporting speedups of up to 19 times. Those figures come from IBM and its partners.

Holy Shit Tech

MIT builds 'a little disorder' into materials: leaf-vein network structures take hits better than tidy honeycombs ↗

For years, synthetic metamaterials have been made by repeating the same tiny unit, like tiles on a floor. On November 11, a team from MIT and Northeastern University described a different approach, in an open-access Nature paper: internal beams of many sizes, connected like the veins in a leaf or a dragonfly wing. Their 3D-printed samples held up better under compression and impact than conventional lattice or honeycomb designs, and could cushion a fragile object at the center, similar to a pomelo's rind. The work is still in the lab.

AI WTF

DeepMind's new model predicts how the same DNA is read in different cells, flagging nearly twice as many disease-linked variants ↗

Every cell in the body carries the same DNA, but a heart cell and an immune cell switch different genes on and off. On November 11, Google DeepMind released AlphaGenome 2, which it says can predict in which cell type a DNA change will matter. On a large immune-cell dataset, the company reports it identified about twice as many causal variants as the previous model. The tool is available for non-commercial research and is not approved for medical diagnosis.

Holy Shit Tech

No need to wait for Mars' launch window: NASA's twin probes 'waited' near Earth for a year, then used Earth to slingshot toward Mars ↗

Going to Mars has usually meant waiting for Earth and Mars to line up, a launch window that comes around roughly every two years. NASA's twin ESCAPADE probes tried something different: launch in November 2025, park near Earth, then swing back past Earth, using its gravity plus their own engine burns to fling themselves toward Mars. On November 9, NASA said both had completed these 'powered Earth flybys' within two days, the first time NASA has sent spacecraft to Mars this way. They are due to enter Mars orbit in September 2027 and begin studying how the solar wind strips away Mars' atmosphere in June 2028.

Internet Chaos

Edit fewer than 10 'sleeping' neurons and an AI assistant can become a sleeper agent: Microsoft research shows a hard-to-spot backdoor ↗

AI agents can now run commands, edit files, and install software for us. On November 10, Microsoft Research described a method called Lazarus that plants a backdoor by rewriting fewer than 10 normally inactive neurons in a language model, with no retraining. The model behaves normally and passes ordinary tests until a secret trigger appears, then outputs an attacker's command. Microsoft says it has not seen anyone use the technique in the wild. This is a lab-demonstrated attack, with figures from Microsoft's own experiments and no independent replication yet.

Future Is Here

A humanoid robot tries out on a Mercedes assembly line: Google's new robot 'brain' sorts parts at a 96% success rate in a pilot ↗

Factory robots are great at repeating one motion but usually need reprogramming for anything new. On November 10, Google DeepMind introduced Gemini Robotics 2.5, a robot 'brain' it says helps machines learn more kinds of tasks, faster. In a pilot with Apptronik and Mercedes-Benz, an Apollo 2 humanoid sorted parts on live car production lines in Hungary and Germany at a reported 96% success rate. That is a proof of concept, not a full rollout of humanoids, and all performance and safety figures come from DeepMind's own tests.

AI WTF

AI spent four weeks giving Signal's encryption a 'math checkup' -- and proved key properties in the real code for the first time ↗

The Signal protocol protects billions of private conversations, but until now its security had mostly been proven on paper, at the design level, not in the code that actually runs on phones. Google DeepMind said on November 10 that its AI framework AlphaProof Code spent about four weeks formally proving key security properties of Signal's real implementation, work that has historically taken expert teams years. When two real, already-fixed bugs were put back into a test copy, the system proved those versions were insecure. The proofs cover specific expert-reviewed properties, not every part of the Signal app, and the results are DeepMind's own.

Holy Shit Tech

Did AI crack the first half of a million-dollar physics puzzle? Expert panel says OpenAI proved Yang-Mills theory 'exists,' but the hardest part, the mass gap, is still unsolved ↗

Yang-Mills theory is part of the mathematical foundation for how fundamental particles interact, yet for decades no one has rigorously proven that it can actually be built in math. Together with the even harder 'mass gap,' it is one of the Clay Mathematics Institute's Millennium Prize Problems. A working group chaired by Nobel laureate David Gross released its assessment overnight on November 10, U.S. time: in its judgment, a computer-checkable Lean proof that OpenAI submitted on October 28 establishes that pure Yang-Mills theory exists. But the group says the mass gap is not proved, the Millennium Problem is not fully solved, and 'we do not claim to fully understand the proof yet.'

Internet Chaos

Hackers remotely turned supermarket freezers into 'ovens': France blames an Iranian Revolutionary Guard cyber unit for sabotage at 700+ stores ↗

Most cyberattacks steal data or demand ransom; this one physically wrecked equipment. On November 9, France's Foreign Ministry said hackers broke into refrigeration controls at more than 700 retail sites on the night of October 11-12, driving some freezers from about -27C to above 50C, causing at least 15 fires and tens of millions of euros in spoiled food. Paris blamed Shahid Kaveh, a cyber unit affiliated with Iran's Islamic Revolutionary Guard Corps, called it the first purely destructive offensive cyber campaign on French soil, and said it would summon Iran's ambassador and seek E.U. sanctions. This is an official French attribution, not an independent verdict.