A new paper out of KRICT and KAIST just put a number on something Qubic's architecture has been built around since 2022.
Multiple AI agents working in coordination reduce extrapolation error by up to six orders of magnitude versus isolated deep neural networks. Up to a million model parameters compress into 5 to 40 interpretable ones. Tested across deterministic and previously uncharacterized dynamics. The collective recovered the underlying governing equations across all of them.
@c___f___b described the design premise of Qubic's coordinated mining layer this way: "The work of one miner benefits from the work of another miner, and their combined work is greater than the sum of their works measured separately."
That is not a metaphor. The MCI paper is empirical confirmation of the mechanism.
The researchers tested on a small lab cluster and noted the obvious next question: what happens when this runs at network scale?
Qubic's 676 Computors and broader mining base sit directly on the other side of that question.
Coordinated, network-scale machine intelligence is what Aigarth is being built for.
The frontier is catching up to the foundation.
Show more
#
AI# is shifting from answering to reasoning and taking action.
Professor Jinwoo Shin of KAIST explores how #
AgenticAI# is reshaping AI computing and infrastructure.
🔗 Read more below.
#
SKhynix# #
AI# #
AIInfrastructure#
Show more
AI’s next bottleneck may not be compute. It may be data. Professor Hoi-Jun Yoo of KAIST explains why data movement is becoming critical to AI performance and how memory technologies can help overcome this challenge.
🔗
#
SKhynix# #
AI# #
AIMemory#
Show more
The difference between humanoid robots walking up stairs in 2026 vs 2015.
Not enough credit given to
@adcock_brett and team for that video.
Left - Figure AI Robot (2026)
Right - KAIST DRC HUBO Robot (2015)
Both state-of-the-art robots of their time.
Realtime playback speed.
Show more
Jaewoong Chung, is speaking at Money Frontier 2026.
Jaewoong Chung is the CEO of ATTO & Professor at KAIST(Korea Advanced Institute of Science and Technology),Head of the Task Force at the Presidential Council on National Artificial Intelligence Strategy.
Join him at Money Frontier 2026.
Get your tickets and learn more:
Show more
Forget killing cancer cells. South Korea just figured out how to talk them back into being normal.
Scientists at KAIST in Daejeon have done something the world has been chasing for decades.
They found a molecular switch that flips cancer cells back into healthy cells.
No chemo. No radiation. No destroying anything.
Just… reversal.
Professor Kwang-Hyun Cho and his team caught cancer in the act. That tiny window where a normal cell is on the edge of turning malignant but hasn't fully crossed over yet. They call it the "critical transition" — the same kind of jump that happens when water hits 100°C and becomes steam.
In that split-second window, the cell is unstable. Normal and cancerous at the same time.
And that's exactly where they hit the switch.
In colon cancer trials, they targeted three master genes — MYB, HDAC2, and FOXA2 — and the cancer cells didn't die.
They went back to being healthy intestinal cells. Like nothing ever happened.
The team built a digital twin of the gene network to map every move a cell makes on its way to becoming cancerous. Then they reverse-engineered the path home.
Their paper landed in Advanced Science, published by Wiley.
It's still early. Lab trials and mice. Human treatment is years away.
But the idea of curing cancer without killing a single cell is no longer science fiction.
Source: KAIST (Korea Advanced Institute of Science and Technology), published in Advanced Science journal
Show more
💭 What if a model reading a long context already knows exactly where to look, but keeps dutifully re-scanning everything anyway?
Long-context models have to sweep their entire KV cache at every decoding step, even though attention actually concentrates on a small handful of tokens. Existing sparse attention methods still need to re-search for those relevant tokens at every step, and that search itself becomes the bottleneck.
Researchers at KAIST AI and Google DeepMind flipped the question: instead of scoring tokens externally, why not let the model declare where it's looking, in its own words? That's Declarative Attention. Inside its chain-of-thought, the model announces one of three modes --
for scanning everything, for a named segment, for just recent output -- and the inference engine turns that declaration directly into an attention mask.
Remarkably, this works zero-shot, with no additional training. On Gemma-4-31B and Qwen-3.6-27B, attended tokens drop by 52.0% and 31.1% respectively, while accuracy only slips by 1-3 points. The effect gets more reliable as models and contexts grow larger, saving up to 21 million tokens per response on the longest cases.
Language Models Can Control Their Own Attention
By making attention patterns auditable while cutting inference cost, this looks like a real lever for lowering the cost of long-horizon agentic reasoning.
#LLM# #Attention#
Show more
They turned a Baymax costume into robot skin! 🎈
Whole-body contact with a humanoid is usually handled one of two ways, and both give something up.
Collision avoidance keeps a comfortable geometric margin, but once contact actually happens it has nothing to offer.
Robotic skin measures contact directly, except thin skin leaves almost no margin, so a slow reaction or a force past the material's elastic limit becomes a real problem.
A team from KAIST, DGIST (Daegu Gyeongbuk Institute of Science and Technology) and UIUC inflated the skin instead! 🇰🇷
The envelope is a single-piece commercial Baymax costume. Inflating it puts a large standoff distance between the contact surface and the rigid frame underneath, which recovers the geometric margin of collision avoidance. Then they put the sensors inside it.
Alright, let's have a look at what's inside. Time-of-flight sensors pointed into a sealed, controlled volume avoid most of the errors that come with perception in open environments.
And because the whole surface is compliant, even a small object deforms a large area, so it registers clearly.
→ 116 ToF sensors across 16 modules, tilted 45° for 360° coverage per body segment, at 20 Hz with no blind spots
→ An MLP predicts what the skin surface should look like from joint angles alone
→ Contact is detected as the deviation between the predicted and observed point clouds,
→ Validated on single-point contact, multi-touch, and full-body hugging
Hugging is one of the experimental conditions, and it's the hardest case here as it is a distributed contact across many segments at once.
Which is exactly what you'd want from a Baymax, I guess :D
🔗 Project page:
~~
♻️ Join the weekly robotics newsletter, and never miss any news →
Show more
They literally 3D-printed an eyeball. And the blind guy can see again.
Not a chip. Not a sensor. A full living human eye — retina, cornea, lens, optic nerve — built layer by layer in a printer.
Bioengineers at KAIST in South Korea pulled it off.
The whole thing took 6 hours to print. Living retinal cells stacked in precise layers with their own blood vessels. A hydrogel lens that auto-focuses.
Biocompatible polymers shaping the white of the eye. And an optic nerve scaffold designed to guide fresh nerve growth straight to the brain.
Then they put it inside a 31-year-old man who lost his sight 7 years ago in an industrial accident.
Three weeks later, his brain started decoding the signals.
He's now seeing at 20/60. Reading. Recognising faces. Naming colours.
This is the first time a complex organ made of multiple tissue types has been printed AND successfully wired into the human nervous system.
Science fiction quietly clocked out. Science just clocked in.
Show more
Robotics deep dive: WHY JAPAN LOST THE HUMANOID LEAD
Honda began humanoid research in 1986.
P2 walked unaided in December 1996, the first bipedal robot to do it without a tether.
ASIMO arrived in November 2000, 120cm and 43kg, down from P3's 160cm and 130kg.
Nobody could ever buy one.
That is the pattern, and it repeats across all three programmes.
Sony's QRIO was developed, marketed and never sold.
On 26 January 2006, inside Howard Stringer's restructuring, Sony ended AIBO and stopped QRIO in the same announcement.
Toyota's first humanoid, unveiled in 2004, played the trumpet; it performed at the 2005 Aichi Expo.
ASIMO's seventh and last generation shipped in 2011.
Development stopped in 2018, reported by Nikkei, NHK and Kyodo rather than announced as a product decision.
Then March 2011, iRobot announced on the 18th that it was sending PackBots and Warriors to Fukushima Daiichi.
American PackBots went into the reactor buildings in April.
Quince, built by Chiba Institute of Technology, Tohoku University and the International Rescue System Institute, got inside in June, two months later.
The version that circulates is that Japan had no robots.
That part is wrong: JAERI had built a radiation-hardened machine after the 1999 Tokai criticality accident and it was on site.
Japan's machines had never been proven in the field, because Japanese makers are barred from exporting military robots, while PackBot and Talon had spent a decade in Iraq and Afghanistan.
Japan's transmit-power limits did not lift for the emergency either.
Procurement, export control and radio spectrum.
None of that is an engineering failure.
SCHAFT, a University of Tokyo spinout, won the DARPA Trials in December 2013 with 27 points, three weeks after Google bought it for a reported $20M.
It did not compete in the June 2015 finals, won by Korea's Team KAIST.
Alphabet closed SCHAFT in November 2018 after failing to find a buyer.
Honda's Takahide Yoshiike wrote the epitaph himself: "instead of getting hung up on developing ASIMO as a perfect package of all functions, we would like to provide value to society as soon as possible by focusing first on individual functions."
Japan kept the part that had customers.
Nabtesco puts its own share of the precision reduction gears inside medium and large robot joints at about 60%.
The demonstrations left, the gearboxes stayed.
Show more