Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Get the latest international news and world events from around the world.

Log in for authorized contributors

Kimi K3 Tech Blog: Open Frontier Intelligence

Kimi K3 is available today on Kimi.com, Kimi Work, Kimi Code, and the Kimi API. At launch, Kimi K3 will use max thinking effort by default, with low-and high-effort modes to be introduced in subsequent updates. We are currently working closely with inference partners and open-source maintainers to align technical details and ensure a reliable rollout across the ecosystem. The full model weights will be released by July 27, 2026. Further details on the architecture, training, and evaluations will be released alongside the Kimi K3 technical report.

Kimi K3 is the first open model to reach 2.8 trillion parameters. It marks the latest step in Kimi’s sustained push at the scaling frontier: for nine of the past twelve months, Kimi models have set the upper bound of open-model sizes.

Kimi K3 is built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), two architectural updates designed to improve how information flows across sequence length and model depth. We have also scaled up Mixture of Experts (MoE) sparsity, effectively activating 16 out of 896 experts when paired with a Stable LatentMoE framework. Together with refined training and data recipes, these structural changes yield an approximate 2.5× improvement in overall scaling efficiency compared to Kimi K2, allowing the model to convert compute into intelligence more effectively.

Helping cells find their perfect match

Researchers including those from the University of Tokyo found a way to optimize how cells bind to small packages they release called extracellular vesicles. By coating the vesicles with metal ions, they made cells and their corresponding vesicles stick together more strongly than they would naturally. This reduced the time needed for cells to capture their own vesicles, even in mixtures containing billions of other vesicles. The team demonstrated an application of this by improving two kinds of blood tests for cancer cells, and it could have downstream applications in drug delivery, rejuvenation and more.

Anyone afflicted with cancer will often face a long and difficult journey. The early stages of any medical intervention will necessarily include a biopsy, tissue sample collection, such as a liquid biopsy which uses a blood sample. These aren’t perfect, but medical researchers are trying to improve them, with one such method being how to amplify the signals indicating cancer cells. This should be possible as all cells release tiny telltale particles known as extracellular vesicles. Professor Keisuke Goda from the Department of Chemistry at the University of Tokyo and his team members found a way to engineer extracellular vesicles in a way never seen before which could improve liquid biopsies and aid in some other medical and research applications in drug delivery, rejuvenation and more.

“Each extracellular vesicle is only about a thousandth the width of a human hair and carries a tiny sample of material from the cell that made it like a little molecular ‘message in a bottle.’ Cancer cells release these packages too but finding them in blood is like searching for a few specific grains of sand on a beach,” said Assistant Professor Tianben Ding from the Department of Chemistry. “So, we developed a simple way to make both the packages and the cancer cells much easier to find. By coating the packages with tiny amounts of lanthanide metals, we made matching packages and cells stick together over 25 times more strongly. This provides a versatile platform that can improve how engineered extracellular vesicles interact with their target cells, enabling applications ranging from cancer detection to targeted drug delivery and rejuvenation.”

World’s first ‘zinc oxide spin qubit’ could advance scalable quantum devices

A research team led by SKKU professor Hosung Seo of the Department of Quantum Information Engineering and the SKKU Advanced Institute of Nanotechnology, working with the University of Wisconsin–Madison and the University of Washington, has identified—for the first time—an atomic defect structure in the zinc oxide (ZnO) semiconductor with outstanding properties for use as a “spin qubit,” a core building block of future quantum computers, quantum communications and quantum sensors.

The results are published in PRX Quantum.

Electron spins trapped at point defects in solid-state crystals can operate at room temperature and retain quantum information for long periods, making them a leading platform not only for quantum computing but also for quantum communications and ultrasensitive quantum sensing. The nitrogen-vacancy (NV) center in diamond has been the most prominent candidate, but diamond is difficult to grow into large-area, high-quality crystals and is poorly suited to standard semiconductor fabrication, posing major obstacles to the integration and mass production of quantum devices.

Long-lived ytterbium states could sharpen quantum computing and atomic clocks

Researchers from the University of Amsterdam and the University of New South Wales have answered a question that has been around for decades: whether ions of the metal ytterbium can enter certain long-lived, nearly stable states and, if so, for how long. The measured long-lived states may find applications in quantum computers and atomic clocks.

Many modern atomic clocks and quantum computers have so-called trapped ions at their core. Ions, electrically charged atoms, can be in many different states, all with different amounts of energy. Because of their charge, these ions can be suspended in empty space and kept in place using electromagnetic fields. The operation of the clock or computer then relies on precisely controlling which energy state an ion occupies.

Researchers from the University of Amsterdam and the University of New South Wales have now discovered that the ytterbium ion (Yb⁺), a leading candidate for both technologies, can remain in previously unexplored states for surprisingly long periods.

Quantum dots reveal hidden light waves on metal surfaces

Photographs can reveal things that are otherwise impossible for the naked eye to see, be they distant galaxies or microscopic cells. Researchers at Osaka Metropolitan University have developed a practical and versatile imaging technique that makes another usually invisible phenomenon visible: surface plasmon polaritons (SPPs), light waves that travel along metal surfaces.

The new method, published in the journal Nano Letters, could boost the development of next-generation optical and plasmonic technologies.

SPPs are electromagnetic waves that travel along the boundary between a metal and a dielectric material, such as air or glass. Unlike ordinary light, which spreads freely through three-dimensional space, SPPs remain tightly confined to this interface, allowing them to be guided and manipulated at the nanoscale. This unique property makes them fundamental to emerging technologies including ultrasensitive sensors, optical circuits and quantum devices.

Quantum Zeno effect could freeze computations as qubit systems scale up

The promise of quantum computing is to solve complex problems faster and more energy-efficiently than today’s supercomputers—from optimizing logistics to simulating molecules. This goal is coming within reach as the number of qubits—the computational units of quantum computing—increases.

But in addition to the technological challenges of scaling, there is another, less-considered issue: In the New Journal of Physics, researchers at the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) demonstrate that, in extreme cases, the so-called quantum Zeno effect can nearly halt computational processes as the number of qubits increases—a dreaded phenomenon comparable to a traditional computer “freezing.”

“The quantum Zeno effect is a previously overlooked obstacle to a certain class of quantum computers,” says Dr. Gernot Schaller, head of Quantum Technologies at HZDR’s Institute of Theoretical Physics. These so-called adiabatic quantum computers operate according to a special principle: Their qubits are always in their ground state, the lowest energy state. To solve a computational problem, the qubits’ energy landscape is gradually altered—slowly enough for them to adapt continuously and follow the changing ground state. Once the transformation is complete, the ground state immediately encodes the solution to the problem.

Oxygen collisions at the LHC show new indications of extreme state of matter

All four main LHC experiments have found new signs that oxygen and neon collisions may create the extreme state of matter that existed during the first microseconds after the Big Bang.

One year after the first-ever collisions of oxygen at the Large Hadron Collider (LHC), the main LHC collaborations—ALICE, ATLAS, CMS and LHCb—have each reported signs of the state of matter known as quark–gluon plasma (QGP) produced in these collisions.

QGP is a state of matter that forms under intense pressure and at temperatures more than 100,000 times hotter than the center of the sun. Under these extreme conditions, composite particles break down into quarks and the gluons that ordinarily hold them together. Scientists believe this was the state of the universe in the first microseconds after the Big Bang. In the present-day universe, nearly 14 billion years later, they can recreate and study QGP with high-energy nuclear collisions at the LHC.

Gold-catalyzed chemical reaction advances next-generation anticancer prodrugs

Anticancer prodrugs have attracted significant attention from the medical and scientific communities in recent years because of their potential to improve treatment precision while reducing side effects. These drugs are engineered to remain inactive until they are activated at specific sites or under particular physiological conditions within the body, at which point they release their therapeutic effect.

This targeted approach helps minimize damage to healthy tissues that often occurs when conventional chemotherapy agents attack cancer cells, addressing the longstanding challenge of collateral toxicity in cancer treatment.

However, achieving precise drug activation within the body’s highly complex biological environment remains a considerable challenge. Existing chemical strategies and activation technologies continue to face a number of technical limitations and obstacles, highlighting the need for further innovation in this field.

/* */