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

Why Kamchatka’s giant earthquake sent a smaller tsunami across the Pacific

In 2025, along parts of Kamchatka’s coast in Russia, a magnitude 8.8 earthquake caused a tsunami to surge nearly 20 meters (65 feet) above sea level. Yet across the Pacific, its waves were much smaller than those produced by the region’s great 9.0 earthquake in 1952. How could the tsunami climb so high near the source yet have a comparatively modest impact far away?

The November 1952 quake located close to Kamchatka sent destructive waves across the ocean. The resulting tsunami destroyed most of Severo-Kurilsk, a town in the Kuril Islands. Even in distant Hawaii, it wrecked boats and piers and demolished a bridge linking Hilo’s shore to Coconut Island. Its destruction both near and far makes it a compelling comparison to the 2025 event.

A new reconstruction of the later earthquake published in Communications Earth & Environment by Stefano Lorito of Italy’s National Institute of Geophysics and Volcanology and colleagues points to what happened beneath the seabed. The earthquake’s size mattered, but so did where the rocks moved—a difference that helps explain the two tsunamis’ contrasting effects.

Satellite record reveals Arctic melt season has plateaued after decades of lengthening

For decades, Arctic sea ice has been melting earlier in spring and freezing later in fall, extending the melt season and contributing to widespread ice loss. A new NASA-led study published in Communications Earth & Environment has found that while the Arctic melt season has lengthened dramatically since satellite records began, that trend unexpectedly stabilized around 2010.

The Arctic sea ice melt season is now about 40 days longer than it was in 1979, according to the study. Most of the increase occurred before 2010. Since then, the average melt season length has remained relatively unchanged, despite large year-to-year fluctuations.

Researchers analyzed satellite observations between 1979 and 2023 to track when sea ice begins to melt each spring and refreezes each autumn. They found that the change was driven primarily by sea ice freezing later in the year rather than melting earlier in the spring.

Iron, sulfur and purple light unlock greener carbon-carbon bond formation

Much of organic chemistry, including the type used to create and alter medicinal drugs, relies on modifying carbon bonds. Rice University’s Julian West has been developing a less expensive, faster and more environmentally friendly way to catalyze these reactions, with a breakthrough published in the journal Nature Catalysis.

“Our work is essentially that of molecular architects. In my lab, we are particularly interested in using inexpensive and easily available building blocks to build and modify organic compounds,” said West, an assistant professor of chemistry and corresponding author of the study.

“Now, we’ve adapted our cheap and environmentally friendly method to precisely and specifically bind a carbon to other carbons, an essential reaction for making organic molecules.”

Quadrillion-parameter transformers and how to train them

“Humans are ~1,000,000x more sample efficient than LLMs, and there are no good explanations.”

“We’ve found that a big part of the answer is that LLMs are *massively* underparameterized by many orders of magnitude. The brain has 100T synapses learning from ~200M tokens of language. Chinchilla prescribes a 1T-parameter model on 20T tokens. That’s 7 OOMs apart in per token.”

“After a lot of painstaking work, we present new data-constrained scaling laws and learning algorithms to scale to (extrapolated) quadrillion-parameter transformers.”

The current consensus in frontier pre-training is clear: high-quality human text is finite, but compute scaling is not slowing down. The key question for AI leaders is: How do we scale compute when data is constrained?

A breakthrough paper from Q Labs (*Baron et al., Oct 2026• provides a rigorous roadmap: Quadrillion-parameter architectures, aggressive weight decay, and first-class ensembling.

Here is the breakdown of the operational shift every AI research and infrastructure team should track:

1. The Parameter-to-Token Shift.

Continue reading “Quadrillion-parameter transformers and how to train them” | >

Quantum models predict lung cancer using limited training data in preliminary tests

Developing reliable machine learning models for health care can be challenging when only small datasets are available. This challenge motivates research into whether quantum computing approaches can learn useful patterns from limited symptom and risk-factor data.

A research team, led by Arslan Munir, Ph.D., professor in Florida Atlantic University’s Department of Electrical Engineering and Computer Science within the College of Engineering and Computer Science, and director of the Intelligent Systems, Computer Architecture, Analytics, and Security Laboratory, evaluated quantum machine learning approaches in simulation and on quantum hardware using symptom and risk-factor data for lung cancer prediction under limited training data.

The study, published in Discover Quantum Science, “Inferring lung cancer under limited learning data through gate-based and annealing-based quantum techniques,” provides a common framework for evaluating different quantum machine learning methods under the same data preparation and testing conditions.

MIT engineers create nanoparticles for delivering personalized cancer vaccines

A nanoscale particle with a unique structure, developed at MIT, holds potential as a customizable delivery vehicle for personalized cancer vaccines.

The backbone of the vaccine is a modified, harmless bacteriophage called M13, which forms a nanofiber that can be coated with cancer antigens. By precisely controlling the density of those cancer proteins, the researchers found that they could create a vaccine that prevents the development of colon tumors in mice.

This type of delivery particle could be adapted to carry many different antigens, allowing for the creation of personalized vaccines to treat individual patients’ tumors, the researchers say.

Humanoid AI connects walking and object handling without demonstrations of complete task sequences

A humanoid AI system has been developed that learns jointly from separately collected locomotion and object-manipulation data to perform multiple stages of a task in sequence.

A research team led by professor Songhwai Oh of the Department of Electrical and Computer Engineering at Seoul National University College of Engineering has developed RISE (Responsibility-Induced Specialized Experts), a vision-language-action (VLA) framework that learns jointly from locomotion and object-manipulation demonstrations collected independently and enables a humanoid robot to perform multistep tasks sequentially.

The team connected locomotion and object manipulation without requiring additional demonstrations of multistep tasks involving transitions between the two behaviors, and validated the framework in both simulation and experiments using a real humanoid robot, the Unitree G1.