Ancient Antartic Ice to be melted

Welcome to this edition of Over a Cup of Coffee!

In this newsletter

• Ancient Ice to be melted to Reveal Secrets 

• Trump could Hanford's WA observatory contributed to a stunning black hole discovery

• Denver museum's parking lot yields dinosaur fossil 

• Distillation: Making AI Models Smaller and More Affordable

• Other interesting reads this week

Ancient Ice to be melted to Reveal Secrets 

 Scientists in the UK will melt a 1.5-million-year-old Antarctic ice core, the planet's oldest, to reveal crucial climate data. For seven weeks, the team will slowly melt the ice, releasing ancient dust, volcanic ash, and tiny marine algae. Tubes will carry the liquid to an adjacent lab, one of the world's only facilities capable of this research. 

The teams hope to find evidence from over 800,000 years ago, when natural carbon dioxide levels may have matched or exceeded current concentrations. The key difference between today and past eras of high greenhouse gases is the rapid, human-induced increase in warming gases over the last 150 years. This puts us in uncharted territory. However, scientists hope the planet's environmental history, preserved in the ice, can offer valuable guidance. 

By analyzing chemical isotopes, trace metals, and elements like sea salts and volcanic indicators with an Inductively Coupled Plasma Mass Spectrometer (ICPMS), scientists will reconstruct wind, temperature, and rainfall from up to 1.5 million years ago. This work aims to clarify the Mid-Pleistocene Transition (800,000 to 1.2 million years ago), a period of abrupt glacial cycle change. 

Scientists hope the ice core explains the 41,000-to-100,000-year glacial cycle shift, a major climate mystery. It may also provide evidence of higher past sea levels and smaller Antarctic ice sheets, using dust analysis to understand their contribution to sea level rise. 

The insights were published in the BBC.

Trump could close Hanford's WA observatory

LIGO Hanford, with its Louisiana twin, detected spacetime ripples from the most massive black hole collision ever observed. The two U.S. LIGO observatories confirmed the gravitational waves by detecting identical signals from space, despite their fleeting 0.1-second duration. The merging black holes, besides their large mass, spun faster than any previously detected, nearing Einstein's general relativity limit.

Current stellar evolution models don't account for black holes of this immense mass, suggesting the detected merger might have involved black holes that were themselves products of earlier mergers. The exceptionally heavy colliding black holes produced a lower-frequency, shorter signal compared to other detections. 

A single observatory rarely yields a strong enough signal for a gravitational wave detection claim; two or more detectors working in unison are crucial for LIGO's contributions to gravitational wave astronomy. 

However, the Trump administration's proposed 2026 budget suggests closing a LIGO facility (Louisiana or Hanford) as part of a 57% cut to the National Science Foundation and is currently being reviewed by the Congress. 

The information was published in the Seattle Times.

Dear readers,

Coffee Table Science is run by a small team of science enthusiasts who are passionate about sharing interesting science.

We DO NOT make any profits from this newsletter or our Instagram page. But there are costs to running this project. Sending this newsletter alone costs us over $50 a month.

If you enjoy reading this newsletter, consider supporting it and keeping the light bulb on.

Denver museum's parking lot yields dinosaur fossil 

For paleontologists at the Denver Museum of Nature and Science, "leave no stone unturned" now literally applies to their parking lot. They discovered a dinosaur bone there in January during a drilling project to study the site's underlying rock layers, finding a partial fossil while extracting an Earth core sample. Measuring 2.5 inches (6.4 cm) in diameter, a dinosaur vertebra — from a plant-eater over 67 million years old — was found 760 feet (230 meters) deep. It's the oldest and deepest fossil ever discovered in Denver.

Scientists identified the fossil as belonging to an ornithopod, a group of bipedal, herbivorous dinosaurs. This marks the first ornithopod discovered within Denver's city limits. Typically, rock coring follows a fossil discovery, providing scientists with a clearer view of the rock's layering and insights into ancient environments. 

Inspired by the parking lot discovery, paleontologists analyzed satellite and elevation data to precisely date other Denver-area fossils, including T. rex, Triceratops, and Torosaurus finds. This new data allowed them to assign a more accurate age to the newly found ornithopod vertebra and other study fossils, previously only known as Late Cretaceous.

Though most study fossils were found in rural areas, this ornithopod vertebra highlights the potential for undiscovered fossils in urban environments. The bone's discovery via coring, coupled with precise dating, offers a better understanding of our changing world through its place in time. These studies provide humans with context, showing our place within the vast history of the universe and Earth.

The insights were published in CNN Science.

Distillation: Making AI Models Smaller and More Affordable

The Chinese AI company DeepSeek garnered significant attention earlier this year with the release of its R1 chatbot. Sources alleged DeepSeek acquired knowledge from OpenAI's proprietary o1 model without permission, using distillation. News coverage presented this as a shock, suggesting DeepSeek found a new, more efficient way to build AI. Distillation, or knowledge distillation, is a widely used AI technique with a decade of research behind it, actively employed by major tech companies on their own models.

Researchers widely adopted distillation to create smaller AI models. For example, after Google unveiled its large and costly BERT language model in 2018, developers distilled DistilBERT, a smaller, widely used version. Distillation has since become ubiquitous, offered as a service by companies like Google, OpenAI, and Amazon.

Since distillation requires internal access to the "teacher" model, it's impossible for a third party to secretly distill data from a closed-source model, such as OpenAI's o1, as DeepSeek was rumored to have done.

Researchers are still discovering new applications for distillation. In January, the NovaSky lab at UC Berkeley demonstrated its effectiveness for training chain-of-thought reasoning models, which use multistep "thinking" for complex questions. Their open-source Sky-T1 model, trained for under $450, achieved results similar to much larger open-source models.  

The insights were published in Quanta Magazine.

Other interesting reads this week-

How a third parent's DNA can prevent an inherited disease

Thanks for reading.

Do share this newsletter with your friends.

Until next time,
Adya