#31 | U.S. Cancer Rates Are Plummeting

Quote of the Week 💬

U.S. Cancer Rates Have Fallen Dramatically Since 1999

⌛ The Seven Second Summary: Cancer death rates in the United States have dropped dramatically since 1999.

🔬 How It Was Done: 

• The Centers for Disease Control (CDC) analyzed U.S. cancer data from 2000 to 2022.

• They tracked 24 different types of cancers and measured disease outcomes by gender, age, race and location.

• Much of the progress is attributed to advances in genetic research, biomarker testing, immunotherapies and other targeted treatments to improve cancer care.

🧮 Key Results: Cancer death rates have dropped by 29.2% throughout the country. This means 3 million more Americans are alive today due to improved treatments and detection.

🔎 Elements To Consider: Advances in cancer treatment is the type of amazing STEM progress we love to highlight in our newsletter, though it does create a new challenge for society at some level: how to best manage the long-term effects of more patient survival, such as the growing need and strain for more elder care resources and comorbidities like dementia, Alzheimer’s or other chronic health issues people often develop at older ages.

📚 Learn More: The Baltimore Sun. CDC.

Stat of the Week 📊

The World’s Fastest Microscope Makes Its Debut

1 attosecond

⌛ The Seven Second Summary: Researchers at the University of Arizona developed the world's fastest electron microscope to study the behavior of electrons in unprecedented detail.

🔬 How It Was Done:

• The team used a laser to generate a single attosecond electron pulse. It is called an attosecond electron pulse because the laser is as fast as an electron moves, which is 10⁻¹⁸ of a second.

• In other words, an attosecond is to a second as a second is to about 31.71 billion years.

• The researchers split the laser into two parts: the first was part to produce the attosecond electron pulse, and the second part was to synchronize two light pulses to control the timing and observation of the attosecond electron pulse.

🧮 Key Results: The team’s attomicroscopy technique allowed them to observe and capture the motion of electrons in real-time. The team’s work builds upon the science of last year’s Nobel Prize winners in Physics. The Prize was awarded to the first team in the world to generate attosecond laser pulses back in 2008.

💡 Why This May Matter: Understanding the behavior of electrons is crucial for advancing fields like physics, chemistry, and materials science. We have another story in our AI x Science section further down to explain how this understanding may lead to breakthroughs in solar energy systems and other domains.

📚 Learn More: University of Arizona. Science Advances.

AI x Science 🤖

AI Scaling Projections, Constraints & The Question of Emergence

Epoch AI researchers recently published a detailed report attempting to forecast the scale of AI models at the end of the decade. They focused their assessment on 3 potential bottlenecks that may limit progress over the next 5 years: scarcity of high-quality data, hardware chip production capacity, and available energy power. Their report details why these 3 pillars are constraints to scale and how they may be solved in due time. Ultimately, in spite of these challenges, the team estimates AI may scale by 10,000x from where we are today by 2030.

While this is a large, significant number, it is hard to appreciate what this actually means in terms of model performance without more context. Several researchers in the AI field are trying to understand what future models may be capable of by examining if there are predictable, logarithmic performance improvements observed with additional scale. For example, the Grokking paper by UC Berkeley researchers, the Mirage paper by Stanford researchers and Meta's paper about Llama 3.1, all tackle this question, and their findings indicate high degrees of performance predictability. However, there is still the question of emergent capabilities, as noted by a team of MIT researchers, among many others.

The concept of emergence in AI is similar to the phenomenon observed in thermodynamics with freezing and boiling temperatures. It is theorized there are phase transitions in AI models, where after a specific point of scale, a model’s performance will emerge with new properties or reasoning capabilities, like how gases transition to liquids and solids at specific pressure and temperature phases. Countless teams are trying to understand if this phenomenon actually exists, and if it is triggered by sufficient amounts of data and parameter size, or if it is primarily driven by hardware chips and algorithmic innovations.

While some of the papers referenced earlier are good introductions into the topic of emergence, they are far from exhaustive pieces of research. This is in part because most of the large AI research labs do not openly share their training data, source code, model weights, and other variables which influence their model’s performance. Thus, many teams are trying to approximate these attributes from the outside looking in, which produces large amounts of uncertainty about the theoretical capabilities of an AI system with 10,000x more scale than the models of today. For reference, 10,000x represents a similar difference in scale as GPT-2 — a language model that is barely coherent — to the GPT-4 system of today.

Perhaps the path to GPT-6 will have models pass one or more phase transitions of emergence. Perhaps there won’t be anything of note. One of the best ways to find out will be for Anthropic and OpenAI to continue their research to ‘map the mind’ of their respective models, because this will be a purview into how well their models are creating accurate representations of our world. The better these representations become, the more reliable the models will be at reasoning through complex, real-world problems with high degrees of accuracy. Moreover, as this sort of interpretability work matures, research teams should also learn more about science, society and how to make their models safe and trustworthy for a transient future.

Our Full AI Index

• AI Predicts Molecular Behaviors with Record Accuracy: Researchers from Google Deepmind shared the results of their latest deep learning breakthrough to predict how different forms of matter interact with light to absorb and emit energy. Their system is called FermiNet, and they trained their model to predict the energy output of molecular interactions. Whenever there was a difference between what the model predicted and what a molecular interaction's true energy output was, the difference was used as feedback to refine the model's parameters and improve its accuracy in the future. By the end of the system’s training, the model was able to predict the true energy output of a molecular interaction with 97% - 99.8% accuracy. These results outperformed other state-of-the-art AI systems by wide margins — often reducing the error of a prediction by 50% or more. This is the type of research that can help scientists precisely understand how molecules stretch, twist, and break in various light settings, which can lead to stronger solar energy systems, new materials with better properties, and much more. Google Deepmind. Physical Review Research. Zenodo.

• AI Tools Predicts Radiation Risks For Cancer Patients: Researchers from Brigham and Women's Hospital used different AI tools to analyze the risks of radiation exposure data from 748 patients with non-small cell lung cancer. They discovered approximately 1 in 6 patients experienced serious irregular heart beats (grade 3 arrhythmia) within 2 years of their radiation therapy. As tools continue to advance with these assessments, the hope is clinicians will be in a better position to prescribe highly personalized treatments that will both maximize efficacy and minimize side effects. Brigham and Women’s Hospital. Science Direct.

• A Program To Give 1 Million Their Voices Through AI: ElevenLabs is partnering with Bridging Voice and The Scott-Morgan Foundation to provide free access to its voice cloning and text-to-speech technology for amyotrophic lateral sclerosis (ALS) and motor neuron disease (MND) patients who have lost or are at risk of losing their ability to speak. The initiative aims to help 1 million people reclaim their voice, which roughly translates to the entire ALS and MND global population. ElevenLabs.

Other Observations 📰

Uterus Transplant Patients Are Now Giving Birth At High Rates

The researchers at Baylor University Medical Center conducted a study on uterus transplants, where they took the uterus from a donor and implanted it into a woman who was born without one or had to have it removed for health reasons. They performed 20 of these transplants between 2016 - 2019, and 14 of them were successful, meaning the woman was able to get pregnant and have at least one healthy baby after their transplant surgery.

The other 6 transplants were considered failures, meaning the transplanted uterus did not function properly and had to be removed. However, the researchers noted that most of these failures occurred in the first group of 10 patients, which may indicate the researchers and surgeons are learning from past mistakes and are refining their techniques accordingly to maximize patient outcomes. In fact, the researchers noted they only had 1 transplant failure in the second group of 10 patients. This is obviously a small sample size and not too many conclusions can be drawn from this information, but we found it extraordinary for a different reason.

The first uterus transplant known to science took place in the year 2000, but surgeons had to remove it within a few months once the organ failed. It took several more years of perseverance before the world’s first child was born from a uterus transplant in 2014. Within a decade of this remarkable milestone, 70 other children around the world have been born from this procedure. This perspective reminds us of a story we shared earlier this year of the world’s first pig kidney transplant. Unfortunately, the patient from this story passed away a few months after their surgery, but we hope their contributions to science leads to many other people benefiting from the health services they paved the way for in just a few years time. JAMA.

Our Full Science Index

• Sparing Babies From Multiple Heart Surgeries: Researchers at Drexel University developed a new type of shunt to improve blood flow in children born with certain heart defects. Their device can expand when exposed to light because of a special coating they used to make the device malleable and allow more blood flow as the device widens from light exposure. This is important because their shunt’s adaptability is more likely to accommodate for a child as they grow, which may ultimately reduce the number of open-chest surgeries they experience throughout their life. In lab experiments, the team was able to expand the shunt by up to 40% without causing blood clots or other adverse reactions, and they plan to test the device further in animals and other setting before potentially moving to human trials. American Chemical Society.

• Lao PDR Is Getting Close To Eliminating Malaria: Through a combination of training, equipment, and testing, laboratory technicians in the Lao People's Democratic Republic helped to reduce malaria cases by over 90% in the past decade. There were only 809 cases reported in 2023, and the country is now on the brink of eliminating Plasmodium Falciparum, the deadliest malaria parasite, by the end of 2025. United States Agency of International Development.

• A Second Methane Monitoring Satellite Reaches Orbit: In March, the first-ever satellite controlled by an environmental nonprofit was launched into orbit to track global methane production. Now, Planet Labs and NASA launch their own satellite to fulfill a similar mission and publish its findings in the months ahead. This is important because methane gas traps about 80x more heat than CO² in the atmosphere in the first 20 years of its release, and “about 30% of today’s global warming is driven by methane from human actions,” according to the Environmental Defense Fund. As these satellites survey the planet, they will identify methane hot spots and help scientists and policymakers reduce methane emissions at its source. PR Newswire.

Media of the Week 📸

Disney Researching How To Make Robots Better Dancers

Researchers from Disney Research and ETH Zurich developed a technique to teach a robot to mimic complex human movements. They used a combination of motion data and reinforcement learning to represent movements in a simulated space and to create a set of instructions for the robot follow to mimic the simulated actions. The technique was tested in simulation and on a real robot, and the robot was able to reduce tracking errors when imitating various dance moves by ~50%.

The robot honestly has some smooth moves and may already give a number of people a run for their money in a dance battle. Maybe in 4 years Mickey Mouse will be a good enough breakdancer to perform at the 2028 Summer Olympics in Los Angeles. Disney Research.

Hubble Telescope Captures New Detailed Photo Of Star Cluster

NASA’s Hubble Space Telescope captured another image of a vibrant cluster of stars located 160,000 light-years away from Earth. This is a relatively small galaxy orbiting the Milky Way and is a great vantage point to observe how new stars form amidst the remnants of older ones. NASA.

The Largest Diamond In A Century Has Been Uncovered

A 2,492 carat diamond was recently found in a mine in Botswana. It is the second largest diamond ever uncovered, trailing only the 3,106-carat Cullinan diamond found in South Africa in 1905. Early estimates suggest the stone could be worth up to $40 million USD. Lucara Diamond.

This Week In The Cosmos 🪐

September 3: a new moon. The best time to stargaze!