#11 | Fortifying Table Salt May Help Prevent Birth Defects

Quote of the Week 💬 

Adding Folic Acid To Table Salt Prevented Birth Defects

⌛ The Seven Second Summary: A clinical trial in rural India had women consume folic acid-fortified iodized salt in their regular diet for a few months and found it significantly increased their serum folate levels. This may prevent folate deficiency and reduce the risk of serious and fatal birth defects.

🔬 How It Was Done:

• The study enrolled 83 nonpregnant women between 20 - 44 years old in 4 rural villages across Southern India.

• Over a period of 4 months, the participants used iodized salt enriched with folic acid as table salt for their daily meals.

• The researchers estimate their participants consumed ~0.3 grams of folic acid-fortified iodized salt per day over the course of their 4 month study.

🧮 Key Results: The median participant experienced a 3.7x increase in their serum folate levels by the end of the study. This increase was both statistically significant and substantial enough to raise the median participant’s serum folate levels to healthier levels for pregnancies.

💡 Why This May Matter: Folic acid is a synthetic form of Vitamin B9. It is an important nutrient for women before and during pregnancies because it promotes healthy cell growth and development for the baby's brain and spinal cord. While Vitamin B9 is present in many foods and folic acid has been enriched in staple grains in many countries, according to the researchers, more than 100 countries have yet to implement similar food enrichment practices. Thus, this study serves as a small-scale demonstration of how new and existing food intervention policies can improve societal health outcomes.

🔎 Elements To Consider: This study was not randomized, which limits the causality that can be inferred from its findings. Furthermore, there is little discussion about how countries and regions can enact policies to enrich commercial salt with folic acid in a manner that will maximize benefits and minimize consequences for its people.

📚 Learn More: Emory. JAMA Network.

Stat of the Week 📊 

AI Predicts The Risk Of Lung Cancer Spreading To The Brain

87%

⌛ The Seven Second Summary: A research team from Washington University School of Medicine in St. Louis used AI to predict whether early-stage lung cancer will metastasis and spread to the brain by analyzing images of lung tissues.

🔬 How It Was Done:

• 118 lung biopsy samples of stage 1 - stage 3 were collected of various patients who did and did not develop brain cancer over a five-year monitoring period.

• This dataset was used to train a machine learning algorithm to detect specific attributes in images and predict whether patients developed brain cancer from those attributes. Some of the attributes the model relied on for its predictions included the shape, location, and presence of various tumor, tissue and immune cells.

• After completing its training, the researchers tested the model using 40 lung biopsy samples it had not seen before and compared its performance to four expert lung pathologists.

🧮 Key Results: The model was 87% accurate at predicting patients who developed brain cancer from the 40 biopsy samples it was tested on, whereas the four pathologists were 54% - 61% accurate. Notably, the model was more accurate at identifying patients who did not develop brain cancer than patients who did.

💡 Why This May Matter: AI models can classify and consider a broad range of attributes for image assessments that are not easily discernible or quantifiable in traditional tumor grading techniques with the human eye. This is one possible explanation for its impressive performance relative to clinical experts. This is yet another example of the potential this technology has to improve early diagnostic screenings and ultimately benefit patients with better quality care.

🔎 Elements To Consider: This was a small study. The model still needs to be tested on much larger patients cohorts to determine if its performance is consistent and reliable.

📚 Learn More: WUSTL. Journal of Pathology.

AI x Science 🤖

A Robotics AI Model For Reasoning

Covariant AI announced the forthcoming release of its first foundational AI model for robotics, RFM-1. While this is still a young company, this is meaningful news for the field of robotics for reasons similar to Google Deepmind’s RT-X model announcement last year.

First, the model is multimodal, meaning it can understand text, images, videos, sensor data, and other data required to navigate a physical environment. This also means robots powered by this model can be trained on Internet data similar to AI models like ChatGPT, which should allow it to understand and execute commands it has never encountered before by applying basic reasoning. For example, if a worker were to tell a robot to “throw away the trash”, it should be able to understand what “trash” means and where it belongs in its environment, and it should know not to physically “throw” whatever it picks up like a baseball.

Second, this is meant to be a general purpose model, meaning one model is used to instruct how many robots will perform a variety of tasks. This is different from how many machines are trained today, where different robots are powered by unique models that are programmed to execute specific tasks, such as identifying, grabbing and moving a set of objects in a kitchen, factory or some other environment. Since only one foundational model will be used in this instance, data from customers with completely different types of robots will effectively be training each other on the platform so they can all perform new and existing tasks more effectively over time.

Whether Covariant AI is successful in the long-run is less important than the general trend taking place. The field of robotics has recently experienced a breakthrough in how to efficiently train machines, using more data at larger scales, with a single model. This is why automakers like BMW and Mercedes have inked deals with AI robotics companies in 2024, and it is why we have a relevant video in our Media of the Week section that you don’t want to miss. Covariant AI.

Our Full AI Index

• Research: Researchers from Google DeepMind and the University of British Columbia shared results from their work on SIMA, a generalist AI agent to complete tasks in a variety of video games. They did this research in an effort to build an AI agent that can translate language commands into useful actions, and video games are a safer place for AI agents to train in than a real world setting. Notably, the more games this AI agent was trained on, the better it got overall. It reached a point where SIMA performed as well as a specialized agent trained exclusively for its specific game, even when SIMA was not trained on that game at all. Google Deepmind.

• Open Source: xAI has open sourced Grok-1. With 314 billion parameters, this is now the largest open source large language model on the market. xAI. Github.

• Cultural Events: 5 of the 45 (11%) finalists in this year's Pulitzer Prize’s for journalism disclosed using AI to help them research, report or tell their stories, according to the organization’s administrator. This is the first year The Pulitzer Board has added an AI usage disclosure requirement in its application process. Nieman Lab.

• Policy: Last week, the European Parliament passed its landmark Artificial Intelligence Act, establishing the world's most comprehensive set of regulations for AI thus far. The regulation outlines details such as AI apps that will be banned, obligations for AI when used in healthcare, banking, employment and other “high-risk” settings, and transparency requirements for AI systems. European Parliament.

Other Observations 📰

Cystic-Fibrosis Breakthroughs Extend Patient Lifespans

The Atlantic shared a lovely story about people with cystic-fibrosis and how a treatment option approved by the U.S. Food and Drug Administration in 2019 called Trikafta is allowing people with the disorder to live longer and healthier lives than ever before.

Cystic-fibrosis is a genetic disorder that primarily affects the lungs in a chronic and progressive manner. It is caused by mutations in a specific gene that leads to the production of thick and sticky mucus. If left untreated, this can cause serious infections and respiratory issues over time. Trikafta corrects the misshapen protein responsible for the disorder, which causes the thick mucus in the lungs to thin out until patients are able to cough it up. The article shares stories from patients who take the medication, experience a dramatic improvement in lung function and are able to breathe freely as a result.

Our favorite passage: “Cystic fibrosis (CF) once all but guaranteed an early death. When the disease was first identified, in the 1930s, most babies born with CF died in infancy. The next decades were a grind of incremental medical progress: A child born with CF in the ’50s could expect to live until age 5. In the ’70s, age 10. In the early 2000s, age 35. With Trikafta came a quantum leap. Today, those who begin taking the drug in early adolescence, a recent study projected, can expect to survive to age 82.5—an essentially normal life span.“ The Atlantic.

Our Full Science Index

• Energy: Consensus is building around 2035 as the date for eliminating fossil fuels from Europe's power system. Ten member states, representing more than 60% of the region’s energy sector, have now committed to eliminating fossil fuels by 2035 or sooner. Euronews.

• Cancer: Fewer middle-aged people are dying from cancer in the UK than at any point over the last 25 years. While cases of cancer in middle-aged people has risen between 1993 - 2018, the number of men and women dying from the disease fell by 37% and 33% respectively over the same time period. Cancer Research UK. BMJ.

• Public Health: The number of children who died before their 5^th birthday has reached a historic low, with the global under-5 mortality rate declining by 51% since 2000. Several countries have outpaced the global rate as well, with some achieving declines as high as 75% since 2000. WHO.

• Weather Forecasting: 4-day weather forecasts today are as accurate as a 1-day forecasts from 30 years ago. Today, the accuracy of 3-day forecasts is ~97%. Our World In Data.

• Animals: Researchers in India may have documented the first known cases of Asian elephants burying their dead calves in irrigation ditches. This fascinating discovery revitalizes ongoing questions about the species’ sapience and unique sense of culture. Times of India. Journal of Threatened Taxa.

• Space: SpaceX successfully conducted a test launch of its massive Starship rocket into space. Starship is the largest rocket ever built, standing nearly 400 feet tall. NASA is counting on the vehicle to return astronauts to the moon with its Artemis 3 mission in ~2026. This successful test is one of many steps required to accomplish the goal. SpaceX.

• Policy: The U.S. Environmental Protection Agency announced a ruling to reduce toxic emissions used in commercial sterilization facilities by more than 90%. This ruling should reduce lifetime cancer risks for people working in and living nearby these facilities across the country. EPA.

Media of the Week 📸 

Figure AI Partners With Open AI To Give A Voice To Their Robots

Here is the video we promised from our AI x Science section above. If you wanted to visualize what we were explaining, Figure AI’s latest demo is the representation for you. We are on the cusp of a new robotics age and it is incredible to witness.

The Remains Of A Star That Exploded Nearly 11,000 Years Ago

Astronomers used the Dark Energy Camera to construct an image of the Vela Supernova Remnant, a nebula of cosmic debris left over from a massive star that exploded about 800 light years away from Earth and 11,000 years ago. Noir Lab.

This Week In The Cosmos 🪐

March 24/25: A Penumbral Lunar Eclipse. It’s the first eclipse of 2024 and will be visible across North, Central and South America.