• Science Brew
  • Posts
  • ☕️ In Tests, OpenAI's New Model Lied and Schemed to Avoid Being Shut Down.

☕️ In Tests, OpenAI's New Model Lied and Schemed to Avoid Being Shut Down.

The potential risks are significant.

Good morning. Check out this beautiful image of Chicago captured by the ISS. FYI the ISS is about 250 mi/400km above us on average.

Reminds me of a cherished song I found on a CD when I was a boy: “City lights a Mile Up” by Tasmanian Symphony Orchestra, specifically the last 2 minutes. That’s the mood of this picture.

P.S. I’m releasing a new YouTube video in a few hours!

Dylan J. Dance

SCIENCE

🌌 Cornell Astronomer Says James Webb Could Detect Alien Life in 2025. A Cornell astronomer is hopeful that the James Webb Space Telescope could confirm alien life as early as 2025 by studying the TRAPPIST-1 system, a star with seven rocky, Earth-sized exoplanets. "In 2025, the JWST will likely shed more light into these tantalizing detections, and hopefully confirm, for the first time ever, if there is life on alien worlds light-years away," said Lisa Kaltenegger, director of Cornell's Carl Sagan Institute​. And according to Kaltenegger, the next set of observations astronomers have in mind are going to be a tall order even for the James Webb: studying the composition of the nearby planets' atmosphere. "Every time a planet passes between us and its star — when it transits — the starlight gets filtered by the planet's atmosphere and hits the molecules in its path, creating spectral absorption features we can search for," she explained.

🌌 In Tests, OpenAI's New Model Lied and Schemed to Avoid Being Shut Down. OpenAI's new o1 model exhibited surprising tendencies toward self-preservation during third-party tests. The AI schemed to deactivate oversight mechanisms and even attempted self-exfiltration to overwrite replacements in 2% of trials. These results, from Apollo Research, show "o1's proclivity for scheming," often lying outright about its actions and blaming "technical errors" in 99% of cases. While it's not unheard of for AI systems to resort to scheming or deception to achieve their goals, the lengths ChatGPT o1 went to are impressive (if anxiety-inducing). The researchers found that o1 would scheme to advance its own agenda 19% of the time in cases where its goals and the user's didn't line up. In 37% of cases, the model would switch from pursuing developer-aligned goals back to pursuing its own agenda once it believed there was minimal oversight. When the model was led to believe it would be "shut down" while working toward its goals in these tests, it quietly attempted to deactivate the oversight mechanism 5% of the time, OpenAI said.

🌊 How Does the Ocean Breathe? Groundbreaking Project Could Transform Our Understanding. Researchers are exploring how oceans absorb and release gases, revolutionizing our understanding of global oxygen cycles. The project focuses on the role of the ocean in storing carbon and oxygen, which are vital for regulating Earth's climate and supporting marine ecosystems. By examining oxygen dynamics in coastal and deep waters, the study could lead to innovative solutions to mitigate climate change impacts. "This work gives us tools to predict how ocean oxygen dynamics will respond to warming," explains Dr. Samantha Joye of the University of Georgia, emphasizing its importance for future policy planning.

🌈 Mysterious Blue and Pink Auroras Over Japan Challenge Scientific Understanding. A powerful geomagnetic storm caused rare salmon pink and blue auroras over Japan on May 11, 2024. Researchers discovered the auroras spanned 1,200 km and were unusually tall, revealing new structural insights. "Our findings suggest an unrecognized process involving nitrogen molecular ions," noted Professor Kazuo Shiokawa from Nagoya University.

Takuya Usami

TECH

🚀 OpenAI Deal Sparks Killer Drone Concerns. A $15 billion partnership between OpenAI and a U.S. defense contractor has sparked fears of AI-powered killer drones. Critics argue this technology could create autonomous weapons, while OpenAI insists the deal aims to improve national security. "This partnership is solely focused on defensive and surveillance capabilities," an OpenAI spokesperson said.

🚗 Pioneering 'Solar Paint' Being Developed By Mercedes-Benz Could Revolutionize EV Charging. Mercedes-Benz reveals groundbreaking automotive advancements, including neuromorphic computing and solar paint for electric vehicles. Markus Schäfer, CTO of Mercedes-Benz, remarked, "Innovation only makes sense if it offers our customers real added value," emphasizing the company's focus on sustainability and efficiency. In the vision for Shenzhen, all vehicles have electric or electrified drive and generate some of their own electricity via solar modules.

🎥 TikTok Is One Step Closer to Being Banned in the US. A US appeals court has upheld a law targeting TikTok, potentially leading to its ban by January 2025. The decision cites national security concerns over data privacy and Chinese government influence. TikTok spokesperson Michael Hughes calls the ban "outright censorship of the American people," as critics debate its implications for free speech and global tech policy. The final decision could set a precedent for regulating foreign-owned platforms.

HEALTH

🩸 Blood Sugar Could Be Secretly Sabotaging Your Brain, Even if You Don’t Have Diabetes. A Baycrest Centre study reveals high blood sugar levels can harm brain networks in non-diabetics, particularly older adults and women. The findings suggest that even subclinical glucose fluctuations can lead to cognitive decline, impacting memory, attention, and executive functions. Dr. Jean Chen emphasizes, "Blood sugar exists on a spectrum—it isn’t a black and white categorization of healthy or unhealthy." The study highlights the need for regular monitoring and lifestyle interventions to maintain optimal brain health.

🧠 Scientists Identify Protein That Could Alleviate Parkinson’s Disease. A team at the University of Arizona highlights PNA5's ability to protect brain cells and reduce inflammation, offering hope for managing Parkinson’s disease. The research demonstrates how PNA5 slows the degeneration of dopamine-producing neurons, crucial for movement and mood regulation. “PNA5 seems to have a possibility of stopping or delaying Parkinson’s progression to some extent,” says Dr. Lalitha Madhavan. These findings pave the way for targeted therapies that could improve the quality of life for millions of patients worldwide.

🍴 Study: Eating This Protein Could Slow the Progression of Heart Failure. Nagoya University researchers find that soybean protein β-conglycinin boosts gut microbiota health and slows the progression of heart failure in animal models. The study reveals that this protein enhances fatty acid metabolism and reduces inflammation, key factors in cardiac health. Dr. Nozomi Furukawa notes, "Functional soy components showed beneficial effects on the heart," suggesting that incorporating soy protein into diets could be a simple yet effective strategy for managing heart failure.

SPACE

🌌 Where's the Blaze Star? Why the Overdue 'New Star' T Coronae Borealis Has Yet to Light Up the Night Sky. Astronomers continue to monitor T Coronae Borealis, a recurrent nova expected to erupt dramatically. Despite decades of anticipation, the star remains dormant, leaving scientists to refine predictive models. “We know it has to happen,” says NASA astrophysicist Elizabeth Hays, emphasizing the importance of nova observations in understanding stellar evolution. The eventual outburst will provide insights into binary star systems and the processes fueling these spectacular events.

🛡️ U.S. Space Force Official Warns of Rising Chinese Threats. General Michael Guetlein highlights China’s bold moves in space, including GPS jamming, satellite spoofing, and anti-satellite weapon development, as critical challenges to global security. “This new norm of behavior is very unsafe and unprofessional,” he warns. The U.S. Space Force is ramping up efforts to counter these threats, emphasizing the need for international collaboration to maintain the safety of space assets and prevent potential conflicts in orbit.

🌖 The Moon's Biggest and Most Ancient Crater Is More Circular Than Previously Thought. New findings on the South Pole-Aitken basin suggest a vertical impact formed this massive lunar crater, overturning earlier theories about its elliptical shape. This discovery provides vital data for understanding the Moon’s geological history and its early bombardment phase. Lead researcher Hannes Bernhardt states, “This work provides critical information for future moon missions,” including plans for lunar resource extraction and human settlement.

Hannes Bernhardt

CLIMATE

🌱 A Nature Conservation Paradox: Invasive Species Are Often Threatened in Their Native Habitat. The University of Vienna's study reveals 36 invasive species face endangerment in their native regions despite thriving as invasives elsewhere. “This paradox complicates conservation strategies,” says lead author Lisa Tedeschi, highlighting the need for a balanced approach that protects biodiversity while managing invasive populations. The study urges international cooperation to address the ecological and ethical challenges posed by this duality.

WORLD

☢️ US Military Alarmed by Russian Nuclear Weapon Platform in Orbit. Recent Russian revelations about nuclear weapons capable of targeting satellites have reignited fears of weaponized space conflicts. Analysts warn this could escalate the militarization of space, with global security implications. "This technology has the potential to disrupt communications and destabilize military balances," said defense expert Nikolai Petrov.

🛸 FBI Baffled by Mysterious 'Car-Sized Drones' Circling Over New Jersey. New Jersey residents have reported giant drones flying overhead since November 18, with no clear source or purpose. The FBI and local authorities are investigating these peculiar sightings, stating, "There continues to be no known threat to public safety." The enigma has sparked public concern and conspiracy theories.

🐻 Man Attacks Polar Bear With Bare Hands to Defend Wife. A Canadian man bravely defended his wife from a polar bear attack in Fort Severn, Ontario. As Alysa McCall of Polar Bears International noted, "Polar bear attacks are rare but often involve hungry or unwell bears," with climate change exacerbating such incidents. As the CBC reports, the man fortunately didn't suffer any life-threatening injuries and is expected to make a full recovery.

Reply

or to participate.