Gencast Weather Forecasting, Venus Ocean History, Dual-Acting GLP-1 Leptin Drug, and Crustacean Pain — Naked Scientists

Episode overview

The Naked Scientists explore breakthroughs across weather forecasting, planetary science, obesity research, crustacean welfare, and cosmology in a single episode. The key segments include the gencast weather model, Venus ocean history, a dual GLP-1 leptin weight loss molecule, crustacean pain experiments, and a Q&A on galaxy formation timescales.

• Gencast delivers fast, energy-efficient, ensemble-based weather forecasts trained on four decades of data and runs on a single TPU chip.

Gencast and the future of weather forecasting

The first major topic centers on gencast, a new AI-based weather forecasting model developed by Google DeepMind. Unlike traditional physics-based forecasts that rely on massive supercomputers running atmospheric simulations, gencast leverages machine learning to extract patterns from around four decades of historical weather data. The model is trained to predict weather states forward in time, learning dynamics directly from past data. It is a global model, necessary for medium-range forecasts up to about 15 days, because atmospheric dynamics are interconnected across the globe. Instead of a single forecast, gencast uses an ensemble approach, generating multiple plausible future scenarios to quantify the likelihood of extreme events and forecast trajectories, including tropical cyclone tracks. In terms of performance, while a single numerical value is difficult to pinpoint, the research reports that gencast improves upon headline metrics in most tasks, and in certain critical cases, such as extreme temperatures and hurricane tracks, it delivers meaningful advantages—roughly a 12 hour accuracy gain in cyclone trajectory forecasts compared with the state-of-the-art models. Energy efficiency is a major benefit: a 15-day forecast can be produced in eight minutes on a single TPU chip, compared with hours on large-scale supercomputers. The work is published in Nature and features insights from Ilan Price and Remy Ilam of DeepMind, who describe how an emphasis on data rather than compute could accelerate research and broaden access to weather forecasting technology.

Venus and the water question

In Cambridge, researchers led by Teresa Constantinou have interrogated the long-standing question of whether Venus could have hosted oceans. Despite Venus being similar in size to Earth, its current surface conditions—high pressure, a CO2-rich atmosphere, and sulfuric acid clouds—are vastly different from Earth’s oceans. The team uses observations of Venus’s atmospheric gases, measured by past and current missions and ground-based telescopes, to infer the water inventory inside Venus and its volcanic outputs. The results indicate that Venus is currently very dry, with little water being released by volcanoes compared to Earth. The interpretation hinges on the idea that Venus, being closer to the Sun, experiences stronger solar radiation that causes water in the atmosphere to dissociate into hydrogen and oxygen; hydrogen escapes more readily due to its light weight, leaving Venus desiccated. This work, published in Nature Astronomy, challenges the notion that Venus might have been a water-rich world like Earth in its past and uses present-day atmospheric chemistry as a window into the planet’s climatic history.

Weight loss and a dual GLP-1 leptin drug

Another major segment covers obesity research, focusing on a double-ended molecule developed by Novo Nordisk that combines a GLP-1 receptor agonist with a leptin-mimicking component. Randy Seeley explains that leptin signals inform the brain about stored calories, but in obesity, the brain becomes resistant to leptin. The dual-acting molecule is designed to hit a specific subset of hypothalamic neurons that express both the GLP-1 receptor and the leptin receptor. The study uses genetic tools to manipulate leptin receptor expression in those neurons, showing that the dual-action signal is necessary and sufficient for the observed weight-loss effects. The hypothesis is that the GLP-1 signal primes the system, re-sensitizing or enabling leptin signaling to drive weight loss more effectively than GLP-1 agonists alone. While direct clinical comparisons to existing GLP-1 drugs aren’t made, the research suggests this approach could enhance weight loss beyond the typical 10–15% seen with current therapies during standard treatment, potentially offering a significant improvement in obesity management in the future. The work is published in Science Translational Medicine and illustrates how combining hormonal signals could rewire brain circuits involved in appetite and energy balance.

Crustacean pain and welfare considerations

Terra Constantinou’s team reports on crustacean pain perception. By applying acetic acid to sensitive tissues and monitoring neural activity, they demonstrate the capacity of crabs to detect a noxious stimulus. They further show that crustaceans will seek analgesia, such as aspirin, after a painful stimulus, indicating not only the perception of pain but a motivation to alleviate it. The discussion extends to industry and laboratory practices, arguing for methods to kill crustaceans rapidly to minimize suffering, such as electrical stunning, and favoring dead crustaceans for sale when possible. The ongoing experiments include exploring the most humane methods for killing large lobsters and crabs and evaluating post-stimulus behaviors as part of the pain assessment framework.

Question of the Week: Galaxy formation timelines

In the science Q&A, Cambridge astronomer Matt Bothwell addresses how long it takes to form a galaxy. He explains that galaxy formation is a drawn-out process driven by gas dynamics and star formation. While there is no single moment that marks a galaxy’s birth, a rough benchmark is a billion years for a chaotic gas cloud to evolve into a disk-like, recognizable galaxy. Our Milky Way continues to form stars at a rate of about four to five per year, suggesting that galaxy formation is not finished even in mature galaxies. The discussion highlights how observations from the James Webb Space Telescope enable us to look back billions of years to study the evolution of early galaxies, and emphasizes that the process of galaxy growth is ongoing across cosmic time.