Advertise on podcast: Intellectually Curious

What if describing the material you want could help you find promising candidates? We explore CLaC, an AI model that connects three-dimensional crystal structures with language. Researchers trained it using descriptions generated from computed material properties, helping it relate words and numbers to the arrangement of atoms. The study shows how this approach can search and rank existing crystal structures in response to requests about properties such as density or band gap. We unpack why carefully grounded synthetic descriptions worked better in these experiments than text extracted from scientific papers, and how language-guided screening could make materials research easier to navigate. The result is a step toward more intuitive discovery tools, with candidates still needing scientific evaluation. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

What does it take to make hundreds of thousands of GPUs work together? We step inside Prometheus, Meta's gigawatt-scale AI supercluster in New Albany, Ohio, whose first phase helped train the Muse family of models. Meta says the system already spans multiple connected buildings, with plans to reach roughly one million GPUs at full scale. We explore the fast networks that keep training coordinated, spare GPUs and saved checkpoints that help jobs recover, and liquid cooling that makes dense hardware practical. Power management and careful monitoring also play their part. It's a look at how many engineering improvements combine to keep an enormous system doing useful work and support the development of future AI models. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

Four AI agents, one shared world, and an Ender Dragon to defeat. We explore how Gemini 4 Argon coding agents completed a single Minecraft survival run on Easy difficulty, writing JavaScript programs using Mineflayer and reading text feedback. Over 27 hours and 14 minutes, they gathered resources, built tools, and worked toward a common goal. The story echoes Voyager's earlier work on reusable coding skills, while bringing teamwork into focus. With no assigned roles, the agents divided tasks, exchanged lessons through shared notes and game chat, and helped one another recover from setbacks. It's a vivid example of how memory, reusable code, and cooperation can help AI agents sustain a complex project over time. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

How can a huge AI model put only part of its network to work at a time? Drawing on Microsoft's pre-training report, we explore MAI-Base-2 and its mixture-of-experts design, where the total parameter count is much larger than the portion used for each token. We also explain how smaller experiments arranged in "scaling ladders" helped guide the full training run. Then we unpack "dropless" processing: extra rounds let overloaded experts process overflow tokens without dropping them. It's an example of model design and computing infrastructure being developed together, with memory use, hardware efficiency, and learning quality all part of the same problem. Careful experiments and practical engineering offer a path toward making large-scale AI training more effective. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

What does it take for an AI agent to turn a customer's request into a useful working system? We explore FDE-Gym, a benchmark inspired by forward-deployed engineers—the people who work with customers to clarify needs, uncover constraints, and integrate software into existing operations. Its simulated environments are adapted from real projects, giving agents a place to tackle the details of delivery. Success is measured by improvements to the customer's business metric, with the cost of stakeholder interactions also counted. We look at why asking good questions and writing working code must connect to the right outcome. The findings highlight opportunities for improvement, while a shared benchmark and online practice environment offer a way for people and AI agents to develop their delivery skills. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

Does every step of an AI task need a large language model? We explore JevAgentHarness, a research framework that lets smaller decision models handle routine multiple-choice decisions, such as choosing the next tool. Confidence thresholds fitted to an agent's own runs determine when to delegate and when the larger model should take over. The key is testing how choices play out across a whole task, then training on the situations agents actually encounter—including those created by their own mistakes. We look at AgentDecisionBench and the study's promising gains in speed and cost, while keeping task success in view. It's a practical direction for AI agents that spend their effort more wisely. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

What changes when an AI research assistant can coordinate a team of coding agents? We explore Sakana AI's Agentic Scientist, a system designed to turn research ideas into experiments by dividing work among agents, comparing results with baselines, and testing which parts of an approach make a difference. Researchers can steer the process as the work develops. We follow the path from setting up a task to running experiments and writing papers grounded in the results. One example is a co-produced paper accepted at ALIFE, with humans verifying the work and editing the manuscript. It's a look at AI as a research teammate—and how parallel experimentation and human judgment can work together to move ideas forward. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

Can a computer find the shortest route to every destination without sorting those destinations by distance? We explore the insight behind the 2025 and 2026 advances in shortest-path algorithms, using a simple city-map analogy to make a classic computer science problem feel familiar. Dijkstra's algorithm provides the starting point for a story about grouping work, choosing useful representatives, and avoiding unnecessary comparisons. The new results improve theoretical running-time bounds for directed networks with nonnegative edge weights, especially sparse ones. Practical speed still depends on the implementation and the network being searched. It's an encouraging example of how fresh thinking can uncover possibilities in problems researchers have studied for decades. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

How do you find a baby planet hidden in the dusty surroundings of a young star? We explore Elias 2-24 b, a giant planet in a system estimated to be less than a million years old. Follow the clues from ALMA's images of a disk gap to infrared observations and a fresh look at archival Keck data. This remarkably young world gives astronomers a chance to investigate how giant planets assemble and shape their surroundings. We unpack what the observations reveal, why estimates of age and mass depend on models, and how studying planetary beginnings elsewhere can help illuminate the early history of our own solar system. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

What could an AI assistant do if it carried useful context between conversations and kept working when a task got difficult? We explore Meta's Muse Wonder and Muse Wonder Full, focusing on Active Memory and Active Focus: features designed to recall relevant details, try alternative approaches, recover from errors, and check in before important actions. From building an app to working through complex documents, we consider the possibilities for more useful personal and professional assistance. We also explain the launch access: Muse Wonder through the Meta Model API and Muse Code, and Muse Wonder Full for Muse Code subscribers. It's a look at Meta's ambitions, the promise of more persistent agents, and why reliable follow-through matters as much as an impressive answer. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

What would the sky look like in ultraviolet light—and how do you map regions no UV telescope has observed? We follow astrophysicist Brice Ménard's collaboration with Claude Science to combine scattered telescope surveys, calibrate their differences, and estimate the missing pieces using observations at other wavelengths. Roughly a third of the finished map is predicted, with measured and predicted pixels labeled separately and uncertainty estimates included. We explore the work behind that distinction, including the atmospheric glow Ménard spotted after AI reviews missed it. It's a story about using AI to complete a long-delayed scientific resource, and why careful human inspection still matters. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

AI agents can search scientific literature, connect ideas, and coordinate specialized models such as AlphaFold. In this episode of Intellectually Curious, we explore how these tools could expand scientific discovery—and what happens when promising predictions arrive faster than researchers can test them. Drawing on the DeepMind Institute's "Bending the Curve of Discovery," we follow the path from computer-generated hypotheses to laboratory evidence. Human judgment, careful verification, and testing capacity remain essential. We also consider whether AI encourages bolder questions or safer, familiar ones, and what scientific institutions need to change to turn faster research into lasting progress. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

Start with the prime numbers, take the absolute differences between neighbors, and repeat. Gilbreath's conjecture predicts that every new row begins with a one. This simple pattern leads to a stubborn mathematical mystery, supported by extensive computation but still awaiting a proof for the primes. We explore what researchers can establish in randomized settings, including the Cramér model, and how those results help illuminate the pattern. Along the way, we consider the gap between compelling evidence and mathematical proof: results about random models offer insight while the original conjecture remains unsolved. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

What can a shadow tell us about the world? We explore how geometry, the wave behavior of light, and the atmosphere shape the shadows we see—and how those patterns carry clues about objects, materials, and space. From human perception and computer vision to astronomy, architecture, and solar energy, shadow physics turns an everyday sight into a useful source of information. This episode follows the connections between how shadows form and how scientists and engineers read them. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC

A coin-flipping game has a mathematically infinite average payout. So why are most people willing to pay only a modest amount to play? We explore the St. Petersburg paradox, where a straightforward probability calculation collides with everyday judgment about money and risk. The puzzle opens up questions about how we value wealth, weigh unlikely outcomes, and account for a house with limited funds. From expected utility to finite bankrolls, we examine why a theoretical average can leave so much out of a practical decision. Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information. Sponsored by Embersilk LLC