The Quanta Podcast

Reversible programs run backward as easily as they run forward, saving energy in theory. After decades of research, they may soon power AI.
The story How Can AI Researchers Save Energy? By Going Backward first appeared on Quanta Magazine.

We already know that what we eat, drink, and inhale can affect which parts of our DNA are expressed, and which aren’t. But recent research poses a shocking idea: A dad’s habits may be encoded in molecules and transmitted to his future kids. On this episode, host Samir Patel and biology editor Hannah Waters dig into the new epigenetic mouse studies exploring whether sperm cells carry more than just genetic information. This topic was covered in a recent story for Quanta Magazine.  
Each week on The Quanta Podcast, Quanta Magazine editor in chief Samir Patel speaks with the people behind the award-winning publication to navigate through some of the most important and mind-expanding questions in science and math.
Audio coda in this episode: Motivation and reward in learning - Produced by the Institute of Human Relations at Yale University, Published by Penn State University, Psychological Cinema Register [1948].

Imagine you’re holding two equal-size dice. Is it possible to bore a tunnel through one die that’s big enough for the other to slide through? It is — but what about other shapes? In a paper posted online in August, two researchers describe a shape with 90 vertices and 152 faces that they’ve named the Noperthedron, the first convex polyhedron that definitely cannot pass through itself. 
In this episode, Quanta contributor Erica Klarriech tells host Samir Patel about how the researchers discovered the shape, and how it solves a centuries-old geometric mystery. 
Audio coda courtesy of the Gemsmen Renaissance Consort.

Studies of neural metabolism reveal our brain’s effort to keep us alive and the evolutionary constraints that sculpted our most complex organ.
The story How Much Energy Does It Take To Think? first appeared on Quanta Magazine.

Ask ChatGPT how to build a bomb, and it will flatly respond that it “can’t help with that.” But users have long played a cat-and-mouse game to try to trick language models into providing forbidden information. Just as quickly as these “jailbreaks” appear, AI companies patch them by simply filtering out forbidden prompts before they ever reach the model itself.
Recently, cryptographers have shown how the defensive filters put around powerful language models can be subverted by well-studied cryptographic tools. In fact, they’ve shown how the very nature of this two-tier system — a filter that protects a powerful language model inside it — creates gaps in the defenses that can always be exploited. In this episode, Quanta executive editor Michael Moyer tells Samir Patel about the findings and implications of this new work.
Audio coda courtesy of Banana Breakdown.