Wednesday, 11 November 2026
13 episodes · 6596 in all
#6537: Expecting Gidi, Act Three: Twenty-Eight Days
A candidate receives sixty-seven recommendations for prime minister, but assembling sixty-one coalition seats requires impossible compromises.
#6536: Why Speech Models Need More than Fifty Hours
A study fine-tuning Whisper on Dutch found word error rate increased after one hour before dropping steadily across fifty hours.
#6535: Who Makes AI Accessible to Everyone?
US software stocks shed roughly two trillion dollars in market value over fears that agentic AI dismantles per-seat pricing models.
#6534: Why People Pay to Drink Terrible Spirits
Jeppson's Malört shot sales jumped from 400,000 to nearly eight million annually despite sommeliers comparing it to pencil shavings.
#6533: How Economics Broke and Rebuilt Israeli Wine
A nineteenth-century Bordeaux baron funded Israel's modern wine industry, installing its first phone inside a cellar wall.
#6532: Train Your Wine Palate with Side-by-Side Tasting
Tasting two wines side-by-side with only one difference between them is the fastest way to train a palate without memorizing jargon.
#6531: Who's Who in Israeli Wine, Negev Included
Modern Israeli winemaking is reviving Byzantine-era desert terraces by using precision drip irrigation and high-elevation night chills.
#6530: Designing Information for Human Brains
Hospital staff spend thirty minutes each week giving directions to lost visitors, mostly to find basic destinations like restrooms and cafeterias.
#6529: How Typesetting Shifted From Metal to Software
Between 1985 and 1995, roughly 4,000 North American trade typesetting firms dissolved as software put page composition directly into designers' hands.
#6528: Who Actually Designs the Letters We Read?
A century of legibility studies found no reliable difference in reading performance between serif and sans serif typefaces.
#6527: Why Prescribed Stimulant Withdrawal Stays Unstudied
While SSRI discontinuation gets rigorous meta-analyses, therapeutic stimulant withdrawal rests on a single twenty-two-person study from 2012.
#6526: Why wav2vec 2.0 Still Dominates Speech Downloads
Despite losing the accuracy leaderboard to models like Whisper and Canary, wav2vec 2.0 base pulls 1.8 million monthly downloads.
#6525: Why Speech Recognition Models Struggle With Accents
A longitudinal speech-to-text audit found minority-accent error rates worsened by up to 65% absolute between model generations.