All Episodes
Page 4 of 320 · November 2026
#6464: Teaching Speech Recognizers to Stay Quiet
Masking just three self-attention heads in Whisper large-v3 cuts hallucinations on non-speech audio by over eighty percent.
#6463: Why Training Speech Recognition From Scratch Fails
Thirty-two hours of speech provides just three hundred thousand words of text, making from-scratch speech models struggle with basic spelling.
#6462: When Adding a Reranker Hurts Search
NVIDIA benchmarked a cross-encoder upgrade that scored three points below its baseline, while other models surged fifteen.
#6461: Why Easy-to-Read Fonts Must Break Harmony
Traditional type design relies on uniform shapes, but Atkinson Hyperlegible breaks visual harmony so low-vision readers can tell letters apart.
#6460: Why Watching a Sleeping Child Defeats Attention
A study showed trained lifeguards could not complete the standard ten-second visual scan even when explicitly instructed.
#6459: Why the Gaussian Distribution Fails in Plain Sight
A single trading day accounted for 78 percent of the kurtosis across ten thousand US stock market observations.
#6458: Why Synthetic Fine-Tuning Data Needs Tooling
Rewriting high-quality data synthetically gives a 3.4-point gain, but running the same process on low-quality data yields just 0.5.
#6457: Where to Learn LLM Training Without Generic Courses
A GPT-2 scale model that took OpenAI 168 hours to train in 2019 can now be replicated on a leaderboard in under two hours.
#6456: The Unsung Builders of Modern Speech AI
OpenAI's Whisper lists famous co-founders among its six authors, yet shipped with no launch event, demo day, or canonical interview.
#6455: Turning Open-Source Triage into Subscription Credit
No software vendor currently applies peer-reviewed contribution credits directly against a customer's subscription renewal invoice.
#6454: Choosing Between MCP Servers, Agent Skills, and CLIs
A controlled Arize evaluation found that bare Claude without an MCP or skill scored higher on correctness than curated tool stacks.
#6453: Who Actually Won the Open Weight Race
With over four billion downloads, the most downloaded open-weight model in history is an embedding tool, not a chatbot.
#6452: Why Recursive File Watching Strains Linux Memory
A recursive watch across a million directories locks up roughly a gigabyte of pinned kernel memory that cannot be swapped.
#6451: Which AI Models Actually Last
OpenAI kept the gpt-3.5-turbo alias alive for over three years, while Google Gemini Flash iterations often turn over in just a few months.
#6450: Securing MCP Endpoints and Webhooks Without Open Doors
Over half of all internet requests at one edge endpoint in a single day came from automated scanners probing for secrets.
#6449: Why PHP2Shell Was a Sniper Rifle, Not a Nuke
A decade-old PHP flaw affected 46 million servers, but attackers bypassed broad waves to cherry-pick narrow target builds.
#6448: Why Fast Model Microtraining Needs an Evaluation Gate
A speech model fine-tuning run can cost twenty-three cents in twenty minutes, but unmonitored updates silently erase older vocabulary.
#6447: Why Uptime Math and Retry Storms Break Cloud Services
Between April and August 2026, GitHub's monthly commits doubled from 1.4 billion to 2.9 billion, exposing massive architectural coupling.
#6446: What Home Fiber Optic Cables Actually Do to Your Eyes
A typical home fiber terminal outputs under 4 milliwatts, far below the 166-milliwatt international safety limit for single-mode fiber.
#6445: Why That In-Wall Fiber Link Refuses to Turn On
On a ten-meter fiber run, a single mated connector pair can cause twenty times more signal loss than the entire length of glass.