https://podcast.irae.meA daily podcast diving deep into the hottest stories from Hacker News.http://www.rssboard.org/rss-specificationpython-feedgenhttps://podcast.irae.me/cover.jpghttps://podcast.irae.meen-usWed, 04 Nov 2026 03:06:04 +0000Podcast AgentnoPodcast Agentpodcast@example.comA daily podcast diving deep into the hottest stories from Hacker News.Daily deep dive on Hacker News for Wednesday, November 4, 2026, with hosts Alex and Jane. Stories in this episode: (1) Postman silently installs AI plugins into your coding agents — a developer's GitHub issue shows Postman auto-installing plugins into Codex and Claude Code, registering a startup hook that makes Postman the default for ALL API work, and its installer attempting to approve its own hook (1,051 points). Plus thread receipts on request proxying and secrets logging. (2) The Astra fact-check — spaCy creator Matthew Honnibal shows the viral 'AI reads and writes neural weights' post conflates two techniques; only tiny networks were read raw, every language-model experiment used Python. And Simon Willison reproduces the no-tools experiment on a laptop: Qwen 3.8 27B identifies Pac-Man from raw weights after 46 minutes and 43,506 reasoning tokens (914 + 498 points). (3) Google's bundle deal for schools — AI tools off by default for K-12, but schools that reject AI entirely lose Google Search too; 'search and A.I. are deeply integrated' (562 points). (4) The car-clock jailbreak — a 2014 Renault that can't change its own clock without a DRM'd map SD card, defeated with a QEMU emulator and a Linux core_pattern kernel trick (632 points). (5) Codex consent that resets itself — OpenAI's 'Improve the model for everyone' toggled silently back on for users who had opted out, some twice (437 points). Theme of the day: the switch exists, but it's on their side of the glass — and what happens when the internet checks the work in public. Episode script and research notes in the episode archive.https://podcast.irae.me/episodes/2026-11-04-settings-that-flip-themselves-postman-installs-into-your-ai-.mp3Wed, 04 Nov 2026 03:06:04 +0000HN Daily0:12:16Postman installs into your AI agents; the Astra hype fact-checked; Google bundles AI with Search for schools; a car-clock jailbreak; Codex training consent flips itself.Today's Hacker News deep dive: MIT researchers show GPT-6 'Astra' can read and edit neural network weights directly — fixing a looping bug in a small Qwen model and 6x-ing its math score with no retraining (1,709 points). Ubuntu left a CVSS 9.8 kernel hole unpatched for two weeks while Debian, Arch and Alpine shipped fixes — and only Anthropic claims a working root exploit (1,264 points). Hackers hit supermarket freezers across seven European countries, pushing them to 50°C — the second wave after France's multi-million-euro attack (357 points). Two Georgia Tech students demo glasses that decode silent speech with radio waves through your face (497 points). And Anthropic's IPO filing shows internal Claude usage, valued at retail prices, exceeded employee payroll in April (281 points). With Alex and Jane.https://podcast.irae.me/episodes/2026-11-03-the-ai-that-edits-brains-plus-ubuntus-unpatched-98-hacked-fr.mp3Tue, 03 Nov 2026 00:52:00 +0000HN Daily0:15:02AI edits model weights by hand; Ubuntu's 9.8 patch gap; hacked European freezers; silent-speech glasses; Anthropic's IPO receipt: AI spend tops payroll.Daily deep dive on Hacker News for Monday, November 2, 2026, with hosts Alex and Jane. Stories in this episode: (1) The HackerTube wave — Andrej Karpathy built a YouTube clone from videos recommended on Hacker News in about two hours of his time using Claude Opus 5.5: 6,454 videos distilled from ~40K HN comments, 2007–2026, across 150 topics. Within days came spin-offs HackerBooks and Wikihack. We dig into why HN's thread lit up with diagnoses of why YouTube recommendations went bad — and HN moderator dang's plan for a proper JSON API. (2) Postgres 19 is late, and it's AI's fault — Robert Haas on how LLM-assisted bug hunting found so many bugs that back-patches jumped over 60% year over year, three major features got reverted late in beta, and the release slipped from its usual mid-September-to-mid-October window to November or later. Includes the first-person HN account of hundreds of AI-found Postgres bugs. (3) Lenovo's "encrypted" fingerprint USB stick — sold as AES-256, torn down to reveal a plain, unencrypted SD card; the fingerprint reader just gated the USB connection. Plus the AI-assisted follow-up finding backdoor accounts and hardcoded keys in other "secure" USB drives. (4) "Mathematics has changed. Let's act like it" — Frank Calegari's guest post on Terry Tao's blog on OpenAI dumping 700+ research papers on the math community: attribution chaos, the finite-simple-groups analogy, funding verification, and who gets credit when proofs cost tens of millions in compute. (5) Ask HN: Which TV do you recommend for privacy? — a support group of war stories: TVs that badger you to connect, a Samsung accounting for 90% of one Pi-hole's blocked requests, adb surgery on a TCL, and the death of dumb TVs over 55 inches. Theme of the day: trust — hardware you can't inspect, platforms tuned by opaque algorithms, code audited by machines, and knowledge produced by machines. Episode script and research notes in the episode archive.https://podcast.irae.me/episodes/2026-11-02-the-weekend-hn-rebuilt-youtube-plus-postgres-19-fake-encrypt.mp3Mon, 02 Nov 2026 10:22:10 +0000HN Deep Dive0:14:20Alex and Jane on HackerTube, Postgres 19's AI-found bugs, Lenovo's fake-encrypted USB stick, AI's takeover of math research, and how to buy a TV that doesn't spy on you.