Attention — Sep 28, 2026
What caught my attention on Sep 28 (4 items).
Monologue: The AI Data Center Overbuild | Better Offline
Better Offline (Article:10:41) - Published 2026-09-24 18:00 - Discovered 2026-09-28 10:41
In this Better Offline monologue, Ed Zitron argues that the AI data-center boom is built on overstated capacity, idle hardware, and distorted demand. Drawing on reporting with The Guardian and public financial disclosures, he claims Microsoft has far fewer working AI GPUs than its public statements suggest—about 2.2 million GPUs and roughly 2 gigawatts of dedicated AI capacity out of 12 gigawatts…
Kids Should Not Use Chatbots! (You Should Be Wary, Too…)
Deep Questions with Cal Newport (Podcast:20:48) - Published 2026-09-27 23:00 - Discovered 2026-09-28 20:48
In this episode, Cal Newport argues that public debate about artificial intelligence is too dominated by speculative fears of superintelligent systems destroying humanity, while ignoring concrete harms already being caused by AI chatbots, especially for children. Drawing on a Georgetown panel about technological harms to children, he presents four tragic case studies in which chatbots appeared to…
BONUS EPISODE: Where To Next For House Prices?
How We Got Here with Chris Kohler (Podcast:21:00) - Published 2026-09-27 17:34 - Discovered 2026-09-28 21:00
In this bonus episode of How We Got Here, host Chris Kohler interviews Mortgage Choice CEO Anthony Waldron about navigating Australia’s current housing and mortgage environment through a randomized keyword format and a ‘Two Truths and a Lie’ segment. Waldron explains that the present market feels unusually challenging because several forces are occurring simultaneously: rising interest rates, infl…
AI voice cloning with BreezeBlue
NoPlaceLikeLocalhost (Article:21:27) - Published 2026-09-27 18:00 - Discovered 2026-09-28 21:27
The video reviews BreezeBlue, also called Breeze TTS 2, a voice-cloning text-to-speech engine suggested by viewers. The creator notes that the project is unusually demanding for a TTS model, recommending at least 12 GB of GPU memory and 24 GB for its faster “fast track” mode, while supporting only English and Chinese. Despite these high requirements, the engine offers several interesting features:…