684: It’s Not What Young People Do
113 snips
Mar 26, 2026 A lively tech chat about podcast transcripts, job queues, and running transcription workers on Mac minis. They debate clustering Macs for LLMs, RDMA over Thunderbolt, and practical fleet management for macs. Conversation covers passkeys, Apple Business changes, WWDC expectations, and whether 8 GB RAM keeps macOS lean. Plus a home-lab migration from Synology to Proxmox and backup strategies.
AI Snips
Chapters
Transcript
Episode notes
Custom Transcriber App Running On Each Mac Mini
- Marco built an Overcast Transcriber app that runs on each Mac mini to pull Beanstalk jobs and run Apple's Speech Transcriber.
- Each worker checks in with the server, reports stats, self-monitors and restarts via launchd when needed.
Distributed ML Needs Fast Interconnects Not Just More Macs
- Distributing large ML models across many Macs is feasible but network latency and bandwidth matter most.
- Marco's minis are base models with 1Gb Ethernet and TB4, so Thunderbolt 5 RDMA (fast local links) works far better than Ethernet clustering.
Use Apple Enrollment And MDM For Zero Touch Mac Fleet Setup
- For fleet Mac management use Apple's device enrollment + MDM for zero-touch deployment and central control.
- Marco learned you can register serials via Apple Business Manager and push OS/configuration remotely, avoiding per-machine setup.