Running AI on a CPU: Because Waiting 14 Minutes Builds Character - (Episode 173)

Running AI on a CPU: Because Waiting 14 Minutes Builds Character - (Episode 173)

Today’s episode is dedicated to one of the greatest technological achievements of the modern era: running an AI model on a CPU because GPUs now cost approximately the same as a midsize apartment in Zagreb. We discuss the breathtaking experience of waiting 14 minutes for a response that confidently explains facts that never existed, cites imaginary research papers, and occasionally forgets what year it is. Naturally, this leads us into the philosophical debate of whether an AI that takes half an hour to answer is actually “thinking” or just emotionally processing its own poor life choices. From thermal throttling and swap-file abuse to hallucinations so convincing they deserve political office, this episode is a beautiful monument to patience, bad decisions, and the unstoppable human urge to run enterprise AI on hardware rescued from a student lab.

The datacenter will be ready right after the roundabout - (Episode 172)

Welcome to The datacenter will be ready right after the roundabout, the only podcast brave enough to ask the important questions: can a hyperscale datacenter survive Croatian paperwork, three ministries, two environmental studies, and a mayor who still thinks “cloud” means rain? In this episode, we dive into the glorious collision of AI ambitions, megawatt fantasies, land permits, power grid realities, and the sacred regional tradition of discussing infrastructure projects for roughly twelve years before pouring a single cubic meter of concrete. We talk cooling, fiber, geopolitics, NIMBYism, diesel generators the size of apartment buildings, and the magical belief that “digital transformation” somehow works without electricity. The future is here — just as soon as somebody finishes the access road.

When your home lab has a better disaster recovery plan than your career plan - (Episode 171)

Somewhere along the way, “learning virtualization” turned into running a miniature Fortune 500 datacenter next to the washing machine. In this episode, we dive head-first into the glorious madness of over-engineered home labs: redundant power supplies for one user, Kubernetes clusters hosting absolutely nothing important, backup strategies more detailed than our retirement plans, and the eternal justification that “it’s for learning.” We talk about the slippery slope from a single Raspberry Pi to racks full of servers screaming through the night, why every homelabber eventually discovers VLANs at 2 AM, and how disaster recovery suddenly becomes deeply personal when Plex goes offline. If you’ve ever convinced yourself that a 100-gigabit upgrade was “necessary,” this episode may feel uncomfortably familiar.

The one where we discuss podcast Saturation: Now available in audio/video form - (Episode 170)

At some point, podcasts stopped being special and became background radiation. Every celebrity has one, every startup founder has three, and somehow every conversation now needs microphones, RGB lighting, and a “don’t forget to like and subscribe.” So naturally, we decided to contribute to the problem. In this episode, we spiral through podcast fatigue, algorithm-driven content sludge, endless “thought leaders,” and the strange realization that we’ve started losing interest in the very thing we once loved listening to. This is not an expert discussion. It’s more like a support group for people emotionally exhausted by content.

The one where Linux and AI were ready but Jasmin was not - (Episode 169)

Welcome to The One Where Linux and AI Were Ready, but Jasmin Was Not — a title that, honestly, wrote itself. The plan was simple: brand-new shiny Linux machine, proper setup, and finally an episode recorded exactly as promised. Linux was ready. AI was ready. The hardware was ready. Confidence was also very ready. And yet, somehow, the actual recording part remained a distant dream. So naturally, we turned that small technical betrayal into comedy and used it as the perfect launch point for a broader conversation about AI, hype, tools, promises, and the timeless truth that even the smartest technology in the world still depends on humans not forgetting the one thing they were supposed to do.