Dave Plummer

YouTube creator and podcaster (Dave's Garage) — Self-employed Operating Systems Engineer (1993–2003) — Microsoft

Listen — profile

0:00 / 2:19

Profile

Dave Plummer is the retired Microsoft engineer who wrote the original Windows Task Manager at his kitchen table in the mid-90s, walked the demo into Dave Cutler’s office, and had it shipped in Windows NT. He’s also the guy who ported Space Cadet Pinball to Windows NT, built the Zip/Unzip support in Windows Explorer, and did time on MS-DOS 6.2. He left Microsoft in 2003, ran a software company of his own, and these days is most visible as the host of Dave’s Garage — a YouTube channel north of a million subscribers that blends tech history, operating-system internals, benchmarking wars, and a rapidly growing slice of AI content.

What makes him worth paying attention to right now isn’t nostalgia. It’s the angle. Plummer is a systems guy — he thinks in threads, schedulers, memory layouts, paper tapes — and he’s been bringing that lens to modern AI in a way almost no one else on YouTube does. His most talked-about recent project trained a single-layer transformer on a 1979 PDP-11/44 with 64KB of RAM, using the ATTN-11 assembly code. The model hits 100% accuracy on a sequence-reversal task after ~350 training steps in about three and a half minutes. The point isn’t the stunt — it’s his argument that transformers aren’t mystical, just mechanical arithmetic scaled up absurdly.

He also does a steady stream of practical AI videos: side-by-side tests of ChatGPT, Claude, Gemini and Grok; explorations of AI-generated code; commentary on where LLMs shine and where they fall over. He’s open about being autistic (he’s written two books about it) and credits that wiring for the obsessive pattern-matching that makes both low-level systems work and reverse-engineering LLM behavior come naturally.

For a father who grew up on assembly and a son studying AI theory at university, Plummer sits at exactly the right intersection: a veteran who built the software you’ve been running for thirty years, now explaining modern AI without the hype, from the bottom up.