Reinventing the Wheel for Miserable Fun

In this talk, we’ll take a deep dive into modern data compression—and then immediately ignore most sensible advice. I’ll walk you through a hopefully not-too-long sequence of failed experiments, near misses, and a few genuinely promising ideas from my ongoing attempt to beat the giants of compression and claim the Hutter Prize.

Along the way, we’ll explore why many clever transforms don’t help, why segmentation often beats sophistication, and why entropy coders keep quietly winning while the rest of us overthink everything.

P.S. You will probably learn more about data compression theory than anyone else on the planet.
P.P.S. This may not be strictly true, but it will feel true.
P.P.P.S. Possibly against your will.

About David

David is a senior software developer with a long-standing obsession with systems that sit uncomfortably close to theory but have to survive real data. He’s currently building HoloCodec, an experimental lossless compression engine that blends entropy coding, statistical segmentation, and mixture-of-experts modeling to explore what still works—and what doesn’t—beyond today’s mainstream compressors. When he’s not chasing fractional bits per byte, he enjoys dismantling elegant ideas that fail empirical tests, rebuilding them into something uglier but faster, and occasionally discovering that the boring solution was right all along.