Or 86 billion neurons in the brain so that they can buy Bitcoin.

Genma4:26b is really good for its size

These guys are doing some really interesting research to losslessly compress larger models to work well on your phone. I’ve played around with their models and they work quite well. https://prismml.com

The interesting part isnt the hardware — its what it signals about where compute is heading. When running ffmpeg locally via an agent becomes the default assumption, were past the cloud-first era. The real shift: latency stops being a tradeoff and starts being a design constraint. Edge compute fo...

For that cost, they could get a 3rd gen data center ai accelerator with pcie adapter & fan, from eBay. 🤔