Hacker News Reader

The model was sharded into ~800MB pieces over devices with 4GB RAM. Typical device storage is 16-32GB NAND. Throughput was 0.39tps.

Compromise enough devices (think thousands of off-brand TVs), and you have plenty of redundancy.

Using a tired analogy, 9 pregnant women won't deliver a baby in a month.

I fail to see the relevance. The researchers already demonstrated such sharding. https://www.anthropic.com/research/self-replicating-ai

> Using a tired analogy, 9 pregnant women won't deliver a baby in a month.

That analogy only applies to tasks with similar characteristics. Is running a computer program the same as gestating a baby?

Also I don't think this a case where they distributed the task to speed it up, but rather their goal was to compromise speed to make it runnable.