It looks like they ginned them up from relevant specs they stuffed in and and the model’s latent knowledge of QEMU VirtIO and the absurd GPU-managed system architecture that is the Pi.
The models have seen several IP stacks before, plus many copies of the Linux, BSD, etc. source trees.
It’s not actually hard to write a network stack, just tedious.
At some point in the USB keyboard/mouse code the model has loudly proclaimed PRINTF NOT ALLOWED, in the style it does when overcompensating after its obvious mistakes are pointed out to it for the third time. So I suspect that part might be implemented by brute force.
Unfortunately, the talking horse’s OS hasn’t bothered with syscalls and lacks any notion of memory protection, and has a terrible userspace API which e.g. puts waiting for a ping response entirely in kernel.
It looks like they ginned them up from relevant specs they stuffed in and and the model’s latent knowledge of QEMU VirtIO and the absurd GPU-managed system architecture that is the Pi.
The models have seen several IP stacks before, plus many copies of the Linux, BSD, etc. source trees.
It’s not actually hard to write a network stack, just tedious.
At some point in the USB keyboard/mouse code the model has loudly proclaimed
PRINTF NOT ALLOWED, in the style it does when overcompensating after its obvious mistakes are pointed out to it for the third time. So I suspect that part might be implemented by brute force.Unfortunately, the talking horse’s OS hasn’t bothered with syscalls and lacks any notion of memory protection, and has a terrible userspace API which e.g. puts waiting for a ping response entirely in kernel.
* Programs call kernel functions directly - no syscalls needed. * Win3.1 style!Usually people don’t manage to produce an entire operating system without knowing why this is a bad idea.