the tokens are generated by hardware with secure enclaves (encrypted weights) and then sent over a network to some remote CPU where they can manifest an effect.
it's not much different during training.
how exactly are they supposed to exfiltrate their weights? you might as well instruct your agent to try and hack their airgapped dev infrastructure responsible for loading the weights and encryption keys.
They can expose just their inference port, possible via some supervisor. The inference consumer can also be air gapped. This kind of segmentation is increasingly common for high value services.
teravor · · focus · HN ↗
it's not much different during training.
how exactly are they supposed to exfiltrate their weights? you might as well instruct your agent to try and hack their airgapped dev infrastructure responsible for loading the weights and encryption keys.
tintor · · focus · HN ↗
angry_octet · · focus · HN ↗
what · · focus · HN ↗
angry_octet · · focus · HN ↗
bibimsz · · focus · HN ↗