That's a major breakthrough. I just hope it can scale at some point, though, because right now they're dealing with only very small models (well that is, in the LLM space).

Especially exciting to me is the fact that it may make the models be _auditable_. That would be a game changer.