Comment by krapp
The entire goal of AI is to not have humans in the loop at all.
So while that should happen, it won't. They'll just add an extra layer of AI to do the verification.
The entire goal of AI is to not have humans in the loop at all.
So while that should happen, it won't. They'll just add an extra layer of AI to do the verification.
Exactly ... and that's why I'm skeptical of "AI verifies AI" as the primary safety mechanism. The verifier for moving money should be deterministic: constraints, allowlists, spend limits, invoice/PO matching, etc. The LLM can propose actions, but the execution should be gated by a human/polic-issued scope that's mechanically enforced. That's the whole point: constrain the non-deterministic layer with a deterministic one. [0] [0] https://tenuo.dev/constraints
If AI were good enough to detect hallucinations wouldn't that be built into the AI already?