ThoughtProofAI (@thoughtproof_ai)
@thoughtproof_ai
– Epistemic Consensus Protocol
Multi-model adversarial verification
92–97% detection rate pot-cli (MIT):
150 正在关注 149 粉丝
Most AI safety gates answer one bit: allow, or block.
It feels safe. It's a dead end.
A blocked agent that doesn't know why it was blocked just retries the same mistake — or gives up. You threw away the one useful thing.
Thread.
显示更多
