2026-10-18 · America/Los_Angeles · 社区动态 · #20

My AI verification tool missed a dangerous medication order. I’m looking for developers to break it.

I’m building an open-source MCP for independently testing AI-generated outputs, starting with synthetic healthcare scenarios. During validation, it missed an acetaminophen order totaling 9,000 mg/day. The verification process failed to account for the frequency but still returned a reassuring result. That’s exactly the kind of failure I’m trying to catch. I’m fixing the issues now, but I’m also trying to understand how other developers handle confident answers that aren’t actually supported by evidence. I’d like to find 5 developers willing to test the MCP when the fixes are ready and try to…

热度 44.5 / 100；排名与评分保留该期记录。