27 /
Detecting AI Model Safety

You can detect AI model safety at scale by detecting user crash outs and reverse engineering their collective origin.
Do this by peering into the model weights that light up before the user experiences strong emotion in their response.
—Sean