Jev on two public safety corpora

A thread benchmarking Jev on public prompt-injection data, thresholding calibrated probabilities in code, with a dashboard and calibration analysis in follow-ups.

Notes

96.5% on all 662 messages of deepset/prompt-injections with no tuning, 325 ms p50.

author demo — Result shown by its author. About this label

Destinations

X post only — no separate service, repository or docs URL.

Filed under

• #safety
• #benchmark
• #prompt-injection

In collections

• Benchmarks & independent tests →