I agree with the scariness etc. Just one possibly comforting point. I assume (hope?) they use more traditional classifiers for determining importance (in addit

I agree with the scariness etc. Just one possibly comforting point. I assume (hope?) they use more traditional classifiers for determining importance (in addition to the model's judgment). Those are much more reliable than LLMs & they're much cheaper to run so I assume they run many of them