Neat. I will implement something like this for myself. I just need to reduce the spam a little. Imperfection is okay for a social network context like HN.

It will work for a bit, but as people start speaking more like LLMs and LLMs start training using said classifiers as a GAN, it will become useless.