Fabrice Bellard's example using gpt-2's prediction to compress english text: https://bellard.org/textsynth/sms.html https://bellard.org/textsynth/sms.html Mayb

Fabrice Bellard's example using gpt-2's prediction to compress english text: https://bellard.org/textsynth/sms.html https://bellard.org/textsynth/sms.html Maybe just run the text through something like that and use the compression ratio to measure nonsense.