It’s frustrating how many of these systems rely on hard coded word or even substring matching, and that in a world of large language models that can evaluate semantics.
Honestly it’s a blessing, hardcoded stuff can be easily circumvented by turning to 1337-speech or inserting whitespace. An AI scanning stuff would not only produce random false positives but also make the censorship actually work
It’s frustrating how many of these systems rely on hard coded word or even substring matching, and that in a world of large language models that can evaluate semantics.
Honestly it’s a blessing, hardcoded stuff can be easily circumvented by turning to 1337-speech or inserting whitespace. An AI scanning stuff would not only produce random false positives but also make the censorship actually work
The good old scunthorpe problem