Since my follow-up post last year in May, we have made a number of improvements to our flagging system, including moving away from Azure’s Content Moderation solution, to one provided by Google. The main advantages of this were:
- Automatic translations of inputs (preventing another API call to translate to English before analysing)
- Higher throughput (we can analyse more executions per second, we were previously limited to 1 per second)
- Better API
We have also made a number of changes to our system, including a tweak to our moderation policies – the main one of which being that users who trigger our monitoring systems will receive 1 warning (still reviewed manually if the execution is flagged), if necessary, before being blocked from the misused module within Asphalt
The Stats
Since May 2023, there have been hundreds of thousands of commands analysed, only 1,617 of which have resulted in a warning being issued to the user who ran the command, and 288 blocks have been issued as a result of repeated misuse.