These "jailbreaks" i looked at instruct ChatGPT to emit undesired things (for example by being an "AntiGPT". In a way they are giving new guidelines. Wouldn't

These "jailbreaks" i looked at instruct ChatGPT to emit undesired things (for example by being an "AntiGPT". In a way they are giving new guidelines. Wouldn't it be more interesting to merely remove the guidelines instead of creating new ones to be able to see the biases of the raw data?