It matters if people figure out how to make your model output inappropriate things and start sharing screenshots on twitter. What if you use a GPT model as part

It matters if people figure out how to make your model output inappropriate things and start sharing screenshots on twitter. What if you use a GPT model as part of a chat bot? Then you definitely don't want to risk any sexual or discriminatory output for example. If that was a person talking, they would be fired.