> Maybe they can fix it with a prompt that makes the model check if the ingredients are safe. Chances are that every prompt can be circumvented but they can pro

> Maybe they can fix it with a prompt that makes the model check if the ingredients are safe. Chances are that every prompt can be circumvented but they can probably find instructions for self poisoning somewhere else with less effort. That seems like a job for a database, not a language model. Every problem is not a nail.