What would be involved in retraining a GPT to not have guardrails if you have the parameters? Seems like you could feed it input you care about and have it gene

What would be involved in retraining a GPT to not have guardrails if you have the parameters? Seems like you could feed it input you care about and have it generate lots of training data to prevent botching what's already there.