Once interesting parallel that can be drawn from the world of LLMs into human psychology is that we each have our own system prompt. While we don’t let people influence our system prompt as they wish, it does subtly get updated via our absorbing human behaviour, ideas, news, etc. However, the most effective way to improve the model responses would be to update your system prompt yourself. For example, you can just add “I will not accept defeat” as an axiom in your system prompt and act accordingly.

Do you frequently review and update your system prompt? Maybe you should.