[TomaszZielinski]

Since this is HN, let me share me an idea I had one day.

Psychology/CBT says that we have so called "core beliefs"--i.e. convictions we hold, that we're not really aware of as they are embedded deep in our minds.

For instance, if one of our core beliefs is "the world is a bad place and people are hostile", we can by default perceive other people as potential threats and feel anxious/afraid/endangered.

Now, ChatGPT has the system prompt "I'm ChatGTP, ...." that's hidden from us.

IMVHO that system prompt is like a core belief. Or vice versa--our core beliefs are like ChatGPT's system prompts.

And CBT can be thought as sandbox escape/prompt injection. If we find a way to override the system prompt, we can effectively change it!

Of course it's just a shallow analogy, but maybe someone will find it useful!