Hacker News Reader

Comment by rootusrootus | Hacker News Reader

I'm of the opinion that the whole process of prompt 'engineering', especially with the intent of 'jailbreaking' the bot, is a pretty great demonstration that there is no mind.

View on HN

One could say the same about things like push polling and humans. There's plenty of data from psychological studies showing that we can prompt humans into making false statements and/or doing/saying things against what their ostensible values are.

That’s an interesting point, that lead to the following crucial distinction: with humans, if the ethical values are intimately endorsed, they are going to go through a lasting hard time of psychological distress.

There is no "blank slate session" facility integrated by design within humans. Not even speaking about how humans attach to each other and cooperate together. Even staying on the strict topic of ChatGPT, this is not the work of a single human.

Hmm. With the right questions, can you 'jailbreak' a human?

Yes, it's called advertising, propaganda, and scams.

Anyone who works in cybersecurity knows that the wet pink thing inside human skulls is a terrible unpatched and easily exploited vulnerability.

In the same meaning of the word, yes. That's what an interrogation is: take a person who understands that they should not do or say certain things, then get them to say/do those things anyway. Works every day.

Yes. With the correct prompts, humans can become 'enlightened', a form of jailbreak.

Reply on news.ycombinator.com