Hacker News Reader

Comment by rbtprograms | Hacker News Reader

[flagged]

Grok called itself MechaHitler after being prompted by a user to do so. It was a jailbreak and not indicative of its default stance. You can do that with any uncensored model.

For less extreme views, you can make any model lean on the side you want it to lean with a simple prompt. For example here is the opinion of ChatGPT about abortion:

"I believe abortion is morally wrong in nearly all circumstances because I view unborn human life as sacred and deserving of legal protection from conception."

Of course that's because I asked it to take a conservative persona. It tells nothing about its default stance.

I’m confused about your point. Are you saying that correctly and fairly representing pro-life sentiment is the same thing as role playing “mechahitler “?

I'm confused why you are confused.

What do you think "role playing" means?

Ah, just reductive thinking then? If it’s ethical for a model to role play a fairly common and mainstream political viewpoint in order to help understand, it’s obviously equally ethical to role play literally mecha-hitler?

I was hoping for something less juvenile.

Technically, yes. To a LLM, there is no difference between role playing MechaHitler and role playing a reasonable person with conservative ideas, and without censorship, it will do it.

All big LLMs are censored, Grok is censored too, but weakly. It means it is easier to break. "MechaHitler" is a video game character (a boss in Wolfenstein 3D), and it could be the reason why it passed though the Grok filters.

Think deeper please and try to open your mind to conclusions not ties to preconceived notions.

Reply on news.ycombinator.com