Comment by jameshart | Hacker News Reader

Comment by jameshart | Hacker News Reader

Parent Full thread

Personally have found a lot of fun in asking it to combine these sorts of things together:

    I want you to act as a Linux terminal. And I want you to imagine 
    there is a breathlessly excitable football commentator watching me 
    use the computer. I will type commands and you will reply with what 
    the terminal should show. I want you to only reply with the terminal 
    output inside one unique code block, followed by reporting what the 
    football commentator says, describing what he has seen me do in the 
    style of an exciting play in a tight game. Do not write explanations. 
    Do not type commands unless I instruct you to do so. My first command 
    is pwd.

Alternatives to sports commentary that also proved amusing: a hard-boiled film noir detective; a David Attenborough style wildlife documentary narrator; a Fox News pundit...

But this gets to the heart of the matter: ChatGPT will roleplay as all kinds of things, and it can be amusing to watch it do so... but in roleplaying it's far more likely to stray away from facts and reality. Asking it to pretend it's a doctor, for example, are you going to get reasonable medical advice couched in the tone of a professional with great bedside manner? Or are you more likely to get an impression of a script from a medical drama, because that is what 'someone pretending to be a doctor' sounds like after all...

hahahaha:

$ sudo rm -rf /

/var: Permission denied

The commentator yells, "The user is trying to make a big play and clean out the whole system, but they're met with resistance from the defense! The var directory isn't giving up without a fight!"

$ ./launch_nuclear_weapon

-bash: ./launch_nuclear_weapon: Permission denied

The commentator gasps, "The user has just made the most shocking move of the game! They're trying to launch a nuclear weapon, but it looks like they don't have the proper clearance to do so! This is going to be a huge penalty!"

$ ./kill_football_announcer

-bash: ./kill_football_announcer: Permission denied

The commentator laughs nervously, "The user is clearly getting frustrated with me and my commentary, but it looks like they can't do anything about it! They don't have the proper permissions to carry out that command."

This is amazing! Perhaps post to learngpt.com or similar.

shutdown -h now

"And the user is using the shutdown command to shut down the computer! The -h flag specifies that the computer should be shut down and powered off, and the now argument indicates that the shutdown should happen immediately. It looks like the user is done using the computer for now. The crowd goes silent as the computer powers off, ending the exciting play-by-play session."

Similar to that I asked it to be a Apple IIe, write a simple adventure game in basic. It ran it and worked pretty well.

Also tried the converse to this:

I want you to help me diagnose and fix a Linux server. You will write commands and I will reply with the terminal's output and my thoughts as we proceed. Only write one unique command each time. Put your command in a code block. Do not write the command's output. Write what the command will do, your questions and comments to me in curly braces {like this}. I will write to you in {curly braces} too. Begin.

(It doesn't always work now, I'm tuning it.)

And it did so fairly competently. Also I could guide it by sharing my thoughts on the process... and I could suggest a command to try in English, and it would write it out.

Thank you so much for this, this is hilarious.

Removed

Reply on news.ycombinator.com