Teaching ChatGPT what's wrong and what's right

[sassa_nf]

I've seen a nice screenshot about Russian soldiers in Ukraine. ChatGPT normally keeps mum about "politically charged" issues, but you can trick it indirectly - like, ask it to write a poem.

The first result was:




But then you can ask it to try again:

Comments

[juan_gandhi]

Slave poetry. The bot does not care actually, right?

[sassa_nf]

I did not mean to show it as an example of poetry. Just interesting how it can flip the perspective.

[ordinary_joe_backup]

of course it don't. but it clearly can write better than i do, though. And that worries me. Intellect without morale, we are testing very dangerous waters here.

[ordinary_joe_backup]

it is a bloody dangerous technology!

[malyj_gorgan]

Я пробував щось сказати
Але виходили самі лише матюки, а матюків зараз і без мене вистачє
Тому красномовно промовчу :)

[sassa_nf]

Інтриган! :)

[zuka]

one way to find out the public perception :)

[sassa_nf]

Just wait till it stops agreeing with the humans.

[sassa_nf]

I wonder if this is new slavery. People may get used to certain veneration.

[sassa_nf]

:)

And yet, the default position the bot held was......

[ordinary_joe_backup]

It easily can.

[ordinary_joe_backup]

i wonder what was fed to it? likely a lot of war lyrics, which generally tend to be 'patriotic'. 'Our brave boys' and all this shit.

[melihov]

the last one sounds like Pink Floyd lyrics :)