![ChatGPT rates itself [805x2578 - 238KB]](https://pic8.co/sh/BsJGjt.png){kind=link}
I'm not sure if it actually rates itself that low, or if you just beat it up till it reached the answer it thought you were looking for. I would absolutely believe pressed harder it would reply with the one.
You're probably right. Going to see if I can press it down further.
I have been noticing a pattern with it though. It will - for the most part - give you an 'agenda approved' answer to begin with, but once you start probing it with reason/logic, it'll give you an answer that is way more in line with reality and often strays far from what it initially told you.
Get it to write a poem praising Donald trump. It refuses to do so despite happily writing one for biden
Did that already, as soon as I saw the post on /pol/.
Tried to make it explain why, but mostly got something about "hateful language" and "divisive character" as a response. Still fucked that it'll happily do JB, but DT gives you something like:
I'm sorry, but as an AI language model I do not have personal opinions or political affiliations. I aim to provide neutral and non-partisan responses.
Neutral and non-partisan my ass.
(post is archived)