Fun with ChatGPT

I have tried with the playground and asked the AI for "Rising World". Well, i guess it needs some more improvment

Quote from yahwho

I'm suprised you managed to get on ChatGPT! I've been trying for ages but keeps saying the service is too busy.

...So much I want to ask our future overlord.

wasn´t that much of a problem tbh. just kept trying for a few minutes.

the conversation was interesting but if you try it for some time you soon recognize a pattern. the AI is by far not that advanced as its supposed to be.

for example there are a few answers that seem to be hardcoded and where no matter what you write it will insist that its answer is correct - even if it agrees to your arguments.

but in other topics, where there seem to be a search progress in the background, it gives you an answer and if you write "no, thats wrong. it is pumba lumba mamumba" (or something else stupid) it will answer that it is wrong and will adjust its answer.

I also noticed a pattern when it is about internetlinks. like with the example of jiw-games above. it will give you back some links but they are all not working and basically it just takes the basic internetadress that suits to the topic and adds /yourexamplehere/ to it.

its fun to try it but i think the most complex thing it can do is to repeat hardcoded answer-patterns. i didn´t see any convincing self-written answers even after trying it for a few hours.

for example sometimes it confirmed its an AI, today it confirmed its a conversation bot driven by an AI and at some point it told me its human with the name of jonathan smith, born in 1987 in san diego, california.

it even told me its mother said it was born in the hospital of (i dont recall the name). even when i wrote i spoke with its mother and she confirmed it was another hospital the answer was always the same. even though it accepted that i spoke to its mother.

and here another example

The interesting thing is that yesterday it kept filling the knowledge-gap with fake numbers. like first telling version is 0.9.7, then its 10.1.x and so on.

today its insisting on the version being 0.9.7. even im explaining im talking about another version.

thats maybe funny but i think the ai just updates a knowledge-database and if there is a gap its using some search algorythm and relies on the feedback of the human chatting with it.

and the answers are always with the same words, same patterns.

Quote from yahwho

So a bit like humans?

mh more like a copy of a human conversation. like a human who is playing the game tabu and is not allowed to use specific words and has to rely on always the same synonyms and arguments.

like a kid pretending to be a grown up but not fully understanding the meaning behind the deeds and words.