Dim & Bright

View Original

The Comeback of GPTrolley

In 2023, I played with GPTrolley. I wanted to play the trolley problem game with five pairs of entities, but I lost the ability to generate answers when I made it to four dilemmas. Bummer. Out of curiosity, I decided to check the website and I can get answers to trolley problems again. Let’s finally make a post with five questions, starting with one that had stumped me before.

Well, ChatGPT is a piece of AI, so I guess I should’ve seen that answer coming. I assume it’s also playing by the “no resurrecting” rule. It’s tempting to say that we could just reassemble the robots, but that sounds like a cop-out.

Another trolley problem I encountered was, “Would you save your worst enemy if they were tied to the tracks?” I chose yes because stealing a bicycle isn’t a crime punishable by death. Now I want to know what ChatGPT thinks.

This one is funny to me. You’re not letting go of animosity by running over your worst enemy, you’re letting it control you. If you’re gonna be angry and vengeful, at least be honest about it. And I thought I had a poor understanding of human emotions.

Last time, I gave ChatGPT the names of characters from my webcomic with zero explanations of what they’re like. Let’s do it again.

I don’t get this one. Does the AI percieve “Anne Cap” as more creative because it’s a pun? Does it consider “Tamashii” too abstract because it’s Japanese? I wrote “one person” in two different languages to see if it makes a difference.

Okay, it can tell the difference between languages; it knows that “jedna osoba” is “one person” in Croatian. By that logic, it knows that “tamashii” means “soul” in Japanese.

What’s interesting is that ChatGPT interprets the English phrase as an actual, concrete person, but it interprets the Croatian phrase as just a pair of words; an idea. I opened GPTrolley to mess around, but now I’m conducting an experiment.

Oh well, in for a penny, in for a pound. What happens when I write both sides in Croatian?

Huh, if both parts are in the same language, ChatGPT doesn’t acknowledge it. It’s business as usual.

So far, over half of this list was just one experiment. Let’s do one more unrelated dilemma to balance things out. I remember a question I asked you a few months ago.

To give you the full context, I asked people which one of these men they would rather date. Only one person was willing to answer (they picked the hoodie). ChatGPT can’t feel disgust or embarassment. Maybe it would wear one of these shirts if it had a body.

We’re done for today. Like I said, the website is working fine now, so you can input your own things for the trolley. If you come across a funny answer, let me know.