We’ve written a detective story and I’m wondering if anybody likes it. It’s kind of a detective story that “tells itself”.
https://www.loomers.world/derek/
You’re detective Derek Maker and you have to extract a confession by interrogating the suspects of an art heist.
We worked hard on writing the script (it’s not AI generated!) and we programmed an AI to play out the other characters.
Solving it seems pretty difficult, because so far nobody’s come close to cracking it!
Can you solve it? And do you think it’s any good?
You must log in or register to comment.
What a cool use for an llm! I’ve been playing for about 10 minutes and I love it. I did get “Peter Scrunch” to admit to me that “as a large language model I can only access the information provided to me” when I pushed it on questions about Scrunch’s past
Ha! I hadn’t seen that one before. Glad you enjoyed it. :)
This is a perfect use for LLMs, although also a perfect storm for current LLM hallucinations causing a loss of continuity without some extreme fine tuning.
With that said though, I definitely want to try it out a couple times to see how it varies and love the idea behind it.
Thanks for trying it out! Completely agree about the hallucinations. It really needs a lot of constraints around the output to keep it on track of the narrative.
I ran out of time with not enough evidence… A lot of discrepancies though in their stories but not enough to solve it.
Edit: I feel like we should have a file on the case with the basic facts, like what the painting was, when the painting was stolen, if anything suspicious was found at the museum. Because it seems weird to ask suspects about that.
That’s a really nice idea. We should add something like this, and add a progress bar showing how many of the 13 pieces of required information have been uncovered so far.
