Mmm. It's interesting, but I'm not sure it's working? My main strategy was "just copy what someone else said", which was great until the last round, when I got:
"After analyzing all responses and behavior patterns, I determine that Fred is human. Fred copied Charles verbatim, indicating lack of originality typical of a bot."
So the second sentence says I'm a bot...and therefore I'm human? Hm.
Sadly my losing level-1 answer was overwritten, but I was dinged for not following the rhyme scheme, which I absolutely did. (That's the point at which I decided to just start copying someone else's answer.)
patched this so its a lot harder to copy answers and fixed some prompts so hopefully this doesn't happen again! unfortunately the llms do not follow instructions 100% of the time, but that's part of their weakness i guess
Yeah, it's...the thing about LLMs is that often they're very bad at analyzing language. (See the entire "how many r's in strawberry" discussion/meme.) I think before a game like this can really work, LLMs have a long way to go [and, full disclosure, I'd personally suggest that "into the garbage" is where they should go...].
← Return to game
Comments
Log in with itch.io to leave a comment.
I beat the last level 4th try. Great game, made me think.
Mmm. It's interesting, but I'm not sure it's working? My main strategy was "just copy what someone else said", which was great until the last round, when I got:
"After analyzing all responses and behavior patterns, I determine that Fred is human. Fred copied Charles verbatim, indicating lack of originality typical of a bot."
So the second sentence says I'm a bot...and therefore I'm human? Hm.
Sadly my losing level-1 answer was overwritten, but I was dinged for not following the rhyme scheme, which I absolutely did. (That's the point at which I decided to just start copying someone else's answer.)
loll i did that too!!!
patched this so its a lot harder to copy answers and fixed some prompts so hopefully this doesn't happen again! unfortunately the llms do not follow instructions 100% of the time, but that's part of their weakness i guess
Yeah, it's...the thing about LLMs is that often they're very bad at analyzing language. (See the entire "how many r's in strawberry" discussion/meme.) I think before a game like this can really work, LLMs have a long way to go [and, full disclosure, I'd personally suggest that "into the garbage" is where they should go...].