Play game
Deduction Pawn (Unfinished)'s itch.io pageResults
| Criteria | Rank | Score* | Raw Score |
| Innovative Gameplay | #31 | 2.582 | 3.333 |
| Emotional Depth | #43 | 1.807 | 2.333 |
| Satisfying Answers | #44 | 1.807 | 2.333 |
| General Enjoyment | #45 | 1.807 | 2.333 |
| Logical Consistency | #46 | 1.807 | 2.333 |
| Overall | #47 | 1.807 | 2.333 |
| Immersive Atmosphere | #47 | 1.807 | 2.333 |
| Puzzle Fairness | #49 | 1.549 | 2.000 |
| Jam Theme | #50 | 1.291 | 1.667 |
Ranked from 3 ratings. Score is adjusted from raw score by the median number of ratings per game in the jam.
What makes your game a Mystery Game?
It's a story driven game with 2 mini mysteries that the player has to solve in order to continue.
How does your game integrate the jam's theme "Last Message"?
The first chapter was supposed to figure a big mystery with the Last Message theme, but I didn't have time to make it.
Does your game use generative AI? And if so, in what way?
AI ART is used in this game in 3 assets:
- One of the blurred backgrounds is AI generated
- The vertically scrolling background in the beginning is AI generated
- The teacher's sprite is AI generated
These 3 assets were supposed to be placeholders, but I didn't have time to replace them with human art.
Other than that, the way answers to mysteries are entered in this game is by typing plain text and a LLM(Large language model) decides if this is the correct answer to the mystery.
Leave a comment
Log in with itch.io to leave a comment.
Comments
With the types of questions and answers, I'm not sure why you needed to use an llm instead of a traditional text parser. I'd imagine that an llm might work better to mark answers correct when they get the gist of it right, but I had the same issue as the other commenter with the inputs being extremely picky about wording/prasing. (For example it didn't accept 'scary movies' for one puzzle or 'they have to/are forced to' for another, despite those being basically the answer)
If it's going to be that picky, there's no reason not to use a traditional parser.
Hi, thanks for the feedback!
Apparently my API key expired so it marked everything as incorrect... I fixed it now. But I do agree that the AI can be a bit restrictive with the answers that it accepts. I plan on either relaxing it more or making it return text that will guide the user to the correct answer.
If you want, you can try to play the game again and review the AI's actual responses.
Unfortunately the answer I put for the first question I think should have been marked correct (it was pretty close), but it said it was incorrect. Also couldn't figure out how to start the game (I eventually pressed Enter and it worked). The second question I thought was confusing and I wasn't quite sure what it was asking me. I felt like I was guessing the answers rather than having the knowledge provided to me in some way that I needed to think through and decipher. Cool idea with the LLM, I tried something similar with a joke evaluator for a small game jam one time, but I didn't get very far with it. I think if you continue working on this, it could be pretty cool!
Thanks for the feedback! Do you remember what you put as the answer to the first question? And I'll try to rewrite the second I agree it's confusing.
Hmm I thnk it was "watching scary videos" and then it said I was wrong. I gave up and it said "they were sharing scary videos" or something along those lines. I know it's not exactly the same (again I didn't know if I had to guess or if maybe I was just missing the exact solution somewhere), but it seemed close.
okay apparently my api key expired so it marked everything as incorrect... I fixed it now but "watching scary videos" is still incorrect because it doesn't contain the "sending to each other" part. I'll follow your advice to relax the AI or make it return text that will guide the user to answer correctly. Thanks!
Yeah that's a good idea. If I give an answer that is related but not correct, it would be good if the AI generated a response that guides me towards the answer or asks me to fill in the missing piece of the answer. "Sharing scary videos" makes sense but what confused me was that they were all reacting at the same time so I had originally answered something about streaming videos. Anyway, cool concept and I hope you can keep going with it. One tip is that you could potentially store the most common answers locally so you don't have to send everything off to the AI, so if someone types exactly the right answer then you can mark it correct, then if it's anything else you send it off for processing. Just to save you some resources/$$$.
I'm sorry, but I don't want to connect a web game with my personal accounts in any way.