LudumData

Collector by primitive-concept

[raw]
made by primitive-concept for Ludum Dare 58 (EXTRA)

This game is a sort of abstracted LLM training simulator. Not at all accurate, and not a ton of fun to play if I'm being honest. However, it does seek to make a few points.

If you're wondering about my own views on AI, whether I'm for or against it, I'll eventually reveal some of my own views at a later date after this game gets some initial feedback. I don't want my own views to bias any of that feedback, because the feedback I get is part of the thought experiment. Be as supportive or brutal as you need to be. It's kind of the point!

Along with my own views, I'll also make the source code public, as that's also part of the experiment and a sort of footnote for the game itself.

All of that said: From what I've already stated here, and what you might gather in later stages of play, you might be able to piece together and infer some of my own thoughts on the matter yourself.

HTML5 (web) https://primitiveconcept.itch.io/ld58-collector
Original URL https://ldjam.com/events/ludum-dare/58/collector-2

Ratings

Given 8🗳️ 3🗨️

Feedback

BlackCat508
Oct 13th · 20:00 UTC
I really like the atmosphere! It kind of feels creepy, which is probably intended. Pretty cool!
Daniel123
Oct 13th · 20:04 UTC
To some extent, I think this kind of data collection could also be considered consistent with the theme of “collect.”
Back to the topic.
In this jam game project, I tried integrating a pre-trained Distil BERT model into the game to interpret player intentions and create a small text-based puzzle game.
I’m really curious — did you also integrate an AI model into Unity to build your LLM simulator?
🎤 primitive-concept
Oct 17th · 03:25 UTC
> I'm really curious -- did you also integrate an AI model into Unity to build your LLM simulator?

Narp, it's all just random chance usually, weighted at times by your AI's various stats. Except during the first run of each phase, those always include specific sentences/prompts to help get certain ideas across and set the tone.
Game-mon
Oct 17th · 08:40 UTC
Really cool, and surprisingly fun. The writing of phase 1 was quite good, and the game renews itself with new phases, which is super nice. I was also engaged because I was constantly curious, wondering - how's it made? is it using a LLM to generate sentences? is there a semantic check on phase 2? etc. And I think having a part of tech mystery/not knowing how a game is programmed helps a lot getting me engaged in the game.
Kash
Oct 20th · 23:19 UTC
Honestly, I found the game to be fascinating and my curiosity definitely pushed me to finish the game. The length is just right, and I like how it's a story based off of certain puzzles. From my understanding, you can improve the model to generate better results before you scale up. I think that's brilliant. I don't know how you did this but well done.
5H08
Oct 21st · 23:06 UTC
This is so damn cool. One of the most unique games in the jam. I love the underlying narrative and how real the process felt.
FireSlash
Oct 22nd · 07:34 UTC
Super interesting little game.

The mix of "work" and narrative is super interesting here.

Would have been nice to have a bit more mood setting (Maybe some text beeps, ambiance, etc) to make the world feel a bit more alive.

Cool entry, nice work!
SergS Workshop
Oct 22nd · 07:42 UTC
It’d be really fun to integrate a simple LLM model into that concept! Good submission overall - just missing some SFX. Adding keyboard clicks or terminal-style letter sounds would really boost the immersion.