I'm glad I joined this jam as it motivated to push something out, regardless of it's polish. That's difficult for a perfectionist like me, but it was motivating in the end, getting some kind of foundation out there for others to see.  Those same foundations I've been working out of since submission, on a different project that uses some of the same ideas. 



Working on the next iteration of my game from the jam. I wanted to strip out the pet/afk/idle stuff and reorient it more towards the core mechanic I liked the most: exploration via words with some light gamification/interpretability sprinkled in. I map neurons 1:1 in a model to star systems in our universe, put it into deterministic mode so that the universe behaves deterministically, and measure peak neuron activationsin real-time from player word inputs. Each peak neuron activation returns an index value of that neuron (0-3071) and in this way we have a little mechanic to discover/unlock new systems via words. In the pet game, this is too slow a process - flying between planets. Better to take a "cosmic" view and ping them real-time, so you can go through ten inputs in ten seconds and quickly realize what works, and what doesn't. This system is working a lot better as I playtest it. 

Each newly-discovered system takes us up one rank, so discovery/interpretability is grounded in the game's core "levelling" system.  This lets me gate access to specific visualizers/filters/analytics so it's not a firehose of information - still working on the tutorial at the moment. 

After discovering enough systems players get access to their first data cache. This does the input for them. At the moment, just to test the system, I have Moby Dick (aka The Sea). It parses ~7,000 pre-computed sentences and their associated neuron activations into the map at various speeds. Additional filters help make more sense of these activations, eventually letting the player start to see some kinds of structure. But that's getting in the weeds a bit and this is already long enough lol~

The visualizations, the way stars are spawned, the cube formation - all of it grounded in meaningful intention about presenting things honestly, all packed with some good lessons about LLMs and interpretability. Teaching all this via the game's system alone would be valuable, I think, so that's my next project. :)