You must be 18+ to view this content

Esoteric ♥ Esoterica may contain content you must be 18+ to view.

Are you 18 years of age or older?

or Return to itch.io

Esoteric Update #160 - The AI Is Skipping School


No lore today, but perhaps I can offer you something else instead?

A Primer On Alchemy
A Primer On The Sefirot

Now, for the update. There are a few things I'd like to report on.

So firstly, let's look at the issue of the covers. Last week, there was a feature I couldn't quite get done, and I only finished it, ultimately, at about the 4th pass at it, today. It's nothing too impressive, so I won't bore you with more generated covers, but it's more of an optimisation that reduced the size of the files associated with the process by around 30% while increasing the number of variations by 400%, even if said variations are not hugely different.

Secondly, we're winding down asset acquisition for the books; I think we mostly have enough of everything and can return to getting things on the fly as they are needed. I'm still processing some of the files, but I'd like to change focus again for a while, as I'm getting tired of image work.

Thirdly, I've re-formulated an up to date design document for the game, but I'm going to stew on it for a while before posting it anywhere. The draft can be seen on our Discord if you are curious. Possibly I'll put it on Itch later this week, and if that happens, I'll attach it to the next update as well.

Finally, I ended up delving into a new theoretical subject, as I wanted to expand the game's AI toolbox with more reinforcement learning methods. I've decided to go with Finite Markov Decision Processes trained by either Q-Learning or Modified Connectionist Q-Learning. Yesterday, I felt pretty frustrated with not making it work in the game, but thinking about it today, I completely changed my approach, and I have a working prototype for both of the learning methods. Anyway, that's really more of a spoiler for next week. I'll try to build some demo. What exactly is the point? Making NPCs learn more complex iterations. The system I had previously only worked in a very limited context, whereas a policy over an FMDP can do a lot more. The issue is mainly the data representation and making the system maintainable, though, as it's not exactly meant to do what I'm trying to make it do.

So anyway, I'm tired, I've been at the doctor again today, and they took more of my blood in exchange for sagely advice. Such is life.

Get Esoteric ♥ Esoterica

Leave a comment

Log in with itch.io to leave a comment.