Esoteric Update #227 - Help, I Am Trapped Under A Pile Of Books

You'll have to excuse me; this update won't be too detailed. Firstly, I am feeling rather tired and under the weather, and there's just so much I have the strength to do today. Secondly, we're in a rather tricky spot right now. That is to say; we're in the middle of juggling a lot of data to pin down the knowledge system before we make it public for testing/verification/feedback.

So what is happening exactly? Following the changes from last week, we've decided to move some things around in the game. Most of the content of the glossary got converted into knowledge. The rest was shifted over to the Locations tab. This cleared the glossary to be, well, a glossary of terms from different languages. Behind the scenes, the list of facts slightly changed as things got removed, added and swapped around. So we hope that by the time we have this out next week, it'll be functional.

For the rest of the update, I'll do something we've not done in a long while. Someone else on the "team" will write a bit about what we've been working on together.

Alright, so I've been told to talk about the two things I've been working on for Eso's game. Firstly I think I should mention the less important of the two things. I guess you could say I'm currently Eso's man-bear-creature for LLMs. And after a good deal of fighting with a few of them, we're at the point where we can get some good, useful content out of them. We went over a lot of new use-cases and some of them started giving good results. We've definitely have a good basis of generative content that needs to be still filtered and cleaned up for the game, but it's a good source of background noise and it doesn't pull away Eso's attention from other things.

But I think what she really wanted me to talk about isn't the LLMs, it's the conversation system. Not really a system, a sub-system. My own spin on her previous work. We've not quite managed to get it working yet, but it's not an issue with the idea but with trying to build some interface for editing these.

So essentially, most conversation systems are vaguely tree shaped, even the more complex results Eso's been getting with her callstack approach, mostly resolve into tree-like structures as you go through it. The issue that these tree-shaped structures cause is the way content is spread through them. You have an entry point, and until you can collapse the paths back into a join, each level requires more and more responses. This places some limitations on the different ways you can navigate a conversation like that. And obviously it should be stated that it's fine for a lot of content to work like this.

But think about cases like small talk or pillow talk. For those, it would be much better if we could find a way to make the distribution of content more equal, because that would lead, in turn, to more possible paths through the conversation. And that can be accomplished by the use of semantic networks (based on directed graphs, with loops). But there's an issue to this and it's not the concept itself, it's building a tool to work with it. It's not for a lack of trying, we've already tested multiple possible ways of doing this with existing software and it's just not working. So we're at the point where I might need to make an interface for this on my own. Fortunately I can program, so it shouldn't take time away from Eso's work.

I think that's all.

That might be all, indeed. For the moment, at least. I think it's true that this is a pretty interesting approach, and we might need it for the above-stated cases. But yes, the first attempt at making an interface for it didn't work that well.

Anyway, there should be some kind of small preview release next week, and I hopefully should be able to work on a few more things too, but I'm not sure what it might be. We'll see. Double lore feature next week, maybe? Technically speaking, we're indirectly dropping a lot of lore through the knowledge feature.