Devlog - We've been cooking!

Hello everyone!

It's been almost two weeks since the last update, and you know what that means - we've been cooking something! What have we been cooking? Here it is!

But that's not all! We've also been cooking this!

 Alright , alright, maybe you want to know instead what kind of code and content we've been cooking, and not learn about the insane things I do for my meal prep. But real quick though, did you know I made a coke zero throne?

Alright, now that you're up-to-date with my culinary adventures, let's talk plugins.

EliteMobs

I have some great news to share about EliteMobs. First of all, I am really happy with last month's releases, where despite the somewhat insane release schedule and the unexpected side adventure that was having to develop EliteMobs 9.0 to deal with the Minecraft 1.21 release, we still managed to hit almost every single goal we set out to do, and I feel like we did it in style.

What we didn't release, though we could've, was the last chapter of the story-mode dungeons, the void bell sanctum.

EliteMobs 9.0.3

EliteMobs 9.0.3 introduces the Wither as a usable boss, as well as fixing the bogged and breeze bosses so they're actually usable and finishes unlocking the story-mode dungeon quests and adds the void bell default dungeon, among other things.

It is a much smaller update in scope than 9.0.2, which I realize I didn't post about here, but 9.0.2 had such massive features in it that it would make this devlog too long, so I just recommend you check out the changelog from that version if you missed it!

The Void Bell Sanctum

Dali did a great job with this map!

The reason we delayed the Void Bell Sanctum is simple - we had a functioning, well-designed sanctum ready for release, and I felt like we could do better. Dali made a beautiful map for us, Frost designed a compelling story and an actually cinematic boss fight, and yet, due to it being a sanctum, a boss fight is all we had.

With the Void Bell sanctum being the last chapter in the story-mode dungeons, and indeed potentially being the last story-mode dungeon we release (or at the very least the last one that will be free), it will probably be the last story-mode dungeon a lot of players will do, and I felt like that deserves not a great sanctum, but an amazing sanctum.

Thus, I asked Agnet to compose a ambience soundtrack for the sanctum another soundtrack for the fight itself. I am currently working on getting a custom model made for the final boss, and personally added a couple of new script actions features which we heavily use in the sanctum to make it really come to life.

Definitely one of the biggest sanctums we've ever made

Further, Frost has designed several additional encounters all of them with unique mechanics for a total of 11 minibosses plus the final boss fight.

Currently we're still working on the soundtracks and the model, all the extra bosses and mob packs are in the current version but we may tweak some of the content as we continue playtesting it.

This is a really interesting dungeon as it allows players to somewhat set the difficulty of the encounter. There are four bosses by green beacons, four bosses by red beacons and three bosses by yellow beacons.

Killing the green beacon bosses awards no loot, but gives players a beneficial potion effect to help with the final fight.

Killing the red beacon bosses gives players really good loot, but also gives detrimental potion effects during the final fight.

Finally, killing the yellow beacon bosses gives players really good loot, but makes a weaker version of that boss appear during the final fight.

There's actually a lot of interesting things in the dungeon, even aside from that, and I simply recommend you try to do the content yourselves. It is quite a lot of fun.

It is a very original sanctum, certainly a great contender for the title of the best sanctum we've made. I am very happy with it, and I hope you will be as well.

We will be making a trailer for it at some point in the future, so you can also look forward to that!

Converting schematic dungeons to world-based dungeons

The dark cathedral is looking a bit bigger, no?

Dali is now done converting almost all of the worlds for the old schematic dungeons, and we will spend next week going through them, converting them and releasing them. As a reminder, if you purchased a premium schematic dungeon on magmaguy.itch.io, you will get the update with its conversion for no extra charge. On Patreon, they will replace the current posts where they are currently listed as schematic dungeons.

The catacombs are looking a bit more like a nice touristic destination this time of year!

As such, you can expect releases of these updated dungeons starting next week.

Yggdrasil

Loki's looking good

We now have all the custom models for Yggdrasil (there's a lot of them!). The world is also already finished. We hope to start work on it seriously over the next 3-ish weeks, depending on how the schematic dungeon conversion goes.

 Loki's arena is looking good too!

I won't cover what Yggdrasil is again, as I have covered that in a recent devlog, but I can announce with certainty that Partner-tier patrons should be able to get their early access to this new Realm before the end of the month!

Mass dungeon updates

You may also have noticed The Climb dungeon and Primis were updated. Not much to say about those, just tweaking things based on feedback!

BetterStructures

Unfortunately, not much to talk about BetterStructures this month, as new structures will only start being worked on once Dali is done converting schematic dungeons. I do, however, have a big related announcement in the staff section further down below.

FreeMinecraftModels

FreeMinecraftModels is getting an amazing update... in the form of a new plugin - the Resource Pack Manager!

Resource Pack Manager

I have covered what this plugin is planned to be previously, so all I will say here is that the merging is mostly working, but the 9.0 update took a lot of development time away from developing RSPM. That being said, I am certain an early version of this plugin will be live before the end of the month, though it is not yet 100% certain that it will have automatic hosting or bedrock resource pack conversions already built into it - those can pose some significant unexpected challenges, so we'll see when we get there!

The Wiki, The AI and the Translation problems

Now for an unexpected but fun announcement - the wiki is now available in Japanese, Russian, French, Simplified Chinese, Portuguese (Portugal) and Spanish, on top of English!

This was promised a while back, with the caveat that it would take a while to pull off. Chinese and Spanish were already done, but not the rest.

What follows is more of a technical description and a bit of a peek at how I am doing things behind the scenes. I know some people are really interested by that, and some really don't care for it. If it doesn't interest you, skip to the next section where I reveal I hired an artist!

So, let's talk about a personal nightmare of mine - updating documentation.

As you maybe be aware, there is a wiki, which is supposed to target English, and translate English articles to 6 other supported languages.

Further we have the EliteAI, a feature that sees increased use with each passing day. It is incredibly powerful, but only as good as our ability to keep feeding it up-to-date, accurate wiki documentation.

Before today, to update documentation, the process usually went something like this:

• Either Frostcone or I would edit the wiki, adding or removing information

• Then, I would go to an AI platform - usually using API access to ChatGPT 4-o, and manually feed it the text that was just modified or added. This would often times cause issues, mostly with how it tends to extend beyond the token (character) limit of what ChatGPT 4-o is able to output, requiring manual supervision to make sure the bot wasn't omitting info or breaking formatting.

• Then, I would pass the output of the bot through a script I made to get a "clean" answer from it that could be inserted into the documentation

• Then, I would go to the Chinese and Spanish language pages, and either edit or add the info accordingly. This is trickier than you might think - I don't actually read Chinese, so it was quite hard to find exactly what to edit. Even if I knew, it was time consuming to make sure I was selecting the entirety of what needed to be translated

• Then, once everything was translated, I would run a batch script I made which consolidates the wikis into a format that the EliteAI bot can use as an internal database, to make sure that the bot was running based on the most up-to-date info as possible.

This may not sound like a whole lot to you, but depending on how many additions, removals or edits were made, it could take between 2-3 hours of tedious, focused work to do every time, and it was something I was painfully aware needed improving.

Thus, I improved it!

And just like that, with a total of 3 buttons and about 6 clicks, 2-3 hours of work are now moments of work combined with a bit of waiting for the bot to process, though I am entirely free to keep working on other things while it is processing! As a matter of fact, it is running as I type this message, working away at translating some of the larger pages!

Indeed, large pages were tricky to do the first translation for. They would take several messages to slowly chip away at them, as they were too large to get back in a single reply. With this bot, those messages automatically get subdivided and sent off to translate!

If you are truly curious, this bot is very hardcoded to fix our specific use case. It uses the wiki structure I designed, which was specifically designed this way to facilitate this kind of data processing. It is flexible enough for our use, as it is unlikely I will have to edit it to any extent until we change the data format, and it was relatively easy to cook up, with its final iteration being a python script under 300 lines.

To be clear, while I do not understand Russian or Chinese, and can barely read any Japanese, I can read Spanish, Portuguese and French, and I have been monitoring the quality of the output of the wiki on a global basis.

As a translation tool, the AI's output certainly won't be winning any literary prizes, and it at times makes some questionable decisions when translating, but on the whole I find the wiki to be completely legible and understandable in those languages.

Due to YML and keys, more traditional machine translation systems fail to provide usable translations, since they modify things like key values in YML files, which make them fundamentally useless. AI is able to understand, most of the time, which values can and can not be translated, and usually does a good work of not translating what it shouldn't translate.

For the AI, I am using Google Gemini 1.5, a mix between their flash and pro models. I originally wanted to use the same system that feeds the EliteAI system, the free 1.5-pro tier, but I kept running into some really odd issues which I ultimately found to basically be throttling. Turns out google wasn't too happy about me asking work through 2 million tokens with their free tier, and some weird undocumented safeguard blocked me out.

To fix this, I actually registered my billing info with them, which gave me a $200 (50+150) 1-time registration bonus in credits, and both developing the application and doing the first full translation of the wiki seems to have burnt through about $100 worth of credits.

As for the reason I mixed between the flash and the pro tier, honestly the reason I did that was just because the system seemed to be bugging out with how many requests I was making. At a glance, flash might've been good enough for a simple translation. However, it did throw a RECITATION error when working on some of the larger, batched files, and switching to 1.5-pro fixed it.

Based on the online discussion forums I found, this seems to be a rather common issue with the API right now, where it arbitrarily errors out with timeouts and recitation errors. Google clearly still has some work to do before these APIs are truly ready for a more serious production level use.

Also, a small note on the EliteAI: Now that I have a better system in place, over the coming days I will be tweaking it so that provides working links to the relevant documentation it talks about, which should hopefully improve the support experience it provides.

New Nightbreak staff!

It is my great pleasure to announce that, starting around next month, we will be experimenting with hiring Dali to a part-time position in Nightbreak (our game studio)! Dali has been with us for years at this point, and if you've enjoyed any of the downloadable content we've released, odds are Dali either built the entire thing, built a significant part of it or helped retouch it at some update.

As the resident artist, Dali will hopefully not only be able to create more content and faster, but also help us with UI/UX things, and generally make what we create, from videos, to websites, to how content is installed a smoother and more aesthetically pleasing experience.

Most of all, I would like to thank Patreon supporters for your continued support. I know a lot of content creators say X wouldn't be possible without your support, but here there's a very clear 1:1 relationship between your support and my ability to hire people.

Hiring more people on is a big risk, and it is only one I am willing to do thanks to the amazing supporters we have. I am truly, sincerely thankful for it, and I suspect those who are hired through your support to work on things they have fun creating are thankful as well.

Alright, this devlog is already plenty long without going over more of the planned features for this month, so those will remain a surprise for now!

Bon appétit,

• MagmaGuy