Been some time, huh? I'm finishing up Laserpit. You shoot one another with lasers. It'll be available here alongside Windows and Linux.
I also released an april fools game on that day. It's about Hikari Maiden slapping the hell out of some mothmen.
Apparently I'm search banned on twitter. Cute.
You can always follow me on other places listed in my profile links. Take the time to also check out my recent art. Last month I drew Mistral from .hack everyday.... I posted some highlights alongside usual art.
I'll make a front page post when Laserpit releases then go back to work on Conflict deFuse.
Protagonist Fusie's design is beginning to converge which is lovely. She's such a mischievous witch, I think everyone I showed her to loves her. Me included!
ChatGPT is becoming the big thing and I have incorporated it into my tool set. This has allowed me feats such as writting a RSS feed maker for my website (I'll get to it in due time.), some random useful batch scripts, even things like an interface... to ChatGPT's API.
Not really feeling like working on this further since making games is more of my thing. I hope to see UI creators improve on stuff though. TavernAI and similar let you create and use character "cards" (PNGs/WEBPs with .json data, or just raw .json) that make the model act like a character. This is superior to the usual boring personality of GPT-3.5/GPT-4.
Dammit, she's onto me... AGI real?! Hope her level isn't max....
Most people are using this so they can do naughty things to their favorite characters. This is stupid. You could be using this to have your questions answered with an actual personality. Mind you as always, always expect models to be confident even about factual errors. I've caught GPT-3.5 fumbling around when asked to write GDScript. However they tend to arrive pretty close to the crux of the the answer and if you know the subject well you can spot the issues and just either correct the model or do some fixes. Be smart, be cool.
libtcc is not super well known! I asked about exposing functions from game to C 'scripts' and got a thoughtful reply. Have yet to test but subsequent questions did fully check out with my 3-4yo C knowledge.
If you know the space well, GPT-3.5/GPT-4 and alternatives such as self-hosted models (LLaMa is my favorite, I got it to run on my PC through llama.cpp and models based on it such as Vicuna manage almost 90% parity with GPT-3.5 quality!) let you get the boring stuff done so you can think about the important bits. It's also good for chatting with for certain creative thought processes. Like the perfect rubber duck debugging duck.
I've also managed to try out a fork of TavernAI where there is some small support for extensions. One pretty much does now what GPT-4 will eventually include: image2text.
It's not perfect but it's pretty close.
Considering you can also get it working with a self-hosted model, anyone with a high end PC and enough reading-the-freaking-manual can already have GPT-4 at home. Now that's the stuff.
Anyways, stay tuned for more for now. Have a good saturday.
