now I get it

no bots, the event is FAR from complete... IK, but I was genuinely so shocked by this I feel like it's worth an announcement post

SO, I was larping with a few Sonic personas (for a bot i will not name because you all will laugh at me), and to be honest...

JLLM is actually pretty good. For being free, that is. I know that things like Deepseek, Gemini, and a multitude of other models significantly outclass it in memory and response quality, but again... they're paid. And yeah, paid things TEND to be more high quality than free things.

Sure you can make the argument that some obscure API provider you're using still has some free top-tier models rolling out, but though die out pretty quickly nowadays, unlike... 2024 to early 2025.

JLLM, at least in my experience, tweak the settings right and use just the right amount of context (yes you need to actually contribute to the roleplay otherwise the bot will narrate for you since it needs data to work with, don't be dead weight), it can ACTUALLY be a fucking decent model!

...but I am getting way off track here.

So, prior to a lot of people talking shit about the LLM, I was one of the ones thinking it's massively overhated and actually decent for what it is at a certain price point (free)...

That was until I kept getting bullshit gibberish. So I got sick of it and moved onto a free proxy I know! But that was bugged, so I moved on another, and the responses were just... dry. And often acted out of tune unlike how I remembered they used too, so much in fact that the few times JLLM did decide to work it was actually more enjoyable and listened to prefills and responses better.

The chat memory worked perfectly fine, and I had no idea what was causing it... up until I tested it on another browser. And indeed, JLLM gibberish responses can also be an issue with your browser...

I use Opera GX mainly. And when I used it on Chrome it worked perfectly fine, decent responses and rare gibberish. Sure, it took longer unlike the constant 10 rerolls I had to do for a lightning fast proper response but even then it could just decide to hate me and not do it anyway, but I'd prefer the former any day. Do I have a fix? No. I'm trying to figure out what jumble of code went wrong and made the thing have a stroke in it's responses.

It was kinda dumb. Since Opera has mostly the same functionality as chrome just with a fancy customization and ui update. So JLLM doesn't work on Opera GX... at least for me. Of course you can mitigate this with proxies but I also find it tends to be a little more dry... I don't know why. And yes the prompts and settings are the same and everything.

moral of the story: check your browser