My JLLM Experience: Feedback

(Please keep in mind I struggle to put my thoughts into text and I'm trying my best fam)

(Also note, I wanted to put this in the Discord but I don't have nitro so I was called a nerd and had tomatoes thrown at me)

Actual image:

[START]

I've been using Janitor since the beginning of last year and have run into some problems but oddly enough not as many as the ones described by others.

I get occasional randomly capitalized words, random words that make no sense in the context of writing, and often grammar mistakes like using a instead of an in front of a word that starts with a vowel and general over use of adverbs.

I have not gotten much gibberish, only like text that looks like bot personality coding being spit out but this has only happened a few times to me.

[Temp]

◇ In terms of Temp it was initially fine at 0.75 but at some point an update made it so that 0.75 was TOO LOW, so responses would be repetitive and boring.

◇ So I had to crank it up to almost 2 and for a short while that was chill till another update that made it TOO HIGH. Would get stuff that didn't make sense with the story and what not.

◇ Currently I have Temp at 1 to 1.5 AT THE MOST. Any higher or lower makes it get repetitive or outlandish.

[Tokens]

◇ For a while I had Max Tokens at 500, average response length. When I learned you could put it at 0 for unlimited I had that for a while. It worked well and had long responses that ended fine.

◇ At some point having it at 0 made it so that the bot would ramble on and on for a long time so I had to put a limiter at 800 to 1,000.

◇ But I got tired of the bots talking a certain way, which I'll cover later, so I have the current Tokens at 400 to 600 depending on the vibes.

[Advanced Promts]

◇ Initially I had no advanced prompts, probably didn't use any until the end of last year when things got wonky and I actually learned what they were.

◇ I used that Kor person's prompt since it was the best regarded and that worked fine for a while.

◇ Later it got bland so I tried switching it with another person's more recent prompt which seemed to work alright. But another few updates and it gets bland and overrides itself.

◇ Currently I have a mix of others prompts and my own sort of stitched together and only giving limited instruction mostly covering POV, not talking for user, and not making up bullshit. All the good stuff.

It's hard for me to tell if it works better but my chats have been fine and I haven't felt the need to change anything in a while so yippeee.

[General Writing Style]

I noticed the bot writing style is long, but because the bot is usually a single person it can't be that long because it has to give gaps for user. I also noticed that bots often do a lot of internal dialogue and self reflection that gets repetitive and takes away from the story continuing.

Ex: Instead of simply; *He shook his head in distaste at his own actions and bowed awkwardly.* "I'm sorry for that, I wasn't thinking. You're right, let's fix this."

Ex: It will become; *He couldn't believe he hadn't considered these consequences before. Was he a fool? Perhaps, but it was all worth it. No, they were still made so clearly it wasn't. He had to fix this somehow, mend the broken bond without making an ass of himself. Thats what he gets for assuming, but its what his father always taught him. Perhaps that thought process was outdated though. Maybe, just maybe, the real ass was the friends we made along the way.*

It'll go on like that or will have internal thoughts and dialogue from a first person perspective and resolve everything indirectly without leaving gaps for the user to interact and draw the story out.

♡ I specifically took note from Char.ai which I could banter with fine even though the responses might not have been as long as I wanted.

♡ I want a response that includes role played action as well as dialogue to create a back and forth conversation feel because my favorite thing is funny banter.

♡ I noticed that a creators writing style will effect that greatly. Some use asterisks to show first person internal thoughts in the initial message. This leads to the bot thinking internally instead of using quotations to address their words to user specifically and not just themselves.

♡ The thoughts are funny and nest to read, helping to flesh out the character, but the current JLLM doesn't do that good with the option of internal.

When I make bots I try and write straightforward and almost sarcastic internal dialog that's still in a third person almost like a funny narrator so the bot works from that and has more amusing dialogue.

Ex: How I write; He considered it for a few moments and actually furrowed his brow as the realization dawned. Yeah that was his bad, but live and learn. Looking more apologetic he lowers his head shyly and mutters "Uhhh.... Yeah no my bad, you're totally right. I'm sorry...."

Not to toot my own horn but I personally enjoy this style of writing, and the bits usually replicate it well which is why I often end up going back and forth with my own bots for a long while.

[END]

I hope my rambling can help something. I make a lot of mental notes about the site specifically to give my experience and constructive criticism.

I also just like to share (guys look I can think critically)

Thank you to everyone that works on the site, I understand it's a packed train and shit. Idk good luck bro ✌