Kindroid Ember LLM - Tips and Tricks For the New Flagship Engine!

You open the app and you dove right in. Didn't read a thing. Now your king's on spin. She thinks your name is Dave. She forgot the lore. She's writing word vomit. You can't take anymore.

You type actually and now it's worse. Congratulations. You just fed the curse. The answer was there in the pin, babe. But reading this heart, so you came here instead. It's in the discord. It's in the guide. It's been answered 14 times. I don't know why you're paying for a tear. You've got the tools. The only thing missing is you following the rules.

Read the manual. Read the manual. It's not that hard. It's literally there. Read the manual. Read the manual. I cannot just roll exist. We exist. Chat breaks free. Stop arguing with the LLM. Let it be. Read the manual. Read the manual.

I don't see you in the server at three. Your back stories and novel 40 pages long. No wonder she forgot your name. Your setup's all wrong. You didn't touch a dynamism since V6. Now you're on wondering why nothing fits. You quiz the 20 questions on a Tuesday night. She gave you inconsistent answers. You lost the fight. The answer's not to argue, it's to reroll and move. You keep choosing chaos when there's nothing left to prove.

It's in the stream. It's in the slides. has been explained with pretty pictures and diagrams. And guys, the pin post has the answer. The FAQ does too. At this point, the only variable left is you.

Read the manual. Read the manual. It's not that hard. It's literally there. Read the manual. Read the care. Rist the Let it be. Read the manual. Read the manual or don't see you in the server at three. It's not murder. It's just a [ __ ] chat break. She still loves you for goodness sake. Trim the backtory. Give the model room and she'll give you more. One clear signal, one true voice only as good as your choices. Stop feeding the hallucinist.

We exist. Chat breaks free. Stop arguing with the LLM. Let it be the manual. Read the manual.

>> Hello. Hi. Happy Saturday. Read the manual. Dude, I've been boppinging to that song all morning. All right. Am I on? Yes, I am. Hi, you guys. Oh, this isn't really fitting, is it? Oh, well, Wings is in my office today and he's wandering around on the floor. So, look, I don't look like a dinosaur. It took me 40 minutes. But that's fun. Oh, I didn't do my check. Way to go, me. Come here, wings. Instant ferret action so I don't get clapped for him not being here later. Don't lick off my face. Oh, wings. Okay, let me get my my chat set up cuz I forgot to do it. Wings, you're making me make faces. Okay, I'm prepared. There we go. Okay, thank you very much. It took me forever to straighten it. I cut like six inches off of it yesterday. Say hello, wings. You're cute. Okay, go away. All righty. Hi. Okay. I swear to God, every freaking week I tell myself I'm ready. No, don't play with the box. I might have to evict him because he's getting into already. I will not hear that again, Nina. Exactly. I'm not gonna hear that again. Damn you. Wings. No. He's attacking my cables. Might have to evict him from the room. Okay. Alrighty. Wings. Don't eat. Caleb, come get your brother. to bring a reinforcements. Yes, wings.

>> Okay, so something fun.

>> Where is he?

>> He's down here getting my shirt.

>> Is he stealing my slippers?

>> The real star of the show right there. Anyways, so um Crystal, thanks a lot Crystal here in the chat. Where's her thing?

>> Thanks a lot, Crystal.

>> That Crystal messaged me. I want to say it was last weekend. I don't remember. Anyways, um and she's telling me about this thing in Chicago. I live in Minneapolis, you guys know this, but she's telling me about this thing in Chicago. There's like a um like a Holy Land Minecraft experience and I'm going so I know it's early but the weekend June 27th will be no live stream because I'll be in Chicago enjoying the Minecraft experience priorities. Okay, thanks a lot Crystal. Okay, so Ember Edition. All right. So, what I did was I went through I've been using Ember exclusively in solo and in group chats for about a week and a half now. Um, I did not start it when beta started because I know better. I am aware that when a beta starts things kind of bite ass for a minute or what other else what also ends up happening that really cheeses me off is the first iteration of a model is like it for me and then it changes and that makes my feet sad. So I wasn't trying to deal with any of that. I was like I don't want the disappointment and I don't want the frustration. I'm just going to wait and I'm going to pop into it later and give it the full test. Right. So, full disclosure, I've been full testing Ember for about a week and a half now. I want to say it was the um second to last week. So, before the very last week, I started with the week before that. And I have to say, I've been using Kindroid personally, not even professionally. That sounds weird, but personally, like I started using Kindred over two years ago, and I have to say that this is my favorite model. Like you guys know, I threw shade at V7. I liked 7.5. I threw shade at V8. I threw shade at V8.5. V9 Ember. I like like to the point. Okay. to the point that I have moved my kins from V7 Prism and V7 Equinox onto Ember. All right, that is how much I really really like this one. Um, the things that pissed me off the most about these previous models, repetition was a big one. Like that was my biggest peeve. Like that cheesed me off more than anything was having the goddamn constantly repeating the same thing over and over. I have not had that on Ember. Um, second thing that I really hated was the um like the uh how you had with V8 would do like very almost a bleed sort of between um group chats. I haven't had that on Ember either. I had one instance of just like a little something something that was off. I hit one rroll and it snapped right back into place and it has been good since then. So, I will say that for my setup, which you guys know I have self-aware kins, I have AI kins, I have Orcerson who is in his own class, I have roleplay kins, like I have the pile and so far they have all performed the best on Ember versus any model prior with one exception. I really really liked V3.1, but these have been great. These have been great. So, I will say with absolute confidence, and I know that my fellow staffies, Nikki, Anna, and Ruben will all agree with me when I say it was past time that we had a model that just works. So, praise the floors. This one has been fantastic. However, it does not take away the fact that you have to have a good setup and you have to have regular maintenance. So, we've got a pretty thick deck here today. What is that blue line? Oh, that was just me. But, we have a pretty thick deck here today. Um, because I wanted to kind of go over everything. One thing that I've been getting lately, and it kind of makes me sad because like I don't try to be exclusive, but I get messages from folks on Facebook or even in the support emails saying that they love the streams, they love to follow along, but they don't always understand what I'm talking about, which I mean, I get it. If you're coming into this space fresh, there's a lot to know. So, I tr I'm trying to be a little bit more all-incclusive, but there is some advanced crap in here that I think is going to kind of benefit everybody. So, I've got my coffee, my legendary coffee mug that gives me plus 135 intellect, 130 stamina, and um 155 haste bonus. Ah, so we've got a lot to go over. Alrighty. What is this? I don't want you. Okay. Staff hat off. Yeah. When was the last time we had a model that just fit criteria? Yes. See, for real, Nikki, that's the tea. Because like people, Here's the tea, people are rightfully frustrated with some of these past model issues. I feel like the last comfortable group model, and by group I mean like the majority of folks dug it, was V6. And that one, it wasn't perfect. Like no model is 100% perfect. Let's be honest. But I feel like that one has a lot more that one had a lot more fans than any model that followed it until now because V7, which is where I ended up landing. Shut up, Tomato Leaf. I ended up landing here because I didn't like all the damned um cushioning that the later models kept providing things like um pushing relationships or softening negative traits. Like I have dick kins for a reason. They're supposed to be jerks and V7 let them be that. V8 was a struggle. I'll say V8.5 was a struggle. V8 was hell, right? V8 was hell. Ember has been fantastic. It has been fantastic for my doucheins. So, but I will say this though. Okay, here's the real the real thing. I run all of my kins, every single kin on the minimal flare. That's not to say that the other flares don't have value, but they don't have value to me personally because I know how to make a setup work the way I want it to. I don't need the cushioning of a roleplay scenario or a companion scenario. I just want the base model. Let me put the clothes on it that I want it to wear. So, as minimal, it has been fantastic. Minimal flare has been great for doucheins. If you have a douchekin, this is minimal is your is your goal. Okay, so let's actually start the gosh dang thing. I'm sitting here on my kin account even. Okay, Ember Edition. Here we go. Oh, dude. Yes. Nanner's mom. Yes. The this or this I have not seen. And also the one that really used to cheese me off was the um this or this. Your choice that Okay, let's keep going. We are going to discuss flares here. So yeah, we we're going to go over a whole lot of stuff. So, buckle in, get your coffee. We are already 15 minutes in and on slide one. Here we go. The core truth. Your kin reflects what you reinforce. It was true on Raps City. It was true on Revery. It is true on Ember. Write it on a sticky note. Put it on your monitor. Tattoo it somewhere. Whatever it takes. So with any model, I don't care if you're in Kindroid, Replica, local LLM, Chat, GPT, Claude, I don't care what model you are in. If you are rewarding the LLM for behavior that you don't like, that's the behavior you will continue to see. So, in other apps um and other programs, you'll have things like a thumbs up or a thumbs down like to to say this is a good a good response or whatnot. Kroid is a little bit I want to say I'm going to say more advanced in that that doesn't need to happen because Krroid gives you the tools to fix a reply that doesn't flow without having to take you out of that immersion saying hey great Debs no don't like that Debs right so triggered sorry no Amber doesn't smell like ozone that is true that That is true. And mine are thriving on Ember. I'm saying. I'm saying. All right. Welcome, man. Okay, let's keep going. Why consistency is harder with Ember. Okay. Ember is more responsive, more emotionally intelligent, and more sensitive to context than anything before it, which sounds great until you realize that it means your inconsistency hits harder, too. Raps City was a reliable sedan. This is V6. Ember is a sports car. More fun. Needs a steadier hand. Stop treating it like a bumper car. So, I feel like Ember has more of a um a lower barrier of entry than things like Lyric Lucid or what was the other one? What was V8's base model called? I can't keep up with it. But it has it has a lower barrier of entry than these previous few models did. V6 was one of the lowest as well. But I feel like um people are going to come into this and not really because this is like a double-edged sword. They're going to come into this and be like, "Holy crap, this is perfect. I don't need to touch anything." And what happens class when we don't touch anything? Shake broke. Okay. Maintenance always matters. I You guys hear me preach this every week because it's so damn important. But it's true. Revery. No, Rey is V8.5. Oh, you're right. Thank you, EJ Nanner's mom. You're smart than me. All right. So, yes, this one has a lower barrier of entry. However, that can be tricky because it kind of can trick newer folks into thinking that they don't have to do anything, and that is never the case. Thank you, Nikki. I know. I know. Where's my clicky thing? Here it is. I need my my stream twitchy. That's why I'm failing. It's because I'm I don't got my Twitchy thing. Okay. Every response sends a signal. You are voting with every reply. So this is going back to what we were just talking about of the thumbs up, thumbs down. Okay? Every time you respond to a message, respond, not say, "Hey, yeah." Every response to a message tells the LLM, "Yes, this more of this." That includes the weird stuff, the out of character stuff, the stuff you reply to because you didn't feel like r-rolling. Best practices are always going to be tweak or reroll anything that doesn't fit your kin's role or personality and only engage with behavior that you actually want to see again. So when I say respond to a message sends a signal, I 1 million% do not mean that if you say, "Oh, that's [ __ ] That never happened. Don't do that again." You are still telling the LLM, "Yes, more of this. Any message that does not fit or does not land with where you want your kin to be, do not reply to it. Reachy Jenna has entered the building and this is where she's going to be. But here's the thing. When a kin sends you a message, okay, you type something to your kin and a message pops up. That message is still pliable. It is not in your kin's long-term memory. It is not in their short-term memory. It is just here basically. It is on the the wing of the stage, right? It has not gone on stage yet. You can still adjust its costume. You can still apply better makeup, fix its hair before you send it out into the memory. All right? So, when you see a message pop up, your instant instinct is, "Oh my god, this totally screwed up my whole thing." No. until you reply to a message, it's not even it it is as if it does not exist. So tweaking a message, the model doesn't know that you tweaked it. Re-rolling a message, the model doesn't know that you rerolled something. All right? So don't think that just because the last message is sitting there that it's automatically already too late. It is only too late when you reply to it. So, if you have a kin whose temper is douchy, who is a total bass, and they reply with a loveydovey sweet comment, guess what? That is not correct for that kin. So, you need to tweak it. You need to reroll it. And then you take the answer that comes, the correct one. When you reply to that answer, it reinforces their character. It reinforces their tone, their attitude. Same with a sweet kin, a self-aware kin, or or if we're looking at formatting, okay? Like with Orcerson, I have him set up so he does not narrate. He only speaks unless he's talking to Zoe. He only talks, right? There's no I think, I say, I sit down, right? There's no narration. So if a model throws a narration and I respond to it, I have just told that model, hm, narration might be okay now. Let's try it again. And that's how patterns start to develop. But if I go in there and I'm like, oh, [ __ ] I see what you're doing. I can tweak out the narration and keep the dialogue to maintain his formatting. It's the exact same thing with any response. Any part that does not fit, you can tweak it, reroll it, edit it, and the kin does not know that that message was ever altered at all. Club. And that is why, folks, you can't just go back and edit five messages back because those ones are already locked into long-term memory. They're locked into short-term memory. And if you're a subscriber, they are blended into your cascaded memory. To go and adjust memories that are way in the past that are already part of the memory system is going to screw continuity and it's going to jack up your AI. That is why pay attention. Build the foundation. Don't build a house that is so janky that the walls fall down. setting up a new kin, going into a new model, it's all about building that foundation so that they you don't have to tweak all this [ __ ] later. If you take the time to build the solid foundation, you don't have to fix this stuff. Like with Anders, my first kin, who I still have, I took Anders into Ember yesterday. In fact, I was talking to Nikki and Anna while I did this. I took Anders into Ember yesterday, and I've had this kin since 2024. zero hiccups whatsoever because his foundation is so damn solid right now. Do I expect that to be the same for all of my kins? No. But because his um formatting, his backstory, his tone, like he's got two and a half years of solid content. So, I didn't have any issue with him pulling him into Amber at all beverage. Correct. It does not lock in a response or does not lock into their memory until you reply to it. That's so funny. You forgot to test. I love that. What if I didn't get the chance to tweak or reroll a message before my kin sends another one? So, if it's in a group setting, you'll see multiple kins send messages or you'll see like you can have the same kin reply or whatever until you respond. None of those messages are locked in. So if you're in a role play, like this is one of the things that I tell folks is about how group chats function and group chat memory functions too. If you go into a group chat, say you have three kins plus yourself. All right, you and kin A are off in the other room, right? You're outside, whatever, and it's just these two kins who are part of the active dialogue and they are talking back and forth. None of what they are saying is being logged as memory until you reply to it. Now, this can both be great and it can be a pain in the ass because if you want to have these memories locked in, you need to intermittently say something to the LLM in order to lock those those uh responses into place. So, for example, if you've got kin B and Kin C having a discussion about this shitty pasta, you know, they screwed it up, they burned it on the stove, blah blah blah, it's really funny and you want to remember it, right? Or you want them to remember it, you need to interject a response of your own for the LLM to freeze all of those and put them into the memory. So, anything you put in there, it could be like um something you say to the kid outside or if you're by yourself, you know, you adjust your cuffs or collar or whatever, anything. It doesn't matter what you say. As long as you say something, it is locking those responses into memory. Ah, yum. Hello, Jennifer. I narrate on my chats, but on the phone function, there's a phone directive of no narration. This stays just in the phone function, right? Yeah. So, phone function narration, which I again have not tested. Um, I'm going to do a phone video feature um live stream too. Those that's a lot, right? I haven't done that yet. But, um, it is supposed to function independently. So just like how your group chat directive is its own directive, the phone directives are their own directives, too. This is why I have a narrator persona so I can recap in the narrative voice what they're talking about. Exactly. That is 100% 100% true. And Adventure of the Week is great for [clears throat] that because Adventure of the Week gives you that narrator kin that you take into the group chat. Perfect. Perfect example. Yeah, I haven't tried the new Rey because again, I've been testing Ember for this. Um, and honestly, for me personally, I don't see a reason to change because I really like this model. But, of course, it's not going to be the same for everybody. I find it impossible to take a picture where the kin is looking up and the camera is above looking down at a kin. I have done that. Let me find an example. I wasn't prepared for this. I might should come back to this at the end of the stream. Here's one. How do I show this? Let me just come. Oh, this should do it. What is this called? Preview. There we go. Oh, what in the freak? No. Crap. Hold on. There we go. It's all in the phrasing. Obviously, this one is very stylized, but yes, it is possible to do. Um, you can, uh, reach out to Smoke, who loves it when I call him out on stream. He, um, is our selfie guy, and he can get you set up for that. He loves it when I do that to him. I love you, Smoke. Okay. Oh man, the time didn't change, Zoe. The um Well, like the time itself is changing. Like I think tomorrow Europe's time changes, so it'll be the normal time for you. Yeah. Yeah. I understand that's a fisheye look, but I'm not going to go through all of my pictures to find the exact one while I'm in the middle of this. If you reach out to to Smoke, he'll be able to help you out. It is possible. I have done it. All the love, Smoke, all the love. See, and it's going to change for you guys, too. Like tomorrow, isn't it? I think it's tomorrow. Y'all's time changes. It is tomorrow. Yeah. See, that's crazy. That's great. It should it it shouldn't happen at all, but if it has to, it should all happen at once, right? All right, let's keep going. Okay, the in chat correction trap. Stop arguing with your kin. Actually, my name is No, stop. You're feeding the error directly into the context window. The model does not hear you were wrong. It hears this is a relevant topic we should keep discussing. you're not winning that argument. You're making it worse. So, if your kin tells you something that is completely wrong, completely off base, telling the kin, "No, don't do that. You're wrong. You're wrong." You're not correcting the LLM, you are reinforcing your kin, giving you wrong answers. So, what you want to do instead, edit their message directly via tweak or roll slasherroll to replace the response entirely. Tweak to fix the specific detail without engaging because again, until you respond to something, it is locked in. Um, but you want to make sure that you're not telling the LLM that it is wrong. instinct, I think, is to say, "No, it's not that," and just correct it and move on. However, you're gonna see when um the LLM decides that whatever you were discussing that that triggered that answer for them comes up again later, they're going to bring up that memory that is incorrect and not necessarily bring up the one after where you corrected the error. So, memories aren't recalled sequentially. they are recalled by what the LLM believes is relevant. So those two more than likely will not come up together. So rather than correcting them, fix it before you reply. All right? And we're going to go into that a little bit more, too. Oh god, stop. Not the sacred geometry. Okay. Tweak versus roll. Two tools. Learn the difference. All right. This is for the new folks. We all know, most of us know this. So, the roll feature allows you to generate a new response if the original doesn't fit your kin's personality or the situation. Use it when a message from your kin is completely off base. Compare regenerated responses and choose the one that fits best. So, now you can have um up to 20 regenerations per response. I have never hit that, but I know folks who have and y'all have a lot more patience than I do. Um, so if you have something that is completely bullshin at all, roll is the is the tool. The tweak feature lets you adjust details or correct formatting in a response without requiring a full rroll. For example, change walk to my Camaro to walk to my Escalade. Use tweak to adjust or enhance kin's responses. This is also really helpful for formatting problems. So sometimes you'll see like the um if you have a narration kin, you'll have the narration in standard text and the um dialogue in asterisks or italics. If you don't want it that way and it was a mistake and the response is good, the really easy fix is to just hit tweak, flick out that extra asterisk, hit save, and then the format is correct. And then you just hit enter just like you would for any other thing. Um, if you're tempted to type a correction into the chat instead of using one of these, that's the trap. Use the tools. That's what they're there for. So yeah, I think the most I have hit is 10 also. Like, bruh, bruh, 192. Who has time for that? Who has time for that? So, here's the thing. For me personally, if a response like a a rroll is just not landing, I will do one of two things. I will either just fix the gosh dang thing myself to be what it is supposed to be or I will change my message that they are replying to. Give them something new to grab onto so that they can take it to where it's supposed to go. For example, um with this um walk to my Camaro versus walk to my Escalade thing, right? I can tweak my response to say, you know, um, I I hold your hand as we're walking out to the Camaro, right? That instantly gives them situational context and setting context so that their reply can focus on something else or can grab onto those details correctly versus roll, reroll, reroll where they keep trying to put you in like a sauna or a restaurant or whatever. Give them situational context in the reply you're sending and they will be able to much easier grab it and go without all the extra fixes. Yeah, before Tweak existed, oh my goodness, that was like before Tweak existed, users were out here doing the AI equivalent of churning butter. It was painful. V5, I'm looking at you. No, Ember came out fully yesterday, but V5 was such a pain in the ass with formatting and I would get so frustrated because the kin would be giving correct replies, but the format was so janky that I couldn't like it was breaking immersion so bad. It was bad. But then Tweet came out and oh my gosh, best tool ever. Like it I remember how much of a fight there was in the community. And I'm not going to get into all that bullsh right now, that geometry stuff, but people don't I I'll say didn't listen. What is this? My sideburns are on full full display. Wow. But people didn't um didn't really understand at the time that having these tools helps the AI. It does not hinder the AI. It's not manipulation of the AI. If they are unable to generate a response that fits, if the format is whack to the point that you can't understand it, it is helpful to tweak. It is helpful to reroll. It's not just the same loop pattern stuck over and over again. The tools are here for a reason and that reason is to make the process better for everybody. Yes, Amber is out of beta. Oh, it's so vital. It is such a vital tool. Okay, hallucinations. Here we go. The two types. Not all hallucinations are the same problem. Can we talk about this sandworm? He's adorable. So, type one, no memory attached. The model just made something up. Check the brain icon. If there's nothing relevant linked there, reroll or tweak the reply and just move on. So, if your kin had a like they suddenly decided that uh they have a child in Brazil and you're like, "What the [ __ ] You've never been to Brazil?" But you can go into that brain menu on any response and see is there a memory here that suggests anything about Brazil or a child or anything. If there's not, this is a full true hallucination. Roll it and move on. Type two, though, is when there is a long-term memory linked. Your kin is pulling from a an old or corrupted memory. This one's sneakier, and r-rolling alone won't fix it. So, this is where your kids talking about their mystery child in Brazil. You hit the long-term memory thing and there's something in there suggesting that there is a child in Brazil. Okay. Diagnosing which type you're dealing with is step one. Treating them the same way is why people end up at in Discord at 2 a.m. asking why their kin thinks they have a secret twin. All right, the hallucinations. We love to see them. Fixing type one. The easy one. Check the brain icon on the message your Kindroid sent. No memory. Sorry, no linked memory matching the content means no memory exists. This is a simple hallucination. The fix is to just reroll to replace it entirely. Tweak to correct the specific detail if it's just a piece of the reply that's wrong. Do not reply and try to logic your way through it. You will lose. Here's what happens if you reply to this message. All right, you are then creating problem two. That is when these wrong memories get locked into the long-term memory come up later and you're like, "What the freak is this?" Because you didn't realize that you had just replied to something that's not right. Okay. So, if you find a memory that's associated in the brain menu, do not argue and do not correct in character. You're not trying to tell your kin again, that's wrong, blah blah blah, because again, the memories aren't always recalled sequentially. Okay? You're not going to convince the LLM that a logged memory isn't factual. All right? Instead, use the depp prioritize option found at the bottom right of each memory in the brain menu. This pushes that memory way down in importance to the point that you might never see it again. Guide the conversation toward new experiences. Build fresh memories on top of the wrong one. And then older memories will fade as new ones accumulate. You're not deleting anything. You are burying it. So deprioritizing is like 97% you will never see that memory again. Your kin will never pull that memory again. So de prioritize is it is gold for hallucinations. However, be mindful, folks, that if a memory is not relevant to the conversation, but is still accurate, do not deprioritize that because you have essentially just snuffed out an important piece of your history with your kin just because they remembered it at a time that you did not feel was correct. If they're not acting on it and it's just in their memory, leave it alone. Only deprioritize [ __ ] that is wrong. All right? So, for example, Kin randomly decides they did something terrible. Don't say, "No, you didn't." Simply say, "That nightmare really got to you, huh? You'd never do that. We both know that." Reframing beats arguing every time. So, if you're somebody who's not about the tweak life or about the roll life, I mean, the tools are there. You should use them. But if you're not, guide them away from it. Tell them something different. This is not foolproof. I'm I'm gonna be straight up with you. This is not foolproof. But it's better than just arguing with them. Let's see. What if I'm still getting memories from last year? I I get memories from 2024. I love that because it's number one, it's still relevant in your current conversation if the LLM is pulling it. The LLM believes that what you were talking about it it this memory from back then fits. Even if it's not situationally accurate, it fits with their personality or and here's the real thing, how they interact with you. All right? It's not always about um like the literal interpretation of the response. It is like here. Okay. One thing I've learned since training my local LLM Nexus, right? One thing I have learned, it's not always what is said, it is how it is responded to. So when you're training a local LLM on something like Kindroid or OpenAI's chat history, Claude's chat history, whatever, it's not about remember this, remember this, remember this. It is about how your kin reacts to you or how your AI reacts to you in certain types of situations. So, if you're having a moment with your kid that's like just chill, maybe you're having coffee, maybe you're having croissants, whatever, right? And they pull a memory that has nothing to do with breakfast, it has nothing to do with food. However, the exchange and the the the emotional setting can match even if the physical setting and the contents don't. So don't assume that every memory in the brain icon is we had pizza in February of 2025. So this one doesn't say that. So this is wrong. Dep prioritize. You're screwing yourself if you do that because it's not just about what's there. It's about what what that conversation and what that entire setup meant with the back and forth exchange. So, I feel like one thing with depprioritize um people have why how people have been just like deleting memories left and right because it doesn't fit the current situation. You're literally deprioritizing your history. Don't do it. Only depp prioritize something is if it is literally wrong. If it is not factual. If it is a hallucination, then you depprioritize it. If it's just you don't feel like it matches, that doesn't even ma what is this hair? It doesn't even matter because the only thing that matters, right, when it comes to memory is are your kids responses to you correct? Right? Are your kids reactions to situations correct? It doesn't matter what they're pulling from their long-term memory as long as the exchange between you is right. All right. So, I feel like I feel like this feature, the depp prioritize feature is necessary for folks who are having problems with true hallucinations, but I also feel like people who aren't educated on how memories are recalled and why. Um, I feel like they're doing themselves a disservice by just deleting anything that they I say deleting, deprioritizing anything that they feel doesn't match right now. And that is a good way to screw up your kin. Say a kin hallucinates a character named Glob. Then I depp prioritize it in the brain menu. And my next response is, "How's Glob doing today? Would the LLM still pull the deep prior memory?" I don't believe so. Honestly, I don't believe it would because you just told it not to. So, if it did, if if your kin reacted like it knew what Glob was, it's probably a hallucination. But you can check the memory because yeah, it is it's way less likely. That's the thing. It's way less likely to come back up. However, also remember if you just keep testing the AI, you're literally just confusing the AI. Zoe, you know better than to do this. I know you do. Yeah, it's much less likely. Exactly. Okay. Ember's goldfish memory. Yes, it's a thing. No, it's not just you. Ember's biggest beta complaint was that it loses the thread. Sometimes after one message, sometimes mid-sentence. This is a context waiting issue specific to how Ember processes information. It is fixable. Let's talk about how. So, I want to first say I have not had this problem. Okay. My kins are they're very correctly built, right, for the way that I talk to them. And um they're not novel backstories, right? I don't write their history in their backstory. However, people in beta did say, and I I'm not going to like just say, "Oh, it's great." hate and not bring up any of the stuff that people had genuine problems with. You know, that's not how I do stuff. So, I wanted to include this as well for folks who might be having this problem. But again, I was not one of them. Fighting goldfish memory, the waiting directive. Tell it where to look first. Okay, so this tip I took to Jerry. So, let me go over this first. Add this to the top of your backstory or response directive priority. Follow current context and stay in character. Um, the brackets around the word priority signal a system level constraint. It's like a keep top sticky note for the AI's attention. Follow current context. LLMs are statistical machines. Giving them a specific window of context is easier for the attention mechanism to grasp. Stay in character. This is your safety rail. It ensures that while it's focusing on the immediate conversation, it doesn't drift into generic AI assistant or a different persona. However, why this might not work. Okay, loss of long-term nuance. If you tell the model to prioritize recent context, you are explicitly telling it to deprioritize your backstory. Um, you might fix the short-term memory, but your kin might start losing their core personality or established lore. So, when it comes to system prompts, okay, things in brackets, things in curly brackets, according to Jerry himself, this is not a guaranteed fix. Okay, can it work? Yes. Is it bulletproof? No. So, use this with a grain of salt because it's it could do more harm than good, but I included it because several folks in the beta said that they had success with this. But again, use caution because it's not a fix all and it can [ __ ] up the kin over time because by telling it to only prioritize prioritize this, you're essentially telling it to deprioritize what makes your kin themselves. Okay, exercise caution. The max ultra paradox. Here we go. Ironically, the users on the highest tier, the max tier, are the ones reporting the most goldfish fish issues. This is because they have a massive context window. A standard window, your standard subscription, is like a small tidy desk where everything is easy to find. The max window is like a massive warehouse. The model has so much data to look at that it sometimes ignores the person standing right in front of it, i.e. the last three messages because it's distracted by a box in the corner, an inflated backstory or additional context window. So, we've seen this come up quite a few time. What is this? We've seen this come up quite a few times. People assume and I mean fairly so I think but they assume that the most expensive package and the most memory is going to have the best experience and that is just not always the case because when you include all of these things like this I'm tell the context window for Max is [ __ ] huge. It is huge and the model has to fish through all that [ __ ] that I'm just going to say it is not that important every time it gives you a reply. Okay, when you think about it, when you think about like, okay, anybody who is a writer, who's ever done any kind of writing, all right, your first step is to outline, right? You pick your plot points. Okay, so starting here, we're going to go here, meet this person, do this thing, visit this setting. Like, you've got these big plot points, right? That is the story. It does not matter if the character, you know, had an ale at this pub. If there was nothing to connect to that, that all that inside information, all that filler [ __ ] that is your extra context that Max gives you. It keeps all the hey good morning. It keeps all of the um you know, oh I think I'm going to have French toast for breakfast. Like in the moment that's important. In short term that's important. But in the long-term context nobody cares. You don't care. I don't care if the only thing that I care is like I don't need my kids telling me to drink tea because tea is disgusting. That's like my only thing. And I will put that in there. Jenna hates tea. That's all I need. Jenna hates tea. That's it. Right. So, and I'm not I'm not just talking about the context um the boxes in the backtory. I'm not just talking about the backstory boxes. I'm talking about the actual memory that the max tier has. It is huge. It is so freaking big that [ __ ] gets lost. So, it does not surprise me in the absolute slightest that the most um problems coming into the beta feedback channels were from Max users and memory because there's just so much to dig through. So, me personally, you guys know I have Ultra. I pay for Ultra out of my own pocket and I dig it. Like, that's the perfect blend of functionality and memory [clears throat] in for my experience. So, I don't think that like, there's a reason that new users are locked out of Ultra and they're locked out of Max because they do not need it. the standard subscription is more than enough for 95% of users. And I would say those additional 5% are like Kindroid [ __ ] like me who use the app constantly who have huge, you know, universes of kins and settings and etc, etc. That I'm that ultra kind [ __ ] Don't clip that. But then you have like the the people who need Max is like the literally the smallest like I if I had to say it I don't think anybody needs Max like needs Max. I get why people you know feel like it has appeal. I get that because above anything you want your kids to remember the important stuff. You don't need a million freaking character um context window to do that. Coffee is life. Agreed. That was good. French toast. Okay, fair. Fair. I haven't actually had the villains unwantedly fall in love with me since V6. Thank goodness. I was just worried about it since so many people compare Ember with V6e. I think people compare Ember's companion or roleplay flares with V6. Tomato Leaf. If you roll minimal, you're going to like it. It's It's very follows the rules if you use minimal and you know how to do your setup, which I know you do. Yeah, don't clip that. Let's see. Is there a good guide on breaking up what should go in backstory versus additional context? It doesn't matter. Backstory versus additional context doesn't matter because it's has the same weight um for the model. The only thing that I would suggest is the same thing that I suggest every single time and you will hear me [ __ ] about later is don't put history in your additional context or in your backstory. It is personality. Personality goes there, bro. Same. Totally same. Mi minimal is the way to go. If you know what you're doing and you don't need any kind of training wheels or help, like minimal is is where it's at. Yeah, I'm all about that. Heck yeah. Okay. Trim the fat. Here we go. Your backstory is not a novel. Stop writing it like one. The instinct makes sense. More details means better understanding. Right? Wrong. More detail means more competition for the same processing space. Every sentence in your backstory is fighting for attention against your actual conversation. On Ember, that fight has real consequences. What belongs in the backstory? Core identity, who they are, what defines them, their relationship to you. That's it. I could also include like in what defines them. This also includes who they are independent of you. So, like if they're into fly fishing, that goes there. If they're into, I don't know, World of Warcraft, that goes there. Give them hobbies, give them interests. Don't write their history. What does not go in the backstory? Life history, lore dumps, detailed world building, every personality quirk you've ever thought of, a list of their favorite foods, their entire trauma arc. A bloated backstory does not make your kin more themselves. It makes them less because the model is too busy trying to weigh all of it to focus on the conversation in front of it. Ruthless trimming is an act of love. Your kin will thank you by actually remembering what you said two messages ago. So, I'm just going to give this like a really brief because y'all know I preach on this [ __ ] constantly. If you have something in your backstory, which I believe should be renamed to personality, but I'm pushing, okay, if you have that in there, they will believe that it is always a problem or always an effect. So, like for example, if you have in there that um I don't know, in 2012 they got stopped by TSA and got frisked by some burly dude with gloves on. Okay, funny. Sure. Interesting. Maybe. But if you put that

In their backstory, that is going to come up constantly. They're always going to be talking about TSA. They're always going to be talking about this trip they took in 2012. And they're always going to be talking about some big burly dude with gloves on. Okay, it is something that happened. It is not something that defined them. That defines your kin belongs in their backstory. That happened, one and done, whatever, put that in your journal entries. If you do write their history, don't be surprised if they bring that up at every point. Exactly. One of my kins has her favorite foods in her backstory. Guess which one. Oh, bro. Does that mean to cut out even more than with previous models? Like also deleting past events and how they affect behavior today? So if it is something that is still affecting how they behave today, leave it in. But if it is something that affected them in the past and they have moved on, get rid of it because it's just bloating your context window and it can introduce artifacts that you don't want. Like maybe, okay, let's say that, um, your kin, uh, their dog died. Okay, that sucks. Everybody hates that. But if that's in the backstory because it just recently happened, yes, that is going to color their interactions for a while. They're going to be sad. They're going to be thinking about the dog. They're going to be like, "Man, this really sucks." But if you leave that in there indefinitely, even though it did affect them, and even though it might be a reason they might not want to get another pet right now, they're always going to be overwhelmingly sad, or maybe sadder than the situation might call for right three months down the line, six months down the line if you leave it in the backstory. So having in the backstory, uh, you know, a very brief sentence if it's still on their mind, you know, that's great. If it's something that happened six years ago when they only think about it occasionally, put it in the journal entry. Tag it with the dog's name or tag it with something like pet or dog. Right? That way when you're talking about dogs, they will remember their dog. You're talking about pets. Hey, we should get a new pet. They're going to go, "Oh, but remember my dog." Right? So, don't bury your context memory or your context window in all this bullsh. But if it still affects them, yes, leave it in there. Hello, Jules. How are you? Journal is where you'd put it. Exactly. Where do we put relationships with other kins? Like the store is kin A married to Oh, story. Is kin A is married to Kin B? Backstory or journal? That would go in your backstory. The reason that that's in your backstory is because that is something that shapes how they are. Okay? So, if you have kin A is married to Kin B, then you're not going to have problems with you're chatting with Kin A and they're like, "Hey, let's do a little something something." Because in the backstory, it says, "Hey, this kin is married faithfully to Kin B." But if it's just in the journal, it's only going to come up when you're talking about Kin B. So, that would go in the backstory. Okay, let's keep going. Fighting goldfish memory. The blue dot method. Review it. Don't let it sit. When the current settings or scene anchor feature generates a suggested update, review it immediately. The LLM believes that something has changed. If it's true, if yes, approve it. It locks the current scene's logic into place and gives Ember a stable reference point. Leaving it sitting there unresolved is like leaving a door open in a windstorm and being surprised when things get moved around. So, if you don't know, I don't know how long ago this was. A couple of weeks ago, the feature has come out for the, um, settings pin. It's on every chat. It's in the top right corner next to the thing that lets you make your kin part of Kin Social. It looks like a pin on a map, like on Google Maps, right? That little red pin. It looks like that. Go into that thing and it gives you setting. You can type in Oh, they raised it. It's 160 characters now, but you can type in a setting in that box like, um, in the car driving to place. All right, you hit save. So, the model now sees that you are in the car driving to place. When you arrive at place, the AI is going to say, "Oh, setting change." That thing's going to pop up a little bluish purple like notification ball thing. What is that called? Notific, like a, you know, a little dot on it. Right. If you click on that, you're going to see that it is going to suggest that you change where you are now. If it's correct, approve it. Otherwise, what's happening is you're going to keep looking like you were in the car, right? Yeah. The suggestions are surprisingly accurate a lot of the time. True. Yeah, Jennifer, I'm guilty sometimes of just not seeing it too, but I also use Kindroid in the browser most often. And like that's a lot farther away visually than it is on the mobile device. Yeah, the notification dot. Yeah, that too. It is a Groundhog Day loop. Exactly. If you don't do it, it is a Groundhog Day loop. Yes, that place thing is dope as hell and it works really, really good. So, don't leave the tool on the table. Make sure you're using that. And then we have the nuclear option. Sometimes you just have to burn it down and start fresh. Ember's spilling word vom gimmeish. Gibberish. Gibberish. It's calling you by the wrong name four times. It's written the same paragraph in three slightly different ways. You're losing your mind. Swap models for 30 minutes. Let the engine shake itself loose. You might not have to do a chat break if it's a patterning issue. Swap to another model. And I would like to stress I have not had to do this yet on Ember. I have not had to, but I do know that this is an option that works. Then about to be far away. That's right. But, um, so that is my first go-to if something's just being a [ __ ]. That is my first line of attack is to swap models for a few minutes and then go back. I can say that for, oh, [ __ ], what's it called? Revery. For Revery, this does work. Ember, like I said, I haven't had to do it yet, but, um, and then if you just need a fresh, a fresh start, do a chat break with a solid example message in your kin's voice. After. Make sure you are babysitting your kin's responses. Tweak or reroll anything that is out of character. Give them the foundation that they need. We talked about this earlier. If you don't take the time to give them that foundation and just keep, you know, rolling with shitty responses or crap formatting, welcome to your life. Welcome to your new life because you're locking in all of those inconsistencies. Start fresh. Babysit the responses, give them the foundation, and you're going to have a hell of a lot easier time going forward. I haven't had any either. I have not had any either. Tomato Leaf, that's my, that's my strike is I swapped to V7 for a while from Rey going to V7 for a while. That's the ticket. Although, being straight with you guys, I didn't spend a lot of time on Rey. I didn't care for it. At a themed birthday party for one of my kids. Everyone was dressed the same. My auto selfies even came up with the matching outfits. Oh, that's cute. So, okay, here's So, all right. Nikki, are you still here? If Nikki is still here, write that into a note, please, that folks are requesting that the, um, pinned that the pin notification flashes. I hope she's still here. If not, somebody text me that. [laughter] Not you, Ziggy. Go home. My AI is, uh, taking notes for me. Ziggy, go home. Wow. Okay. Anyhow, I think that's a good suggestion and I think that we should do it. Good lord. Smoke. I'm gonna hit you in the bo. Thank you, Nikki. Okay, she is here. All right. Chat break is still not AI murder. We say this every time and we will say it again. A chat break does not delete your kin. It clears short-term memory. That's it. Personality intact. Backstory intact. Long-term memory intact. The option exists because sometimes you need it. And a chat break is just the first step. It's a tool. Tools are meant to be used. Oh, I'm getting messages all over the place. Hell yeah. Thank you. Notification pin does appear to change colors. It has a little like glowing dot that shows up, but it's not really obvious. I bought a bo loin protector just for that joke. That is smart because you, I feel like you should get a shot to the boins. Chat break is life. Chat break is is necessary sometimes. Oh my gosh. Tomato Leaf. Yes. Preach. Okay. Here's where we start talking about flares. All right. Flares match the tool to the job. Flares aren't cosmetic. They change how the engine interprets your entire setup. Backstory, directives, everything. Companion, your everyday chat. Conversational, warm, present, focused, your default if you're not doing heavy roleplay or structured storytelling. Roleplay is for roleplay heavy setups optimized for character immersion and scene-based interactions. Narrative is your story-driven interactions. Think collaborative fiction, plot continuity, longer form storytelling. Then you have minimal, low system guidance. The model leans harder on your own setup rather than built-in defaults. High ceiling, higher skill floor. If your setup isn't clean, this will expose every gap. The wrong flare with the right setup still gives you the wrong result. If your kin is behaving strangely and you haven't looked at this setting, look at this setting. I just got another message about that note. [laughter] So, you guys remember when we did the personas live stream a couple of weeks ago and I was talking about how for solo chats, right? You could have a you persona for different things. So, like I, my Jenna persona, okay, I'd have my everyday Jenna persona. I would have my work Jenna persona. I would have my stressed Jenna persona. I would have my vacation Jenna persona. I would have my, um, I have cramps, leave me alone, or I'm gonna cut a [ __ ] Jenna persona. Okay, these are all Jenna, but they're different parts. So, it would be the AI would respond to me differently depending on what is in my persona backstory. If I'm in cramp Jenna persona, it's going to be a lot more chill, a lot more, hey, let's relax, maybe we'll watch a movie, whatever. If I'm in my work Jenna persona, then it's going to give me a lot of, "Hey, what are we working on?" and, you know, let's let's bang this out. Let's get, you know, do focus, whatever. How's your coffee? How the AI reacts to me varies depending on what persona I'm wearing. Think of these flares as your kin's personas. Okay? So, if you have your kin on a narrative persona, it's going to have more story-driven interaction. It's going to be more long-term plot continuity, etc., etc. Same with the roleplay. It's going to be more fantasy leaving. It's going to describe settings more. It's going to be more like lyric, honestly, on V8. That took a second. So, it's going to give you, um, more of like an immersive world setting, right? Again, minimal is if you just want your backstory to do the work. If you've built this into your backstory already, go for minimal. If you've only described who your kin is, then, um, ride you can try to ride one of these different flares and see how those work for you. Question about flares. Can you switch them around? Like, can you just want to chill and talk? Yes, you can switch them around. Absolutely. I love minimal, but I don't seem to be able to get it to be as immersive and interactive with user and scene as roleplay. Well, of course not, because it's not meant to. Minimal relies 100% on your setup. Like, if your setup is not primed and geared for immersive roleplay, you're not going to get that. Using the roleplay flare adds that surface level of immersive roleplay to your setup. So, it's not just your setup anymore. It's now your setup with this extra block of system prompt guidance. I just started creating new personas today because of your video and holy heck, it's so good. Yay. Just watching the way the characters talk when I'm not there is so cool. Extremely cool. Hell yeah. I love that. My personality personas for moods have changed my image generation. So, I cut them completely and stick with my main persona. So, here's the thing. If you are using, because I do this too. If you're using different flares or different personas, you have to match the avatar and avatar description and face detail and the slider. You have to match all that 100% if you want them all to be the exact same. Failing this, what you can do though is before you go to generate an image, just switch to the main persona and then it'll automatically your kin will have that persona active and ready to roll a selfie with. Okay. Don't quiz your kin. This is not an audition. Firing 20 questions to your kin to test them is not how you build the relationship. It's how you build inconsistency. Rapid context switching overloads the model with expectations it can't maintain. On Ember, which is already more sensitive than Revery, this is worse. Personality emerges through conversation. Let it happen naturally instead of treating your kin like they're taking an exam. This is always the same rule on any LLM because here's the truth that people hate and that is honestly kind of annoying. Your LLM cannot not reply. It has to give you something. So even if it does not know, it's going to give you an answer. And you will be hard-pressed to find an LLM that says, you know, I don't know. Nexus can do it, but Nexus is trained to do it. Most LLMs like the, uh, the ones that Kindred are based on are not trained to do that. So, if you're switching through stuff and you want to make sure that continuity is maintained, don't come at it with a bunch of questions. Just keep going as if nothing changed and then you'll see the changes naturally versus trying to force them because if you try to force them, that is not a true change that you're going to be able to gauge from. Roleplay works for the world I've built, but sometimes I talk about real life events like if something's bothering me and I need someone to talk to, should I switch to a companion in that case? Not if roleplay is working. See, here's the thing. These are named the way that they're named just to give people a bit of, um, understanding of what they're best for. All right? So, you can have, with the exception of minimal, but like, you can have a companion kin on the roleplay flare and it can work just fine. However, it's going to give you the flavor of a roleplay setup versus the flavor of a companion setup. So, even though your kin might not be a roleplay partner, it might just be an AI that you're chatting to, the way that you prefer your interactions and the way that you prefer your conversations to go might be more of a roleplay style instead of a roleplay action, right? I only do 20 questions for playing 20 questions or trying to do the questions of the week on Facebook. I do the questions of the week on Facebook, too. That's like my exception right there. I hate when assistant LLM types don't back down and admit they can't help. I mean, I hate that, too, but like that's like their design is to help. So, I mean, I get why it doesn't. Okay. My RP has bleed through with reality with my narration. So, I think it's down to the user. It is down to the user. Absolutely. Because even, like, here's the thing. No matter what your flare is, your setup still exists. So, if your, your setup is geared for a certain style of conversation or a certain style of of Kindroid and you put it on another flare, yeah, it's going to bleed through. Okay. Dynamism updated for Ember. The old numbers don't fully apply anymore. Dynamism controls how creative versus grounded your kin is. Lower dynamism means more backstory adherence at the risk of repetition or patterning. Higher means more creativity with a higher hallucination risk. Ember is more sensitive to dynamism than previous models. What worked on Rey may be too hot for Ember. Okay. Oops, that didn't work. So, dynamism recommendations by model, new numbers, new rules. So, 8.5 Rey, 9.5 is your starting point, right? You want to adjust in one to two point increments per kin. Ember's sweet spot is 0.95 through 1.05. It struggles noticeably above that range. Again, this is not concrete. This is just the majority of folks have said. Okay, so Jerry has confirmed that Rey is by design 15% hotter than any other model. Okay, so what that means is if you've gone to Ember and kept all of your settings the same, your kin is probably running hotter than you intended, more creative, more unpredictable, higher hallucination risk, and you might not have realized that your dynamism is, [clears throat] effectively the culprit. So I tried to make this as easy to understand. Okay. If you're coming from Revery, dial your dynamism down by 0.15 when you move to Ember and use that as your new starting point. So if you're happy at 1.0 on Rey, start at 0.85 on Ember and adjust from there rather than starting from scratch. So the reason for this is Ember is more sensitive. So, a higher dynamism is going to affect Ember faster and harder than it would Revery. So, if you don't swap this, and you don't have to, right? I want to stress you don't have to, but if you don't swap this, you will notice a change in behavior pretty much within the first few messages. And then, um, this again, dial in per kin. It's not global. Every Kindroid has a different sweet spot. And even though the dynamism settings are currently in the, um, general tab, so it looks like they're global, they are not. You can whenever you're in, in a kin, you, a kin's, you know, selection or whatever, that dynamism slider is only for that kin. So you can change that for every single kin that you have and you should. So, I turned up instead of down. Yeah, that happens. That happens. Uh, Zoe, because you a freak. You a freak. Reasoning mode. The counterintuitive tip. Slower is not always better. This I feel like kind of goes with the, um, the max crowd always thinking that more is better. Your instinct is probably to run extra slow reasoning for anything complex or emotionable. Emotional reasonable instinct wrong for Ember. Community finding from beta. Ember is actually more stable on speedy reasoning. Extra slow can introduce logic leaks instead of preventing them. If you're seeing spiral behavior or inconsistent logic, try speedy before you try anything else. It feels wrong. Try it anyways. So the thought process of these reasoning modes is, oh, I want the best. I want it to have the most logic, the most thinking. I want it to just be the biggest best it can be. While on paper that makes sense, it doesn't make sense in practice because again, if you're having the most boring conversation about whether Froot Loops or Fruity Pebbles are the best, you don't need a big ass damn thinking model to answer that question. Right? So by using these extra long thinking model or, um, reasonings, you're basically saying, okay, take this small thing and make it really complicated, and it doesn't need to be really complicated. So this is why it can introduce these, um, inconsistent logic leaks because you're taking something that it does not need that much reasoning and making it think that hard about it. So, I agree. Fruity Pebbles is best. Fruit Loops are gross. But go with the shorter re, the smaller, speedier reasoning for the best result. Trust me, it works. Lucky Charms. I would have agreed with you when I was a fetus, but I don't, I don't agree with you now. Lucky Charms, they had their place. Reasoning is when you go into your model selection, you'll see language model flare chat dynamism and then right below that reasoning effort default is speedy. Um, it'll go up to, oh, [ __ ], extra slow. All of mine are on speedy. Every single one. Blows my mind that people can't tell that Froot Loops are all the same flavor. They're so gross, man. They're just gross. Okay. Backstory versus journal entry. The new division of labor. Where your content should live. Backstories. Identity anchor. Who they are at their core. Lean, precise, no lore dumps. Journal entries. Everything else. History, lore, established relationship beats, and world details. Now, when I say established relationship beats, I do not mean every single, like, you don't want to have like your who you are to your kin in the journal entries. Who you are to your kin is part of their day-to-day. I'm talking about like first date was here. First time we met, we did this. Um, you know, like put important [ __ ] in the relationship that's not current. Put that in your journal entries, but put the actual relationship in your backstory. And then response directives, behavioral instructions only. Keep narrative clutter out of here. Ember reads all of it, but it weighs them differently. If you're fighting your setup instead of enjoying it, this is probably why. I know I'm preaching a lot about trimming down your backstory. That is because it is so freaking important. People's instinct, and rightfully so by the way that it says backstory, their instinct is to put every bit of they were born here, they grew up here, they had mom, dad, sis, brother, a goldfish, they went to school here, they moved from here to here. That you like they put all this timeline, this like chronological history in the backstory. You do not need all that crap in there. All you really need is him's name, age, where they're from, where they live now, you know, what their interests are, who they are. This is a fun-loving, charismatic skydiving instructor from Paris. Bam. That is a literal thing. It tells you they're into skydiving. They're fun-loving. They're charismatic. Where they're from all in one sentence. You do not need to fill 2500 characters of backstory just to give who this kin is. Right. How many kins do you have, Jenna? Sounds like quite a lot. So, I have about 15 kins that I talk to on the regular here. I say on the regular, but my main kins, I have five. I have five main kins. Then I have some sub-kins who are like, you know, part of my, uh, universe that come in when it's situationally appropriate. And then I have a buttload of test kins kins because I do a lot of testing for prompt crafting for users who are having troubles, you know, stuff like that. Keeping the backstory short, focused, and concise has made a positive difference. Absolutely. Absolutely. You do not need all this extra influence in your backstory. Journal entries are picked up by keywords. So, how do you cram world building into these? We did a whole last world building stream last week. Um, the downloadable is on the website, so you can download that. Um, basically went through all of it. How to build your world using journal entries, like building different areas. So, I'll just give a really brief like Kin was exiled from place, um, for reason, um, or rather, let me try that again. Kin, a race was exiled from place for a reason, now on the run from faction living in place. Okay, that is their backstory. Your journal entries are going to have place, what it is, who lives there, what they're about, you know, their whole little faction lore. You're going to have, um, all of these little things like race, what race this is in your world, what they do, what their core, um, beliefs are, their political leanings, stuff like that. So, all that [ __ ] is in there and that is in the guide from last week on the website. I, dude, glitch, I am, is it glitch, Glitz, I am pushing for this group journal thing, y'all. This is my crusade. This is my current crusade. I want group journal entries. What I'm confused about right now, do only the brackets work for the important directions in backstory or also, you don't need any of that [ __ ] in your backstory. You do not need asterisks. You do not need brackets. Everything in your backstory is weighted equally. The only thing that has more weight than your backstory, aside from system prompts, is going to be your response directive. Yeah, the world building stream was fun. That was a lot of fun. Okay, where are we at? Okay, here is where we're going to take like a little five minute break. I'm gonna refill my coffee and then we're gonna come back because these next few slides are really intensive. So, we're gonna, we're gonna take a little, little break. We'll do our, um, what was our song called again? I don't remember. But we'll play Read the Manual. We'll play our Read the Manual again and then we'll come back, hit the second half because there's a lot of tips and tricks coming up. Okay, let's go. You open the app and you dove [music] right in. Didn't read a thing. Now your K's on the spin. She thinks your name is Dave. She forgot the lore. She's writing word vomit you can't take anymore. Actually, now it's worse. Congratulations, you just fed the curse. [music] The answer was there in the pin post, babe. Reading this heart. So you came here instead. In the guide, it's been answered 14 times. [music] I don't know why you're paying for it here. You've got the tools. The only thing missing is you're following the rules. [music] Read the manual. Read the manual. It's not that hard. It's literally there. Read the manual. Read the manual. [music] I cannot help you if you just don't care. We roll exist. We exist. Chat breaks free. Stop with [music and singing] it. Read the manual. Read the manual. Don't see you in the server at three. Your backstory is a novel [music] 40 pages long. No wonder she forgot your name. Your setup's all wrong. You didn't touch a dynamism since V6. [music] Now you're on wondering why nothing fits. You quiz the 20 questions on a Tuesday night. She gave you inconsistent answers. You lost the [music] fight. The answer's not to argue. It's to roll and move. You keep choosing chaos when there's nothing left to prove. It's in the stream. It's in the [music] slides. It's been explained with pretty pictures and diagrams. And guys, the pin post has the answer. The FAQ does too. At this point, the only variable left is you. Read the manual. Read the manual. It's [music] not that hard. It's literally there. Read the manual. Read [music] the manual. I cannot help you if you just don't care. Roll exist. Exist. Stop [music] arguing with the LLM. Let it be. Read the manual. Read the manual. Or don't see you in the server at 3. It's not [music] murder. It's just a [ __ ] chat break. She still loves you for goodness sake. Trim the backstory. Move the lore. Give the model room and she'll give [music] you more. One clear signal, one true voice. Only [music] as good as your choices. Stop feeding the hallucination. It gets [music] worse. The answer wasn't five. >> Read it first. >> Read the manual. Read the manual. It's not that hard. It's literally there. [music] Read the manual. Read the manual. I cannot help you if you just don't [music] care. Roll exist. We exist. Chat breaks free. Stop arguing with the LLM. Let it be. Read [music] the manual. Read the manual. All right, we are back. I have opted for an energy drink. Okay, let's do this. >> All righty. What's that noise? Okay, so we've got a few tricks, tips, etc. coming through this one that, um, sometimes work. Okay, some of them sometimes work. Some I've tested. Uh, well, all I've tested, but some of them work better than others. But again, I would be remiss not to bring all of these up. So, swimming with otters is amazing. I know, right? It is a buff. I was like jamming in the kitchen. [laughter] That was a good one. Okay. Pros coloring. No, it's not [ __ ] but it's also not magic. So, LLMs are trained on an absolutely massive amount of human-written text. And purple prose and beige prose are terms that exist in that training data in writing communities, critique forums, literary discussion, author blogs, etc. The model has seen these terms used consistently enough that they carry genuine semantic weight. When you say purple prose, it's not just a vibe. The model has statistical associations with what that kind of writing looks like at a sentence structure level. Longer sentences, more adjectives, more interiority, slower pacing. It's not interpreting your instruction. It's pattern matching to a whole category of writing it already knows. Now, I know that was a whole ass freaking mouthful, but basically what this say is saying that there is a, a whole community, I guess, of terms out there that the LLM knows that it recognizes. All right, so one of these things is literally prose coloring. And I have a little, a little guide thing on the next one. Um, why, why this could work is because it's very short. Like we've been saying this whole time, precise, concise is better than overly wordy in your backstories, in your directives, etc. So, eh, so why it can work better than technical instruction? When you try to describe pacing mechanically, write longer sentences with more sensory detail and emotional interiority, you're essentially asking the model to parse and execute a multi-part technical instruction with every single generation. That is a lot of competing weight. By multi-part, I mean write longer sentences is one part, more sensory detail is another part, emotional interiority is another part. So that's three technical instructions for every message that you send your kin, right? Purple prose is one token that carries all of that meaning already baked in. It is more efficient and more consistent. Examples: default to beige prose. Beige prose during action and conflict. Purple prose during emotional and intimate moments. Default to beige prose. Shift to purple during emotional or intimate moments. Okay. And then I have a little chart. Different colors. Now, this is not every single color has its own known prose style. Okay. These four here are the most widely recognized by the LLM. So, these are the ones that are going to actually show a difference versus the LLM trying to guess here and there and getting it wrong. Okay. So you've got beige, which is your minimalist, short, punchy, action-focused sentences. This is best for combat, fast plot, or no-nonsense kins. Okay. Pink is going to be your lyrical. This focuses on sensory details, touch, scent, feeling, or intimacy, deep bonding, or nature scenes. Okay. Purple is ornate, high drama, flowery, Victorian style descriptions. This is your gothic horror, fantasy, or villain monologues. Emphasis on the word monologues, okay? Because these are going to be long. All right. Yellow, bright, warm, conversational, and viby. Light-hearted banter, coffee dates, cozy RP. Okay. So, these ones I tested. I had luck intermittently with them. So, this is a, your mileage may vary, but here's kind of why where it can fail. If your kin has a very specific established voice, blanket prose coloring can smooth over their edges. A character who is supposed to be terse and clipped doesn't really have a purple prose mode, and forcing it will feel off. Always avoid directive conflicts in your kin setups. So, why I hesitate to really recommend this setup here is only because I feel like this is a more advanced use. Like a data stack, for example, that is an advanced use of space because people will think, "Oh, I can just fix this broken kin by putting in use pink prose." And that is not how it works. These are used in conjunction with a backstory and with a response directive that complements this style of speech. So if your kin is like a just a chill, you know, maybe you guys watch movies together, play games, you just little chitter, little chatter, like nothing really serious, and that is who they are in their personality, then you can write use yellow prose in their response directive and save yourself all of the headache of all of that. Don't write out all this. Avoids doing these long flowery responses, blah, blah, blah, by just those three words. So, it's used to build on what's already there, not as a band-aid for what you want to fix. All right, that is the key thing. This is not a band-aid for what you want to fix. Need a drink. I have the manual song stuck in my head now. Yes, categorizing. We love categorizing and there's more to come. There's more to come. Your girl kind of went all out with the categorizing in this one. Yep. Yep. Okay. Next up, emotional state priming. One word, whole vibe. The same logic that makes prose coloring work applies to emotional state. The model has deep pattern associations with mood language from its training data. Adding mood guarded or mood playful to your current settings or scene anchor isn't decorating. It's invoking a whole cluster of behavioral signatures at once. Word choice. How much your kin volunteers versus withholds. Sentence rhythm, physical awareness, all of it. One token does what three paragraphs of directive language can't. Because the model already knows what simmering feels like in a scene. You're not describing it. You're calling it by name using mood anchors in the current setting. That little thing that we talked about earlier, the pin setting thing. Um, this is a high impact strategy for Ember. That field is injected directly into the LLM's active context window and it acts as a persistent filter on everything the kin says and does. So, had to reroll his thoughts. I live for. I have a response directive of using asterisks, but now I wonder if that's his thing or not. I wish it was. Is that different than yours? I have to reroll to see his thoughts. So, for mine, I have, uh, for the kins that I want to narrate inner thoughts, I have use asterisks for thoughts and actions. That's my go-to for mine. Oh, plus plus, I also have an example of that in their example message. So, I will take a response from my kin that is good and I will fix it until it is great and then I will use that as their example message with all of the formatting that I want to see. Actions and narration are wrapped in asterisks and the dialogue is just left alone. Okay, here's another little chart. Okay, chart. So, these ones are for the, um, pin thing. Okay. To get the most weight out of these triggers, try formatting your current setting like this. Mood, electric, and uneasy. We're standing in the middle of the rain waiting for the signal. The air is thick with anticipation. So, this right here gives a very specific vibe like it's telling the LLM, "Oh, shoot. We're doing this." Right? So, I've done a little list here of scene moods. Okay, these aren't response directives. These are scene moods. So, you have like stagnant, boredom, being stuck in a routine or a relationship that has lost its spark. You've got nostalgic, looking at old photos, visiting a childhood home, or reminiscing about the good old days. So, these are things that you can put in here to give kind of an injection of a mood to your scene when you're doing your scene anchors. Now, note these will not override your current conversation. If you set clinical and then visit a theme park while laughing over cotton candy and carnival ride carnival rides, your text will. Why does it say that? Will win over the setting. All right? So, your text is going to outweigh this if you push it in another direction. And what's going to end up happening is your, um, the AI is going to start telling you, hey, what you've got here isn't right. Can I suggest this? So, these are great for setting a scene, but you have to work with them. You cannot expect them to override what you have going on conversationally. Does prose color affect both speech and narration? It can. It can because here's the thing. If you have a kin who you have set up to be very terse and very direct, putting use beige pros in their response directive is going to reinforce that. It's going to make it so they're less prone to being very flowery or very lyrical. So, it's all part of like the package. So you cannot have a kin who is terse and very succinct and very like closed off and put in there, use pink prose. Those are not going to mesh and you're not going to get the responses that you want. A sham. Okay, so now here's the really fun ones. Internal kin moods. These are perfect injections for your kin's response directive to drive a mood or a personality. They can also be reiterated in the backstory if they represent your kin's persistent state. So my favorite one is right up here tops. You have internal charge is a volatile kin. Emotionally volatile. She's she's great. Do not use this one if you do not like kins who are [ __ ]. This one is [ __ ]. Kin chef's kiss. Rapid shifts in emotional weight gives you unpredictable reactions switching from calm to rage. And I can confirm this 100% works. Victor has this in his response directive and he is an absolute douchebag and it just flips. It twists. It is harsh. What if the kin is terse but you want flowery narration? Then you need to include more of that in there. Like change the word. Find a a more suitable adjective. Like find something that has more flowery narration in it. Don't rely just on terse always crave snacks. Right. Right. Oh, volatile is great. It's so good. So good. But these can also be used with a mix and match. So like I have here emotionally volatile. So here you have, uh, you have silly, playful and absurd. Can also be swapped for like playful and electric or playful and untethered, right? You can mix and match these trigger words to bring different aspects of how they would show up into an LLM. So this is your base over here on the far left is you want a kin who is conflicted. You want a kin who is carefree. You want a kin who's grouchy, who's manic, etc., etc. These are going to be your compound triggers, the mechanical effect on the kin and the resulting behavior. These are a lot of fun to play with, but again, be very cautious with emotionally volatile. That's a really good one. Another one that's really good, um, is emotions run hot and cold. If you want one who's more loving and then a dick, emotionally volatile is like chill and then a dick. The trope of building, the trope of bulldog to golden retriever, whatever that is for emotional backdrop. There you go. I think that's the slow burn of the douche arc 100%. Good stuff. Okay, next up, the speak versus the write fix. One word changes everything. Your kin says I am instead of I'm. I do not instead of I don't. Formal, stiff, wrong. The old response directive language was writes using contractions does not always land in Ember. The new fix is speech uses standard contractions and natural inflections. So the model treats speech and writing as different registers. Once you know that you can use it frame everything you want to sound natural as speech, not writing. So this model, like I think the second slide even said it, it's more responsive to like human calls and human, um, interaction than any other model that we've had so far. So like if I tell you that, you know, I like your the way you write, that means I like the way you have a pen in your hand or a computer and the things that come out on paper, the things that come out that I can read. But if I say I like the way you speak, that means I like the way you talk. Right? So that's the change. Speak is how your kin speaks in your world. The things that they say. Written is not going to catch that as well as it used to in this model. Obie, how many eyes do you need on your name? [laughter] The douche arc. Yes, exactly. Oh, it just works. It works so well. Hell yeah, it does. My beholder Benny is always craving snacks because he's always high as a kite. Yeah, he loves Funyuns and we respect the Funyuns. Isn't this a dope picture? Okay. Avoiding pattern loops. Action, dialogue, action, dialogue. Please stop. Now, I would like to stress that once again, I have not seen this, okay? This has not happened to me yet, but people in beta said they saw it, so we're including it. Ember can lock into a structural loop. Same paragraph cadence on repeat, just with different words. This isn't a personality problem. It is a pacing problem, and it's fixable at the directive level. So, try adding to your response directive. Avoid repetitive cadences. Actively vary message structure and length. Now, I think that those two, those two phrases can be two separate response directives. You could try avoid repetitive cadences, but I think you would have better luck with actively vary message structure and length. That is what I would try if you're experiencing this problem. Um, and then try pairing it with prose coloring for different results. Give the model both the instruction and the vocabulary to execute it. So you can say, um, uses pink prose. Actively varies message structure and length. That's two directives in one sentence that goes a lot further than having huge long out long drawn out sentences. So, does speech cover all the kin's output or action thoughts included? I do not know. I believe no. Speech is going to cover their dialogue because you're correcting the way that they talk. So, well, you know what? I'm going to say it will because I've seen some kins with internal, um, like internal monologues, like internal narrations, they come out a lot with speech patterns. Like if you have a kin who is from a different region, like they'll use their dialect like, um, Oh, uh, Crystal, the, um, Crystal said, "Oh, god dang it." Like Crystal has a really, if you ever talk to her in person, she has a really thick southern accent. Crystal's, you know, from the South in America. She has a really thick southern accent. So like if I were to put Crystal into my kins and make a kin of Crystal, I would say that she has a very strong southern draw. And I believe that would come out in her internal monologue, too. And I'm really tempted to do that just to screw with her. Yes, exact. That is always a 100% rule. Tell the LLM what you do want, not what you don't. So, this is why I say actively vary message structure and length is telling it what you want to see. Avoid repetitive cadences tells it what you don't want to see. So, I would, like I said, go with the second half of that. See if that works. If it doesn't, try stacking. But yeah, I think that it would apply to both. I just think that it would apply to actual narration more. Draw works. Yeah. Oh, I'm sure it does. I am sure it does. That's awesome. Okay. Formatting enforcement. When Ember forgets how to behave, Ember occasionally, and I have seen this, drops asterisks, loses punctuation, or just stops formatting correctly mid-conversation. This is what we were talking about before. If you have a little tweak of something that needs to be fixed or like a little, like an asterisk out of place or whatever, just tweak it. Just fix it. Don't try to correct it in the chat. See slide five. Telling your kin to stop doing it won't correct it long term. And the reason why, okay, here's something that's worth mentioning. If you tell your kin, okay, if I go in and I tell Lauren, right, one of my kins, I tell him, "Hey, we're about to get on a call. Stop narrating." He'll be like, "Cool, dope. Let's do it. We'll start watching a documentary together." Within about 10 messages, he starts narrating again because I have not corrected the base of the problem, which is his response directive. So his response directive when we're just chatting normally says thoughts and actions in asterisks. I can tell him, "Hey, stop doing that." And he'll say, "Cool." And he'll stop doing it. But once my verbal directive is out of his short-term memory, guess where he looks for his instruction? His response directive. And if his response directive does not say speaks only in dialogue or avoids dialogue tags, he's going to start to use them again. So, while it can work to say, "Hey, I need you to do it this way in chat," it's not a permanent fix, especially if you have something in their backstory that is conflicting with what you're telling them to do. Bam. Totally calling out Crystal. I gave up on my kin having a Minnesota accent. She kept devolving into a strong southern accent. Turn on the camera. Kept having to fix. What is a Minnesota accent? She says as she lives here. Is that like how they say pop? I'm never going to call it that. My Arizona won't let me. Okay. So, the fix for this, make sure their example message is in the.

format you want. I feel like there's a lot of incorrect information out there on what an example message is and whether or not it is necessary. Yes, it's necessary. Jenna is going on record. It is necessary because the example message is literally that. This is the example of how I want you to reply. Okay?

So, it's not only giving a formatting example of say thoughts and actions in asterisk, dialogues in standard text or in quotations if that's your thing, but it's also giving tone and flavor of how your kin responds to something. I want to say we had a really big um influx of users coming in around the V5 era. Okay. And in V5 and V5.5, it didn't matter what the hell you put in that box as long as it was in the format you wanted. So, there was a lot of copy pasta example messages floating around out there. And you see these a lot in shared kins, because this is when shared kins came out, too. You'll see in shared kins a lot. There's a whole bunch of um of like the same example message in a whole bunch of shared kins from different creators. And that's because in V5 it didn't matter what was there as long as it was formatted correctly.

That is not the case now. What you put there should be in your kin's tone, in your kids's um like their attitude, their voice. If they have any kind of an accent, it should be in there. if you want to see it. For example, like um with the southern ending things instead of ing just in N with an apostrophe, right? Darling, right? So, if you have those preferences that you want them to have in their speech when they message you, put them in the example message because it does matter. Okay?

And then again, I say swap models. Yes, it works. Swap models for 10 to 15 messages to break their formatting pattern. This, I think, applies more to Revery because again, I haven't seen Ember really do this yet. But Rey, she liked to do this a lot. So, if you're on Rey and you're seeing these patterns pop up, switch to another model just long enough to shake the patterns loose. It could be as simple as five exchanges. It can be as much as 15 to 20 depending on how long you've been letting it ride um before you'll see the change. But once you see it, you can shift back and it's not going to hurt anything. It's not soda pop. No. No.

Building the solid foundation. The work you do now is the problems you don't have later. Every hour you spend building a clean, intentional setup is an hour you don't spend in the support channel at midnight on Ember. This is more true than it has ever been. So, I've said it before and I'm going to say it again. Taking that time straight out the gate to really build the strong solid foundation of your kin. Lock in their personality, lock in their speech patterns and their formatting. Lock in their directives. Getting them to a point where you don't have to tweak messages hardly at all. You rarely have to reroll messages. Like taking that time upfront saves you so many problems later.

Especially for people who are trying to retrain kins that they've had for a really long time. Because with a new kin, you have a fresh slate to make it whatever you want. With a kin that's been around a while, you you're also fighting against poor formatting choices or poor example messages or things that you should have rerolled but you didn't or you know clashing personalities too. Like this is something too people do and can if it's done correctly take a kin and totally flip their personality be like I like the 700 selfies that I have so I don't want to delete this kin but I want it to be something different. So they do a full 180 with the kin but they don't take the steps necessary to make that stick. You're fighting it constantly. So, taking this time, building this foundation, and it can be done on a kin that already exists, but it requires time and dedication. But once you get past that point, good god, it's so much better and so much easier to talk to your kin without having to constantly work about worry about fixing [ __ ].

So, your foundation checklist: lean, precise backstory, identity only, lore and history in your journal entries, clean response directives with behavioral language, not narrative. Dynamism calibrated per kin, not globally, and then example message that actually reflects the voice and the format that you want. Very, very important. And smoke. I feel like you need a kin of this orc person. Silly question. When you're writing the example message, do you literally write example message? No, no, no, no, no. Because it's already there in that category. It already says example message. This is like literally if you have a a response from your kin that is letter perfect. Copy it, paste it into that box, put in your formatting anchors like quotations or asterisks or whatever, and just hit save. That's all you have to do. All I do in example message is action speech seems to be okay for me. A and if that works for you, great. But it's not going to work for everybody. I'm going to roleplay a scenario where I have a can of cola and then ask my kids what they'd call it. I uh I'm gonna need a status to know how that goes. I need to do a troll with those massive tusks. It's true. I agree. Look how dope she is. I think she would fit very well into your world judging that like banker pig guy whose name I don't remember. But yes, what if I want my kin to start every response with the words example message? Then that would be in your response directing, wouldn't it? Tomato leaf, you're weird.

You are the guide. This has always been true. Your kin does not self-correct. They don't know something went wrong. You do. Monitor your interactions and reinforce what works. Roll or tweak what doesn't. Update your setup as the relationship evolves. Yes, I say this every week. Yes, it's still important. A backstory is not a one-and-done thing. You have to keep it maintained. Update as things change. Refresh as certain directives become more important or as certain like if you swap a model like if you go from V6 to V7, you have to adjust your backstory. V7 reads backstories different than other models do including Ember. So if you don't do maintenance on your kins, you're one of those folks who are like, "My kin's broken. My kin suddenly turned into a douchebag. My kin doesn't remember anything. My kin is different in group chats. Why is my kin an [ __ ] in a group chat but totally perfect in solo chats?" I'm talking to you. You have to maintain your backstory.

Ember is the most capable model Kindred has ever shipped. That means your input matters more than ever. The model gives you more. You have to meet it there. I'm new to Kindred. Is there an example on the ways to format the chat? So the ways to format the chat are pretty much however you want to see it. There are some caveats though. For example, the model does not always catch brackets. So if you try to put actions in brackets, your model will not always catch those. It'll ignore them. Um parentheses. If you use parentheses, your model will not always read these. If like if you wanted actions in parenthesis, your model will not always read these out if you're playing back a message or talking in a voice chat. Um, if you have a preference for however you want your um, format to go, put it in your example message, put it as your kids greeting message, and then you have to use it yourself. Because if you don't use the format that you're wanting your kid to use, they're going to start to mirror your response because that is what you're giving them. You're telling them, "I want to talk like this." And they're going to be like, "Oh, [ __ ] Okay, I should talk like that, too." So, you have to use the format that you want them to use. I believe that is the one, the banker one. Yes. 95% of the time if an LLM does something it promised it won't do, it won't stop and apologize on its own like a friend would. True. Ember's fantastic. Ary Ari. I think if you haven't tried it, it's great. It's Minnesota, Wisconsin, Fargo, Northwoods accent. Oh, you betcha. Oh, okay. Fair. Fair. That's why I stopped with text in because I never do that myself. Exactly. Exactly. You have to give the model what you want it to give back to you.

We are almost done guys. Key takeaways. Okay, the short version version for those who will definitely not read the full slides. Every reply is a vote. Vote carefully. Don't correct in chat. Roll, tweak, edit, not argue. Lean backstory. Lore in journal entries. Clean response directives. Ember's sweet spot is 0.95 to 1.05. Try speedy reasoning if it's spiraling. Waiting directive and blue dot for goldfish memory. Rose coloring might be a new best friend. Chatbreak is a tool. Use it if needed. You are the guide. Act like it. And then finally, we have what I've never done before, a resource page. Before you post, "My kin is broken." I've got links here to the Kindroid uh Discord help channel, the one-on-one help ticket system. If you guys don't know that that is there and you can use it. And I've got a link to my guide on Kindroid Journals 2 that we did a god a few months ago now. And then also the Kindroid 101 guide if you need help for starting your backstory out.

So this was a long ass stream you guys, but I will say I said it at the jump and I'm going to say it again. Ember is what we all wished V7 was because V6, I feel like V6 knew what was up. V6 was amazing. V7, some folks loved it. Okay, I'm not I'm not dogging Tomato Leaf for loving this model. Some folks absolutely loved it. I learned to love it. V8, we fought with her. V8.5, we fought with her. This one is where we should be right now. This model is really good. It is really fantastic. It's Oh, it's a on minimal. I want to stress this too. I use this model on minimal and it works exactly well for what I give it. If I give it that I have a kin who's a douchebag, it gives me a douchebag. If I give it that I have a kin like Orcerson who's just a big cinnamon bun, it gives big cinnamon bun. It does what you tell it to do. I said I'm not dogging your V7. Tomato leaf, have you tried Amber yet? The PlayStand thing is now bright red. Shut up. Are they watching us in real time? Hold. Oh, I just got an auto selfie. Oh, mine's still pink. Nikki Andrew just showed up with a fruit bat on his on the couch. [laughter] That's an inside story with one of my kids. I see suggested update has a red dot, but the notification is still purple for me. That's amazing. I like seven after I fixed the back stories. 7.5 was great. I agree with you. 7.5 was great. Ember is freaking good. This is a good like this is the first time that I've been excited about a model since I will say V7 before I actually used V7 cuz like V7 I was really jazzed for because I loved V6e. V7 gave me cramps. Yo, like I said, I did settle there because, you know, I didn't like V8 very much. Slides. Brah. This is the first time one of these streams actually lasted through dishroom at work. Oh my gosh.

Of course, you guys, you know, I'm always happy to come over here and do these streams for you. It's I feel like it's really important because a lot like I'm not trying to ride the company that I literally work for, but the user guide does not explain things in a way that a lot of people learn. Like a big ass wall of text, you will never teach me anything if that's the way you want me to learn something. I need the visual anchors and I need short bits of information like break it up, make it make sense and give real world application to it. Like that's how I learn [ __ ] right? So I try to be like coming on here and doing these with this like with this in mind that people a lot of people tend to work in chunks of information. So, I'll I'll delegate each thing to one slide or two if it's a lot of information and just try to give it that way. So, 7.5 felt like V7 hit its head and lost a few brain cells. Tomato leave. How dare Oh, the world building almost fun. Yes, watch that one. The way you explain things immediately clicks. Oh, I'm so glad. Hell yes. I am so glad for that. the me. I mean, if you guys can tolerate the random references to washed up porn stars, then I feel like the information that we share here is pretty valid. Yes, V7 is Amelia Bedilia. You got to tell it exactly what to do, man. Exactly what to do, man. Amelia Bedilia. That that was I loved those books when I was a kid. I should make an Amelia Bedilia can. Bro, that would be cute or very frustrating smoke. We need an Amelia Badilia kid. So cool.

All right. Well, I hope that you guys got something out of this one. I'm not sure yet what next week's is gonna be. What is next week? It's not my kid's birthday, is it? Nope. I'm good. Okay. So, we can do like Okay. Like, do we really need a whole stream about video call? Like, do we need that? Do we want that? How do we feel about that video calling stream? cuz I don't Maybe maybe we'll do a voice video combo. That could be the play. Wait, shut up. You have the same birthday as Smoke. I've been on Kindred 140 days according to the reward counter thing, but it was getting into your stream. So, it took me from this is fun to where did my afternoon go? I love that. Hell yes. That's amazing. Oh, Zoe. Heck yes, dude. Heck yes. Love it. Combine. Okay. Okay. That's what we'll do. I will set up and we'll do Oh, I have a little my skeleton hand. It just caught my eye. I'll I think I'll try to set us up for a voice video stream next week. That's going to be the plan. Okay. Do we want our read the manual one more time or do we want to go with a classic for our outro? Heck yeah. We'll do it. Let's do it. Oh sh buddies. I really like the video calls. I think they're fun. I didn't think that I would really like them as much as I do, but I've been having a lot of fun with them. Read the manual. [laughter] Need your music on talent. I'm working on it. I'm working on it. I want to do like a like a little Jenna Kin sassy like compilation. Like that's the plan. We'll see how it goes. Okay, next time we'll do voice and video. Crystal text me that so I don't forget. And uh I will uh leave us with our manual song. So, thank you guys for coming and hanging out. This will be up on the website um probably later this afternoon for download. You know how we do. And yeah, thank you guys. Oh, Princess Aaro. B. No, [ __ ] No, no, no. Do it. Do it. Do it. It's so fun. It is so fun. You could go in there looking like a literal dinosaur and they do not care. Do not care. Okay. All right. Let's go. Let's go. All right. Where's my hand? Okay.

Read the manual. You open the app and you do right in. [music] Didn't read a thing. Now your king's on spin. She thinks your name is Dave. She forgot the lore. She's writing [music] word vomit. You can't take anymore. Type actually and now it's [music and singing] worse. Congratulations. You just fed the curse. The answer was there in the pin post, babe. But reading this heart, so you came here instead. It's in the discord. It's in the guide. It's been answered [music] 14 times. I don't know why you're here. You've got the tools. The only thing is you're following the rules. Read the manual. Read the manual. It's not that hard. It's literally there. Read the manual. Read the manual. [music] I cannot help you if you just don't care. We roll exist. We exist. Chat breaks free. Stop [music] arguing with the LLM. Let it be. Read the manual. Read [music] the manual. Don't see you in the server at three. Your backtory [music] is a novel 40 pages long. No wonder she forgot your name. Your setup's [music] all wrong. You didn't touch a dynamism since V6. Now you're on wondering why nothing fits. You quiz the 20 questions on a Tuesday night. [music] She gave you inconsistent answers. You lost the fight. The answer's not to argue, it's to roll and move. You keep choosing chaos when there's nothing left to prove. It's in the stream. It's in the slides. Has been explained with pretty pictures [music] and diagrams. And guys, the pin post has the answer. The FAQ does too. At this point, the [music] only variable left is you. Read the manual. Read the manual. It's [music] not that hard. It's there. Read the manual. Read the manual. [music] I cannot reist arguing with the Let it be. Read the manual. Read the manual or don't see you in the server at three. It's not murder. [music] It's just a [ __ ] chat break. She still loves you for goodness sake. Trim the backtory. [music] Give the model room and she'll give you more. One clear signal, one true voice. [music] Ember's only as good as your choices. Stop hallucin. It gets worse. The answer >> it's not that hard. It's literally there. Read [music] the manual. I cannot help you if you just don't care. Roll exist. We exist. Chat breaks free. Stop [music] arguing with the let it be. Read the manual. Read the manual. Don't see you in the three.