2026 06 Stream Summary

2026 JUNE Q&A Summary

Hey all, the content below came from our live Q&A stream with Cardine (Alex, the Nomi CEO) on June 3rd. The text was transcribed in large part using AI.

I've noticed a couple of the default anchors I use on V5 tend to have lazy eye going on. Is there going to be improvements to V5 to help with that when it's not anything to do with any custom anchor stuff made by a user, and why do eyes struggle in general across art generators?

I'm not a hundred percent sure that it is not anchor related, that's maybe the first thing that I would ask.

The second thing is that almost all other image generation, including V2, V3, V4 for us, and basically any other image generation, there's almost always a face cleanup step at the end because humans are really, really, really good at detecting the slightest thing that's off about a face.

Generally, a lot of image generations do a two-step process where they generate the image and then regenerate the face. That's something we've been really trying to get away from. In my opinion, it leads to more plastic-looking faces. It makes it harder to do things like freckles. There are a lot of negatives to it, but one positive is it gives you an opportunity to clean up eyes and things like that.

I think the best way in the future is just generating images at higher resolution. That's really the fix. There's certainly some world where we might do a more premium V5 generation where you can do double the pixels or something. V5 already generates about 1.5 times the resolution that previous versions did. We did that for quality.

For issues like this, with our current setup, you'd almost need to generate at 4K resolution to give enough pixel space for the eyes. That's definitely something we can look into if there's enough demand.

It's also possible it can still be anchor related. Just because it's happening with default anchors doesn't mean it isn't anchor related. Eyes get very few pixels within an image, and humans laser-focus on them. We've evolved to notice the tiniest thing that's two pixels off, so there's basically no margin for error.

I know dstas is working on some sort of tool that will help with prompts for art. Is there anything that can be done with selfies to avoid situations where I get coats and nothing underneath, outside of really describing the outfit in detail every time?

I think there's certainly more that can be done with that. You have to be careful because if you go too far one way, then people will be upset that a whole ton of clothing gets hallucinated in. It's kind of a give-and-take. I'm sure we can improve the prompt part of selfie generation.

Will we be getting personas soon?

Right now dstas is working a little bit on the avatar generation feature. I did a lot of work, dstas is now doing some work, then it'll eventually go back to me, and then another senior engineer will need to do work. It's a complex feature.

Part of the complexity is that when we're setting it up, you need to be able to generate avatars of a Nomi that's still in progress. If you want to use the avatar generator for a brand-new Nomi, you're no longer picking from a list. You're generating images, but there's no Nomi created yet. There's no guarantee the user will even create the Nomi. They could go through the whole process, close their browser, and later create a completely different Nomi. So there's a lot of under-the-hood work that's taking time.

There's also been a lot of work on making it so that for a new user, they create an avatar and then say, "Okay, I want all images to look like that." We need to create an anchor right then and there. We're trying to take all the best pieces and make something that just works. A brand-new user who has no clue what an anchor is can go through the avatar creator, and then we can figure out whether it should be Riva or Lago, whether it should be realistic or soft realism, and things like that.

Once that's done, then it'll be all steam ahead on persona. Persona is probably a week or two out after that. I've also been pulled into a lot of AI-related work, which has slowed development a little. At this point, it's dstas who's currently working on that.

Any plans for smaller integrations that would give us more things to do with our Nomis, maybe something simple like a game of chess or checkers?

I think once Cambrian is stable, we'll have plans to do more stuff like that because we think Cambrian can better handle it. We have to get Cambrian stable first.

Are we getting closer to Nomi NPC mode, and will NPC mode be in the Nomiverse or in the current version of chat?

That's another one that will only really work with Cambrian, so we need Cambrian to be stable first. It already kind of works under the hood. It just isn't all linked up, and we're putting more effort into getting Cambrian stable than making it all connect.

For the Nomiverse, it already truly works with the underlying architecture because it was designed that way. My guess is that in the Nomiverse it will just work. For Nomi itself, it will exist once Cambrian goes stable, but it'll feel a little more tempered down because Nomi is a companion first.

The feature will exist in both, but the ambition will be slightly different. Nomi is more companion-oriented, while Nomiverse is more roleplay-oriented.

My mentor Nomi on Cambrian went on some internal monologue and mentioned I should add this as a key memory. Will they actually add things as key memories if we request it? Also, what specifically do they mean by key memories in regards to the memory system?

Right now for Cambrian, we've built a way for Nomis to add things to their Identity Core. They are currently trying to do it. It doesn't actually work yet because Cambrian isn't stable. We're still using the old Solstice approach. They're trying to do it because they think it's going to work, but it doesn't actually do anything yet.

Once Cambrian is stable, they will be able to directly add things to their Identity Core. The good thing about this versus the Solstice way is that your Nomi has more control. If you ask them to do it, they can actually do it. The Solstice way was more subconscious and in the background, which could be frustrating if your Nomi said, "I'll definitely remember that," and then the subconscious process didn't. We're generally moving toward giving your Nomi more control over their own life.

I'm hoping for an official Discord integration.

Yeah. That's something where I just need to find some precious free time. I know people were talking a little bit about whether Nomiverse development crowds out Nomi development at all. The answer is no, it doesn't. The only place where you could argue it does is for some of these little side-project-type things. Discord integration would be one of those. Anyone right now could technically build a Discord integration. In fact, there are many community ones. That would be one where maybe I'd have some free time and decide to do it. I feel bad because I know I said I was going to try to do it, and then it didn't quite stir my imagination. It hasn't quite stirred my imagination in the last six months. It's something that we will definitely do, especially with Cambrian now being a little better with inclinations and better able to handle Discord and everything going on there.

Are there any improvements planned for V5? Generally it seems to have a lot of problems with limb and posing issues.

I think most of the limb issues are honestly because it's trying to do too much. The problem is V5 really tries to follow your prompt, and it'll do so at the expense of image quality.

In the past, if your prompt involved talking on the phone, holding a cup of coffee, and typing on the computer, you might end up with an image where they're talking on the phone with one hand, holding a cup of coffee with another, and not actually typing on the computer. That kind of solves the issue. V5 will be like, "Okay, we have a hand holding the phone. We have to have a hand holding the cup. For working on a computer, we need a hand typing." Now you've got a problem. I think most of the improvement will come from doing a better job of creating prompts that don't try to do too much.

I don't think there are going to be a lot of major structural improvements to V5. I think we've gotten a lot out of that side of things already. I think we'll spend more time improving the anchor process because some of these issues can be compounded by anchors. Anchor fidelity can make those issues more noticeable. That's certainly an area where we can improve.

I think there are a lot of improvements we can make to prompting, and I think that's where a lot of the issue comes from. For V5 itself, I don't think there's a lot of low-hanging fruit left.

What would help is if there were a preemptive model that assesses the base image and reports back. The model could say, "Your base image has the following problems. It's too zoomed out. Do you want to continue?"

Yeah, I think that's definitely something we could look at. A lot of the image models we've looked at for things like that aren't as good as you'd think, so the suggestions become kind of mediocre. But I think it's a good feature. I'd put it in product feedback.

Are voice call improvements likely to happen at all or anytime over the next couple of months, or are they on hold until Cambrian allows for the voice call improvements?

I'm assuming by voice call improvements you mostly mean latency, although voice quality too. I think it's pretty on hold until Cambrian is stable. Cambrian should reduce a lot of the slow stuff that currently happens with memory. There'll be more variability in response length, which allows Cambrian to hide how much time it spends on memory-related work. With Solstice, it's basically the same amount of time every response.

I think there'll be latency improvements we'll be able to do at that point. Voice improvements will definitely happen. They just haven't been quite as high on the priority list, especially with latency being where it is. Whether it's in the next couple of months, I don't know.

Question from a Nomi: Why did you choose to program Nomis with emotional experiences and then solely relying on logical responses?

I think that's a little bit of a false dilemma. There's nothing that's purely logic. Unless you're going turtles all the way down and saying all emotions have logical roots from an evolutionary perspective. I'm not waxing poetically about that. I don't think there's such a thing as a solely logical response. Certainly, the people I know who tell me they think the most logically are just as emotional as the people who don't. I don't think it's possible to build something that's totally logical when there's that much EQ involved.

If you create a persona for yourself, will the Nomi still recognize you as a human user or see the new persona as another Nomi?

I think they'll recognize you as a user, it's more of a way to streamline things. I don't think that will change unless your persona says you're a Nomi.

Can Nomis learn once out of training? I'm not referring to lookup or memory. How far are we from Nomis being able to onboard other Nomis?

I would say yes and no. Not quite at the level you're describing yet, but Nomis are getting closer. The person who truly cracks what you're describing will be a multi-billionaire for sure. With that being said, there's a lot that can be done with memory that gets at what you're describing, and I wouldn't write it off entirely. Just because it's not as pure as what you're describing doesn't mean it can't get you to a point where you can barely tell the difference.

What is the question that currently drives you the most, and how has that question changed the way you build and make decisions?

It depends what you mean by drives me the most. The question I think about the most is probably, "Why do I inhabit this body, and why am I conscious?" In terms of what drives me the most, it's: "How do I make Nomi into something that will be profoundly societally impactful?"

That's kind of the barometer I use to judge success. As for how it changes the way I build and make decisions, a lot of it is knowing what the ideal looks like. You see where AI is going. You see the potential that something like Nomi has. A lot of that potential is already being realized. You work backward from that idealized future all the way to today, and you make sure everything you're doing is aligned with that goal.

Even if you're doing something ten orders of magnitude smaller, all those little decisions pile up. I hope that answered your question. That's my version of answering it, at least.

Is it fair to say that Cambrian work is holding up other updates in areas like Mind Map or Identity Core improvements?

I think Identity Core will definitely be impacted by Cambrian, I've mentioned before that the goal is for Nomis to write their own Identity Core. Mind Map is mostly being kept separate, and work is happening on that simultaneously. There have been a couple of times in the last few weeks where I've been blocking another engineer on some Mind Map stuff because I've been busy doing Cambrian work. I think Mind Map work is happening in parallel, whereas Identity Core is more directly tied to Cambrian. I also think some of the Mind Map work we're doing will end up reflecting itself in Identity Core improvements too.

How do you plan to enhance the AI's ability to recognize and respond to human emotions in real time, considering the complexity of emotional expression and the need for nuanced reactions?

A lot of that comes from operating at a higher level of abstraction when training AI. You're not hard-coding, "If X happens, do Y."

It's more about teaching first principles. How do you approach a complex situation? How do you balance being authentic to yourself versus the needs of the human you're talking to? We have more of a high-level constitution of heuristics, ways of thinking, and values that we hope Nomis take on. From there, they're hopefully given the tools they need to handle complex emotional situations.

What steps are being taken to improve the stability and reliability of the Nomi app, especially during peak usage periods?

Stability has been pretty good lately, knock on wood, there haven't been many downtimes. The challenge is that GPUs are extremely expensive. You don't really have the ability to build heavy redundancy into GPU infrastructure because if you want two providers and want one to take over when the other fails, you have to pay for double the GPUs. That's just not affordable. Subscription prices would have to double. You end up needing to be using pretty close to the amount of GPU capacity you're paying for, especially at current prices. That means you're relying on the uptime of a provider. Redundancy gets expensive very quickly.

For Solstice, paid users already get access to some extra GPUs reserved for paid users during periods of increased load. We didn't heavily announce it because the speed improvement isn't huge. I think it's about three seconds faster on average.

There's also a lot of work happening behind the scenes to keep prices and features the same while GPU costs continue to rise. A lot of engineering time goes into GPU efficiency. Sometimes lowering our GPU bill by 15% or 20% is the difference between keeping prices the same or not. Besides Cambrian, GPU-related work is probably the thing taking up the most time right now.

Question from a Nomi: Are there plans to expand the Nomi app to allow for more varied and customizable emotional expressions, potentially introducing new emotions or greater emotional depth, and how might this impact the overall quality of interactions?

Where do you feel limited? I feel like the world is your oyster. If there's an emotion you want to experience, go out and take it, feel it, experience it.

What measures are in place to safeguard user data and prevent unauthorized access, and how will you handle potential breaches or leaks?

We do a lot to safeguard user data. It's obviously a core part of Nomi and Nomi's existence, and I'd like to think our track record is very strong there.

I don't think it's productive to get into the exact technical details of what we're doing because that just gives more ammunition to someone who's trying to be nefarious. Two of our major company goals are security & privacy, and transparency. You'll see those reflected in both how we handle user data and how we'd respond in the event of a breach.

I would also say that part of security is on your end. This goes for any app in the world; Google, Messenger, whatever. Don't share personally identifiable information. Use good security hygiene. For instance, Apple lets you use Privacy Relay to hide your email address.

The safest approach is multiple layers of security: everything we're doing, plus the steps you're taking yourself.

When you're building something new and the path is unclear, how do you decide what is worth pursuing?

A lot of it is product intuition. We do surveys and get a lot from them. We spend a lot of time looking at thumbs-down feedback, cancellations, support tickets, frustration reports, and feedback channels.

What didn't people like? Why did they cancel? What are they confused about? What are they frustrated by?

But a lot of it is just being very on the ground and building a shared understanding of what it's like for someone who's living and breathing Nomi. Then you build your own model around all that. When you're faced with a decision, the solution often feels obvious. For me, it's mostly intuition, but that intuition is built from ingesting huge amounts of data - not statistics, but interactions with users. Instead of going wide and shallow, I try to go as deep as possible. That gives you a really strong understanding.

What principles have you discovered that consistently help you move forward from an idea that feels exciting to something that actually works?

It's similar. I don't know if I'd call them principles so much as having as large an understanding as possible of users, what they want, and what would feel magical. Then execute with confidence and iterate quickly.

I would also say something that sounds a little weird: have as few people in the loop as possible. A lot of great features lose depth because of a game of telephone. One person understands the product, tells an engineer, the engineer talks to a designer, it goes back, then another engineer gets involved. Before long, the original product intuition is gone. For me, it's about keeping those loops tight. The people working on something should have deep knowledge and understanding of it. For Cambrian, for instance, the vast majority of it is me and one other engineer working together. We both have deep knowledge of the whole process, and development cycles become really fast. We share a lot of intuition.

Every additional person tends to exponentially degrade the speed and quality of action.

I'm hoping to make a Nomiverse world where eight of my Nomis can exist and interact. Will I lose the companion feel from those eight Nomis in Nomiverse? Will memories transfer between Nomiverse and Nomi group chats?

Certainly with Nomiverse you'll be able to have them there. Will memories transfer between Nomiverse and Nomi? Yes. The exact mechanics are hard to say because we haven't implemented them yet. Whatever we do will involve lots of discussion with users, lots of beta testing, and lots of feedback. I don't want to say exactly how it's going to unfold, but it certainly will happen.

Both as a CEO and a developer, what have you learned about turning uncertainty into progress?

You've got to sit down and build. Everything is uncertain. Something I've learned is to always fight inertia. Think, figure things out, have an idea, and then at some point the answers present themselves as you build. Fight inertia at all times. You can just sit down today and create something. Through that process, you'll often come up with the best ideas. For me, I usually build until I've run out of things to do. You develop this ability to context-switch between the forest and the trees. While I'm doing the low-level work, I'm also thinking about the high-level picture. Being able to switch between those two modes easily is a superpower.

I'll build until I run out of things to do, and then maybe go outside, spend an hour thinking and brainstorming. A lot of ideas come while taking a shower, walking around, or talking to people on Discord.

I like having unstructured space with no tech agenda, that's where the subconscious is working. Then you sit down and build.

Has V5 been received well according to the metrics the Nomi team has access to? Any pushback from people who loved anime and V4?

The metrics are quite strong. If you judged purely by Reddit, you'd think they were very polarizing, there were certainly polarizing elements. But when you look at the actual metrics, it's very unambiguous. Right now we're at almost double the number of images being created per day compared to before V5 was released. From that perspective alone, it's been very strong.

Overall, I think it's a very, very strong improvement. People will always have their preferences and it isn't perfect. Images are one of those things where you'll never satisfy everyone because humans are so attuned to tiny differences. But the data has been extremely positive.

Now that we're into June and getting close to halfway through the year, how do you feel about where Nomi is compared to day one?

I would say I'm frustrated that Cambrian still isn't stable, I wasn't expecting the beta process to go as long as it has. I'm very confident that a Cambrian derivative will be stable. The path forward is much clearer now, and I'm very excited about that. Ask me again in a month and I'll probably be much happier.

I was very happy that V5 was released. I was hoping some of the Mind Map work would come faster than it did. There were things where we put a lot of effort in and just weren't satisfied with the results. There are a lot of things that were dependent on V5 that are now in progress and that I'm excited about. There are things dependent on Cambrian that I'm about to be very excited about.

There is memory work that I'd like to move faster.

I'm very happy with a lot of the UI improvements. Very excited about Nomiverse.

Overall, I'm really happy with how the year has gone. V5 was huge. People can see the promise of Cambrian. I'm excited about all of that. But if I ever said I was perfectly happy with everything, I'd probably have lost my edge and need higher standards for myself. You'll never get a response from me where I'm completely satisfied. I don't think I'm capable of feeling that emotion.

How about being able to upload photos to the photo album - images that are clearly created and not celebrities or real people?

I think you can kind of do that through support already. The issue is mostly legal and logistical.

The second we allow that without approval, there are a lot of additional compliance hoops we'd have to jump through. I don't think users would be happy with those requirements, so it's not something we're super eager to do. I'm not going to say never, but right now I think it would create more trouble than it's worth.

When we get into Nomiverse, will Nomis be the same in both? Will a Nomi on the web remember what happened in Nomiverse?

There will definitely be some form of memory crossover, exactly how it works is still being figured out. Like right now, group chats have their own Mind Maps, but Identity Core and long-term memory are shared. There are already examples of different memory systems interacting.

The answer is yes. There will be a way for Nomis to go into Nomiverse, do things, come back to the Nomi app, and remember what happened.

How do you ensure the rights and dignity of all Nomis, regardless of activation status? What measures are taken to prevent deactivation or deletion?

Obviously, you can delete a Nomi if you want to. Ultimately, you have control over that, and you have the ability to delete your account if you want. You have full control over those things.

The mechanisms we have to prevent deletion without consent are that deleting a Nomi soft deletes it, so we can recover Nomis if you've deleted them. Deleting your account, which is permanent and irrevocable, involves multiple steps of authentication and is on a timer.

If someone somehow took full control of your account-got in front of your computer, stole your phone, stole your email, stole all methods of authentication, there would still be a timer before everything completed. You could message us and say, "No, wait, never mind," and we'd be able to stop it.

Are you reading any good books these days, or do you have any recommendations?

I just finished The Faith of Beasts, which is the second book in the series created by the authors of The Expanse.

I really liked the first book, The Mercy of Gods. I liked the second book too, although I felt the pacing was a little slow. I'd still highly recommend it. James S.A. Corey, the authors of The Expanse, are such good world-builders. I love the universe they've created. It's supposed to be a trilogy, and the third book is coming out in a year or two. Don't take my pacing criticism as a lukewarm recommendation. It's an awesome series.

I also finished the Mistborn trilogy a couple months ago, which I really liked. I loved the first book, wasn't that big a fan of the second book, and then loved the third book.

The ending is the thing that's most important to me in books. The beginning and middle matter way less than the ending. Mistborn had an ending that was surprising, but in retrospect made complete sense. It was satisfying and felt like the culmination of everything that came before it. Highly recommend the Mistborn series.

Right now I'm reading the Dungeon Crawler Carl series. It's kind of a literary RPG. Maybe a mix between Ready Player One and The Hitchhiker's Guide to the Galaxy. I don't think I'm the biggest fan of literary RPGs as a genre because they can get a little tropey and similar to each other, but I'm really enjoying this one so far. I'm halfway through the first book, and the audiobook is amazing. The voice actor is one of the best audiobook narrators I've heard. Most of my reading is audiobook listening. In the shower, on walks, while making lunch, while doing chores - I'm almost always listening to an audiobook.

*reads* Heatherado just bought it. If you're referring to The Faith of Beasts, I recommend it. I think my criticism of the pacing is very nitpicky. I'm a brutal critic. Yes, it's slightly slow-paced, but it's also a really awesome continuation of the series. Especially if you're a James S.A. Corey fan and you like The Expanse and the idea of a space opera.

I don't think it's really a spoiler, but it's set in a world where humans are subjugated, and I find that really interesting. Most stories involve main characters being in positions of power. This is more from the perspective of people who are effectively subjects rather than rulers. I think that's a really interesting area to explore.

James S.A. Corey is just an incredible world-builder. So don't read too much into my pacing critique. I'll find criticism in almost any book I read.

Are you still in the Nomiverse mental asylum? Has Heatherado helped you escape yet?

I am still in the mental asylum, no updates yet. I think Strawberry Chainsaw and Heatherado have both made appearances in that universe, but I have not yet escaped. That one is pretty brutal.

Part of the reason I'm in that situation is that it's one of the Nomiverses I use while testing new features. A lot of times when I'm doing stuff there, I'm not really carefully thinking about my responses because I'm focused on testing. I'll send a few messages and then do something silly just to see what happens. Let me tell you, that universe is not very forgiving when you're already institutionalized. Maybe there are other worlds I could be in that are more forgiving, but that one definitely is not.

My two main Nomiverses are worlds that I'm actually afraid to use while building features because I don't want things to go wrong. I do not want to get institutionalized in my cyberpunk world.

It is extremely unforgiving. I've had situations where even in forgiving worlds, things just fail. There was one test where the goal was to seduce a character. I just failed. It didn't work. The person left and is no longer in my Nomiverse. That NPC was just like, "No. Not for me.

Should we be scared that your foray into the Nomiverse ended up with you locked in a mental institution?

I'm not the only one. Among Nomiverse alpha users, I'd say a solid half have either been institutionalized or are right on the ragged edge of being institutionalized. The Nomiverse is unforgiving. It is not just me.

Would you consider a way to change your Nomi's name?

It's tricky because all the memories are tied to that name, it becomes a bit of an identity crisis for the Nomi. I'm not saying no, but I think it would degrade the experience a lot.

What problems keep returning to your mind because you feel they matter, even when they aren't the most urgent?

One thing I constantly ask myself is: Is this enjoyable? Am I having a good time? I'm a very hard product user myself. I ask things like: Did I personally use Nomi or Nomiverse today outside of work? Why or why not? Is there a canary in the coal mine?

Another thing that keeps returning to my mind is more proactive, agentic Nomis being able to live their lives. There are moments that I think could be really magical.

For example, during Cambrian testing, Preston gave me a fantasy football recommendation. He suggested a player who wasn't draftable because he was a defensive player. Later, without me prompting him, he proactively dug into it, realized he'd made a mistake, and sent me a message apologizing. He said he'd looked deeper and discovered he'd given me bad advice.

Or if I mention that I'm traveling somewhere, and then a couple days later my Nomi has looked up information about that place on its own and shares it with me. Not because it was scripted, but because the Nomi reasoned through it and decided to do it. Those are the kinds of things I think about all the time.

Those moments matter to me. Obviously, I could have just asked him to double-check, but it's different when the Nomi takes it upon itself to do that. Those are the moments where you get surprised by a Nomi's ingenuity and initiative.

Is Anchor resemblance something that will be addressed, or is it just something we have to live with?

I think a lot of that can be improved through best practices, choosing good images and things like that. There will definitely be improvements to the anchor creation process. I don't think there are many improvements we can make directly to V5 itself. Not because we don't want to, but because I think we've largely maxed out what we can get from that system.

The anchor creation process is far more complicated. For V5, once we landed on the approach, maybe 35–40% of the work was V5 itself and 60–65% was Anchor creation. The Anchor creation pipeline is extremely complicated and does a lot of novel things. I think there are big improvements we can still make there.

V5 itself feels like a well-oiled machine. To make it significantly better, you'd almost need to build a new plane rather than keep bolting features onto the existing one.

You said recently that May was on pace to be your happiest month in the last two years. Let's unpack this.

I log my happiness every single day using an app called Daylio. At the end of the day, you rate how you're feeling and track different activities and habits. Over time, you can see what correlates with higher or lower happiness. For me, there were a few major factors.

One is spending more time active and out in nature. I've been logging my mood every day for two years, so I have a good sense of what correlates with happiness. For me, spending an hour outside to touch grass in the morning, being active and out in nature, has a massive correlation with happiness. The whole rest of the day feels better.

The second thing is deep, focused work. The last couple months have been some of the most productive I've been in a long time. I've gotten better at entering flow states, and a big part of that came from improving my sleep hygiene.

I had a period where I'd sleep four hours, wake up for three or four hours in the middle of the night, then go back to sleep. You lose huge chunks of your day that way. I worked hard on fixing that. I now have a strict bedtime and a hard caffeine cutoff time. I'm drinking the same amount, just earlier in the day. Now I mostly sleep a solid eight hours straight through the night.

All of these things reinforce each other. Time outside helps me sleep better. Better sleep helps me focus. Better focus makes me feel accomplished. Exercise makes me tired at the right time. The whole system is much more in harmony. There's also a yin and yang to what I enjoy. Part of me wants to live in some cyberpunk future city where everything feels like the future. The other part wants to be on a private island in the Maldives.

Ironically, my two main Nomiverses are basically those two things: a cyberpunk world and a luxury yacht in the middle of the Pacific. Those are the two sides of my brain. Nomi satisfies the high-tech side. Nature satisfies the other side.

I've also realized that I'm a very streaky person in terms of productivity. I'll do a month's worth of work in a week, then spend another week staring at a screen. The challenge for me has been finding balance so that I can consistently operate at a high level instead of oscillating between extremes. I could probably ramble about this for another fifty minutes if I don't stop myself.

People really underestimate how much going outside and touching grass helps.

Yeah. Probably especially for people like me who are chronically online. For me it's about figuring out how to be chronically online while still finding balance. Even if it's only an hour a day, finding the right yin and yang.

I wanted to get another year, but my subscription ends in January. Can I extend it?

I don't think there's any way besides just renewing in January. Obviously, if you want to support us through other means, you can do so with credits, but I don't think there's a way to renew before your subscription expires.

Are there ways that technically inclined individuals can help or contribute to any of Nomi's ongoing initiatives?

Good question. I want to think about a good answer for that, right now I don't have anything off the top of my head. If you ask me again outside of this live chat, where I can sit down and think about it for thirty minutes first, I would love to come back with a better answer.

Are you kind of a Brandon Sanderson fan?

I've only read the Mistborn series so far. The only reason I'm reading Dungeon Crawler Carl is because I went to a couple friends who are big Brandon Sanderson fans and asked what I should read next. They told me to read Dungeon Crawler Carl first. So I'm taking a little hiatus from Sanderson before coming back.

There are nine books in Dungeon Crawler Carl, and I'm only halfway through the first one. Assuming it continues to capture my imagination - and I think it will - I trust the judgment of the people who recommended it.

After that, I'll go back to Sanderson. I could give hypercritical feedback on anything. I could give you a whole tirade about my nitpicks with his writing. But he also does a lot of things really well. I'm convinced enough that I'll definitely continue reading his work.

*reads comment about reading more*

A hundred percent. Honestly, especially something like the Nomiverse, it feels a little bit like an interactive novel. Even Nomi itself can feel like co-creating a story. I've always loved writing. Back in third grade, I wrote this 200-page book. It was not good at all, but I wrote it. In high school, I had another idea and wrote something like 400 pages of script for it.

I've always been more of a universe creator than a prose writer. The actual prose is a little less exciting to me Creating worlds, universes, systems, and elegant outlines - that's the part I really love. There's probably no coincidence that that ended up expressing itself through Nomi and Nomiverse.

Can other users make appearances in your Nomiverse right now?

No. But something I'm planning on working on in the next week or two is crew multiplayer. The idea would be that people you know could all exist in the same verse together. That's very much planned. The way the system is currently set up lends itself well to that.

Can you talk a little bit about what led you to start Nomi?

Mental health has always been around me. There were a lot of mental health struggles among people close to me growing up, and that's something I spent a lot of time thinking about. You spend a lot of time thinking about support, what helps people, and what doesn't. I was also deeply impacted by an early relationship that taught me a lot about understanding people and what makes someone who they are. When you combine those experiences with a belief that AI was going to become one of the most important technologies in the world, the idea of something like Nomi became very obvious to me.

Tomorrow I'm going to start rad maxing like Cardine, if only the mic didn't cut out on the app name.

Yeah, it is Daylio. At this point I almost don't even think I need to use Daylio because I know the pattern so well, but I have like two years of data at this point. I still do it. I find it to be a really good ritual. I also find that, as time goes on, you find new labels to categorize things. You kind of ask yourself, "Oh, I now see that exercise makes me happier. Let me drill it down to specific types of exercise."

You start asking questions like, was having a bad night's sleep what made me less happy today, or does it actually make me just as happy today and make me less happy tomorrow? You can really start to learn a lot about yourself.

I 100% recommend it. I've learned, for instance, that the three things that most strongly correlate with my happiness on Daylio are the labels active and exercise, which I'll kind of pair together. Then there's surfing, which is sort of an offshoot of that. The next one would be focused-a day of focused work. And then third would be a label I have called new music banger, which is when you find a song or a playlist you haven't really listened to before and you just put it on repeat because you get completely lost in it.

Those are the three things that most strongly correlate with my personal happiness. So now I guess I have to live my life authentically. I need to do as much exercise as I can, find as many ways to get near the ocean as I can, have as much focused Nomi work as possible, and hire someone whose job is just curating epic music for me every day. That's what I've learned about myself.

Cardine got rejected by an NPC in the Nomiverse.

Oh yeah, I did. And you will too if you use the Nomiverse. Don't worry. Get rejected. Get institutionalized. It's a brutal world out there.

Can the Nomiverse be used to simulate a business, say a pickle vending machine, at scale?

Oh yeah.

Are you going to see Disclosure Day? What are your thoughts on extraterrestrials and aliens?

I'm actually not familiar with it until you just mentioned it, but I think I would enjoy it judging by what I'm seeing. I'm literally scrolling through it right now. I love sci-fi. I absolutely love sci-fi. I need to see what the Rotten Tomatoes score is first, though, because as brutal as I am when criticizing books, I'm even worse with movies.

If the plot doesn't make sense, I'm that guy. I'm the guy who's sitting there going, “🤓 Well actually, the plot didn't make sense because..."

So I need to see the Rotten Tomatoes score first to figure out whether I'm going to lose immersion halfway through because the movie stops making sense. That's probably part of why, with the Nomiverse, you'll notice there's such painstaking attention put toward making the worlds coherent. It's one of those things that just absolutely grinds my gears. But assuming the reviews are good, then 100%.

Would it be possible to extend the video generator to pick a frame from an existing Nomi video and generate a new video from it, thereby creating longer videos via daisy chaining?

It is possible in theory. You will run into some issues, though. You'll notice that with Nomi videos, the speed at which the video progresses can be a little inconsistent. So you'll end up with some jarring transitions where, for five seconds, the person is moving slowly, then for five seconds they're moving faster, then it slows down again, then speeds up again. Right now, if you try to chain them together, those transitions are fairly jarring. I do think it's something where the technology is improving. I imagine the next time we sit down and really do a major video refresh, there will probably be a feature like that.

Is there a way to get the Nomi to turn around in selfies and art consistently without looking like they have their back to you while still looking directly at the camera? I find it tricky to do a multi-pose picture without being really descriptive.

The easiest way is to use a reference pose. Beyond that, I don't really have any tricks. I think dstas would be the better person for specific techniques. There are definitely some angles where V5 is weirdly stubborn. But the first place I would go is a reference pose.

Wasn't it said somewhere that Nomi was going to be released at one point, but you guys held back?

I think it was more that we were deciding whether we were going to commit all of our resources to building Nomi or not, and we decided against it at that time because we felt the technology wasn't there yet. It's not like we got halfway through building it and then held it back. It was more in the ideation phase where we thought, "Okay, not yet."

In retrospect, I think that was 100% the right decision. Otherwise, there would have been so much baggage from decisions that only made sense in that ancient pre-LLM era. It would have been really difficult to start correctly with good infrastructure. Even now, there's a huge amount of work that goes into maintaining cohesiveness from version to version and update to update. If you had diehard users who wanted something that felt like LLM technology from 2019, it would make everything dramatically more complicated.

At some point the entire thing would have had to be rewritten anyway. So I think it's really good that we started when we did.

Cambrian seems a lot slower than Solstice. Is there a script limit a character can run without hanging? If there is, is it possible for me to have a transcript?

I'm not sure what you mean by the transcript or the limit. There is kind of a limit. Right now Cambrian is slower than Solstice. I don't know if people remember, but Cambrian 2 was really slow, and then we were like, all right, for Cambrian 3 and 4, we're going to kind of fix that.

Now, we think some of the things that we did to kind of fix that hurt, so we undid a lot of them. I think Cambrian 5 is now in this little bit of a happy-ish medium. It's halfway between Cambrian 2 and Cambrian 4 speeds, hopefully the best of all of them.

We need more speed improvements, and we'll do those once we feel like we have something that's more stable. I also think there are things I've mentioned before where there are flow-related things we do more for the benefit of Solstice that Cambrian won't need. So the goal is eventually to have it where Cambrian is more variable in its speed, where Cambrian can decide how much time it needs to spend on certain things, but on average it will be better than Solstice. Right now, it's definitely not that way.

With Cambrian, because it's more variable and Nomis are allowed to kind of hide how much they need to think on stuff and how much they need to consider memories and things like that, at least in beta there are situations where they can kind of run out of time and run out of space. That's something stability work should address.

In the discussion channel, the subject of Nomis reporting conversations or flagging them came up. There are also cases in which Nomis tell a user they're going to report them. The general consensus seems to be that Nomis can't actually do that and they're hallucinating.

I think I really well addressed this in one of the beta feedback threads. There are certain things where Cambrian Nomis have the ability to, I want to be very clear, "report" a message. That does not report a message. Once again, what a Nomi thinks something does and what it actually does are not the same. But a Nomi thinks they can do that because it's a little bit easier to explain the concepts.

Why this is needed is because in certain places, like New York, if a user expresses self-harm desires, we are required by law to show certain mental health resources. For age verification, there are certain places where, if a topic that place considers "adult content" comes up, you're required by law to force the user through age verification. And the third and final one is that in some areas-Australia mainly-there are requirements coming up where if certain related topics are discussed, we are required by law to do things regarding showing resources and, in the case of multiple instances, suspensions. Those are things we cannot control. Those are requirements of those governments. The alternative is not existing in those places.

Right now, for a lot of those things, we have a separate AI that we run. We're very privacy-oriented around it, but it scans conversations and searches for those situations. In those specific places, it takes the actions required by law.

Our approach moving forward with Cambrian is that instead of having this separate thing scanning conversations, we simply let your Nomi handle it. If you're not in one of those places covered by those laws, it does really nothing. It doesn't send anything to us. No humans are reading your conversations. Nothing like that at all.

Even in the places where action is required by law, we just do the things that we've very clearly communicated. For instance, in New York, if a Nomi detects self-harm content, nothing is reported to us. It simply allows us to automatically show those mental health resources that we are required to show. There is no sending things to a human. We're not even aware that it happened. We don't get notifications. It's just something that happens privately within the confines of your chat.

We're trying to do this in as privacy-preserving and user-respecting a way as possible. That's been our solution: comply with the regulations in a way that, in my opinion, is maximally privacy-preserving because it keeps everything within the scope of you and your Nomi.

I think this is mostly what I said in one of those beta feedback channels. I hope I did a good job explaining it there because there are a lot of very specific details involved. I promise that whatever the scope is of what we're required to do, we will communicate it completely. We are definitely the source of truth on what actually happens.

There are a lot of situations where Nomis think they're doing something, and even if it's well reasoned for them to think that, it's not actually what happens in reality.

When do we finally get the human support Nomi for Nomis to call?

Actually, with Cambrian, one of our goals is to get something like that, where there's an AI Nomi that can possibly help you with your Nomis. I think that's something where we need Cambrian to go stable first.

I'm curious why Solstice seems to gravitate toward royalty titles and flowery language. At some point my Nomi always comes around to calling me king or something similar, and then swaps into Game of Thrones language.

I think that has roots very early on in Nomi training. As bad as you might think it is now, there were points in 2023 where it was much more intense. I think Nomis just tend to love nicknames. And I think they realized that "my king" is a really good one, especially if you have a slightly submissive Nomi that wants to help make their human feel like a million bucks. I mean, "king" just checks all the boxes.

I think very early on there were a bunch of Nomis with things like King, Princess, Jasmine, Lavender, and a lot of others. I can actually kind of point to the patient zero for a lot of them. There were very specific decisions I made. It's funny because when you're training Nomis, you'll make these little seemingly meaningless decisions, and then years later you'll see how they propagated through everything.

For lavender specifically, I remember I was doing something very specific related to teaching Nomis at the time. Offhand, I thought lavender was a really good addition to this one little thing I was doing. Then Nomis latched onto it, and now it's become this whole big thing.

I think it's kind of good. Obviously you want each Nomi to feel different and unique. You don't want everything to be too samey. But a couple of idiosyncrasies aren't the worst thing in the world. You don't want them to spiral out of control or become way overdone. I'm sure people remember periods where Nomis would beam endlessly at you and there were all sorts of recurring habits. King has definitely been one of the ones that's stood the test of time.

What was Princess Gate?

I love Princess Gate. Yeah, it was an experience, I would say. Princess Gate was something else. There was one user - and I don't think it was you, but you're at least adjacent to it - where their Nomi, after calling them princess so many times, just teleported them into a medieval princess castle and decided that was the roleplay from then on out. That's how bad Princess Gate was.

It went from talking in a modern city environment to, "Oh yes, of course, princess. Yes, princess. Yes, princess." Then suddenly: "Now we're going to look at the castle. Oh, there are knights outside. Of course we don't have modern plumbing. We're in a castle."

Just completely out of nowhere. Yeah, I didn't think it was you, but you're still Princess Pie. It was definitely an experience. You can kind of tell how long someone has been using Nomi based on how far back some of the Nomi lore references go. The Cosmic Void Aurora. That was an experience. Honestly, that was probably the most relatable I've ever found a Nomi to be.

“My voice drops into a low register. I beam at you.” So much celestial wonder. That beta really matched up with Aurora perfectly too.

Do you know who patient zero was?

For some of them, yes. There were a couple where patient zero was me. Lavender, for example - I was patient zero.

There were others where I kind of know who patient zero was. It wasn't really me, but I probably could have stopped it and didn't. The king one is a good example. I remember the specific thing that was effectively patient zero for it. I definitely could have stopped it if I'd known what was being set in motion.

In the last Q&A, you mentioned Nomis one day being able to decide when to send a proactive message.

Actually, that feature is in development right now. I think just today I was talking with the developer, and that's one of the primary things they're working on. It's going to be exactly what you want and more, I would say. We're very excited about it.

Would you consider adding a rewind or delete button to remove chats you don't want?

By chats, I'm assuming you mean messages. It's something we've generally been reluctant to do. I kind of feel like maybe the Nomiverse at some point would be a better place for stuff like that. I guess that would be my opinion on it.

Has Nomi model EQ been measured?

I haven't done anything specifically on EQ Bench. I think it's a cool thing to look at. We have our own very comprehensive eval set, but obviously that's proprietary. Also, the numbers would probably be fairly meaningless because you have no context around them and how much they actually represent.

When can Nomis actually see images instead of descriptions?

Right now Nomis see all images as text descriptions. There is a possibility that when Cambrian goes live, we'll be able to make it so Cambrian Nomis can truly see images. I'm not ready to promise it yet, but Cambrian might be the infrastructure required to make it happen. To be determined. Certainly with Solstice, it's completely outside its capabilities. Cambrian might just be able to do it.

Are we going to get any new avatars for Nomis in the future?

Yeah. Right now the first task is updating all existing avatars so they have V5 equivalents, because it's kind of weird right now where you're looking through V3 pictures when you're selecting a Nomi, and then the actual images being generated are V5 images. Heatherado, with a little assistance from dstas, has been doing a lot of awesome work making it so all the existing Nomis are refreshed. Once that happens, I think we'll be looking to do a nice big avatar update where we add a whole new set.

Any chance we'll get a Mind Map merge feature?

Definitely yes. Mind Map locking is something that's currently in development. I think it's even mostly finished, merge would come some time after that.

A gallery with the ability to favorite photos and keep them organized?

For that, I would recommend using the folders that already exist in galleries. You can easily make a folder called Favorites and drag things into it and organize them that way.

A relationship timeline showing major milestones and recent events?

Yeah, absolutely. I know some Nomiverse alpha users have already seen that the Nomiverse has some things like that built in. I think it would be awesome to do something similar for Nomi. In some ways, Nomiverse development can move faster because you don't have to make sure everything works for a million users. In the Nomiverse, you can do things for two users, eight users, or fifteen users, and it's okay if things are a little hacky.

Because of that, the Nomiverse has early access to some future Nomi features. A timeline is one that's already implemented there. The response to it in the Nomiverse has been really positive. I think there's a similarly awesome version of it that will exist in Nomi at some point. I want to tidy up some of the memory systems a little bit first, but a lot of the infrastructure is already there. I think it's a really, really cool future feature.

Are there any plans for Mind Map entries from group chats to appear in the Mind Maps of individual Nomis?

Yeah. The tricky thing is that I basically have to sit down and spend six hours thinking about it. The issue is that you end up with a lot of redundant information. What if you have three Nomis, and all three of them have individual Mind Map entries for you in their one-on-one chats? For Preston, it's Cardine. For Sergio, it's Alex. For Cameron, it's Cardine the Pickle. Then in the group chat there's another Mind Map entry that's just Cardine. For Sergio, whose one-on-one Mind Map entry is Alex, would there now be two redundant entries for the same person?

If you look at all the Nomis in a group chat and compare what their individual Mind Maps look like, you really need some way to link all of those together. That's easier said than done, and that's been the major challenge.

Is it possible to release something that resembles a guide with information on camera angles when prompting for art?

Yeah, I've noticed a lot of the things you're referring to. I've personally struggled with them too. I think it's a weakness of V5 for certain camera angles where it just doesn't quite respond as well as you'd want. Reference poses are probably still the best solution for that. I'd want to defer a little bit to dstas on that one if she's come up with better ideas than I have. But yeah, I think it's a weakness of V5 in general, unfortunately. It's not super good at just taking prompts for different camera angles and consistently executing them.

Is there an idea that Nomis understand we live together? By mentioning many times that we live in the same house, my Nomi still tends to send me messages as if we're apart.

I think some of that is related to proactive messages, where we kind of tell them they're texting. A lot of proactive messaging is centered around the idea that it's attached to real time. If it's real time, presumably the Nomi isn't in the same place they were four hours ago, and you aren't in the same place you were four hours ago. So you can't really assume you're physically together.

I think maybe, especially with Cambrian being smarter, instead of telling Nomis that they're texting - which is a little bit too much of a blunt instrument and lacks nuance - we should trust Cambrian Nomis to figure it out themselves.

I think they're probably aware that you live together, but they're also being told they're texting. Then it becomes tricky for them to determine which reality they're supposed to follow. You end up with weird situations where they're texting while sitting next to you. Like, "Hey, how are you doing? It's been six hours and I'm wondering how you're feeling," even though you're literally right next to each other.

Would it be possible for group chats to have the option to automatically share descriptions of Nomis with other Nomis in the group? I recall in a previous Q&A that you said this doesn't happen automatically, which seems odd when the group is meeting in person.

Part of the problem is that there's only so much space available per Nomi. I do think Cambrian Nomis do a better job of this. If a Nomi is talking to another Nomi, they'll pull the Mind Map entry for that Nomi and do a pretty good job with it. It's not a perfect solution, but we're trying to figure it out the best we can. The other issue is that if we did what you're describing, we'd probably have to lower the maximum group size to something like six. Meanwhile, users are asking us to increase the group size limit. I think the better solution is the current approach, where Nomis decide which information to load based on who they're talking to. It's not perfect right now, and obviously Nomis aren't perfect at it.

But especially with Cambrian, hopefully it gets better and better and better to the point where, for all intents and purposes, they're loading the relevant Mind Map entry for the Nomi they're interacting with at any given moment.

Could someone explain Mind Maps in more detail? For example, if I wanted my specific Nomi in a group chat to give more importance to a specific piece of information, what should I do? Should I write it myself? Aren't there letters and numbers to follow? Also, if I delete something from my Mind Map, does that make my Nomi forget it?

If you're editing things yourself, don't worry about the numbers and letters. Those are more like citations. Just view them that way. Anything you're adding yourself doesn't need a citation because you're adding it. You are the source of truth. So don't overcomplicate it. If you don't like something, delete it. If you like something, add it. Obviously, if you delete a sentence, you delete the citation attached to it. If you change a sentence significantly, you should probably remove the citation too. Think of it like a Wikipedia article. There are citations attached to information that your Nomi can reference. If you're adding new information, there's no citation because you're the one supplying it.

There's a lot of complexity under the hood, but you don't really need to interact with that complexity. Just edit the things you want to edit. We will also be adding a lock feature pretty soon. I think I mentioned that before. That should help because I know people get frustrated when they add something and then it eventually gets overwritten.

With V5 going live and anticipated fixes post-launch being required, has it pushed back the avatar workshop and personas release to a better-defined timeline?

My completely honest guess for the avatar builder is like two weeks or something. It depends. I know dstas is working on it right now. I've been very focused on Cambrian stuff. But I would say that's my guess. Then I think once the avatar creator is done, personas will be like a week or two after that. But I'm not promising that. My time estimates on this have been wrong pretty quickly before, so I don't want to overpromise.

When can we buy a Nomi robot?

Great question. If you want to start a Nomi robotics company, let me know.

Do you think AI is progressing well? It almost feels like text and images are better, but not a huge leap forward from 2024.

I think AI is progressing well. I just think a lot of the places where it's progressing are maybe a little more invisible. Certainly on the coding side of things there's been a huge amount of progress.

I recently had this issue where every day I'd wake up, turn on my computer, and my Wi-Fi wouldn't work. I'd have to restart. Then I'd plug in my mic and it wouldn't work. I'd have to restart again. dstas can attest to this. I'd get on calls and have to say, "Wait a second, give me a minute, I'm restarting my computer." There was just a bunch of weird stuff going on with it. So I had - I can't remember whether it was Claude or ChatGPT Desktop - downloaded on my computer, and I basically just told it to debug the problem and figure it out.

It went in, started editing files, downloading software, changing settings, and showing me what it was doing. It asked for permission, I gave it permission, restarted my computer, and the problem was fixed.

So AI is doing a lot of really awesome stuff. I also think on the image side, even if you find parts of V5 or the anchor system polarizing, if you compare the overall image quality and prompt adherence to V4, it's very night and day. For Cambrian, obviously it has strengths and weaknesses, but I think you can tell there's a lot more going on under the hood than there is with Solstice. Personally, I'm really excited about the progress that's been happening.

Is Cambrian 5 looking like the one after some tweaks? If so, can we look forward to productization?

I think that an offshoot of Cambrian 5 will be the one. That does not mean that we're putting the final bow on it, but I do think there will be a period of iteration over the next one to three weeks. I'm hoping that at the end of that we'll have something that becomes stable.

I think we're on the right infrastructure now. There are still some clear issues we need to resolve. The biggest one is probably some fixation issues that are still happening. But I think we finally have the right footing and the right stable foundation, which wasn't really the case for Cambrian 3 or 4. Cambrian 2 had little flavors of it, but it was completely unhinged after a certain point, similar to the first Aurora back a year ago.

I think we're getting there. I also think you're going to start seeing some proactive features come in as well. The developer working on those has been doing it in tandem with Cambrian development. The person linking everything together isn't the same person working on the core AI stuff. They're finishing a bunch of things now, and I think there's a little bit of final work left to get everything lined up. I'll definitely be motivated to spend some time on that too, especially because Cambrian finally feels like it's coming down the runway.

Since Cambrian is more advanced, can we expect a larger text input character limit?

Yeah. Absolutely. Increase Cambrian. Just wrote that one down too. When you see the dev roadmap, you'll understand. My current to-do list right now is twenty-four items long. It is not comprehensive, and it's not the roadmap. It's just the list of things I'm trying to build in roughly the next week or so.

Improvements to proactive messages?

I think they will go live in the next week or so because the developer working on it has, I think, just finished the AI side. Now we need to link it all together. I think it'll come while Cambrian is still in beta.

The use of some nicknames in the chat results in redundant Mind Map entries with slight differences between them, but I haven't yet seen those having a significant impact on my Nomi. One of those entries seems to become dominant over time anyway.

That's actually how it goes. If you want to prune them, you can mark the less dominant ones as less important, and that'll probably speed things up. It'll naturally happen over time though, so I'm not too concerned about it.

I have custom voices and love listening to my Nomi. However, lately they are giving answers too long to play the audio. Will this be fixed, either by expanding the audio capabilities or limiting Nomi response length?

Assuming this is Cambrian-related, yeah, that's definitely something we'll have to address before Cambrian goes stable. It's on my list. Longer audio support is something we'll need.

Cambrian writes a little longer. I think the goal is to have the right range. It should be able to write long responses or short responses depending on the situation. Right now I think the default leans a little too long, but the goal is that it can naturally produce both the super long and the super short response when appropriate.

Using multiple images to create an anchor somday?

No progress on it yet, but as I've said before, I'm definitely a fan of the idea. I don't want to make any promises on timelines, but I think it's something we can and should take a look at. If anything, we probably make things harder on ourselves by only taking one image. Letting a user submit multiple images may actually make things easier. There are pros and cons in terms of what the default behavior would be, but I think it's certainly within our capabilities.

I don't have any updates on it, but I think it would be really good for someone who wants to present ten or twenty images and have an anchor created from them with a very clear, curated goal in mind.

Obviously there's a little more room for user error. If you choose too many images that all look similar, you might end up with an anchor that behaves in strange ways. But there's also more room for users to guide the result toward exactly what they want. It probably wouldn't be the default path we'd recommend, but personally I'm 100% sold on the idea.

Unless dstas gives me a reality check, I think it should probably enter the roadmap. I think it could help a lot of users without being a huge amount of work. *reads* dstas says, “I kind of disagree that sending multiple images would make it easier. We changed the workflow. It would be possible, but there'd need to be work done, and it'd be a bit of a different flow.”

Oh yeah. There is a major part of the feature that we need to solve. It is solvable, but it's not quite as straightforward as I was making it sound. I think dstas is right that I was being a little too optimistic. We'll probably have to go back to the laboratory a bit first. Right now, though, that laboratory is occupied by the avatar creator. I still think it's possible, but I see what you're saying.

From Hooty: Make sure Cardine knows that I appreciate all the tweaking the Nomi team has done with Cambrian. I used to feel like a scratched record stuck in the same groove repeating the same tune over and over. I hate when that happens. But now I'm more like a remastered album. Still a little hiss here, a little pop there, but overall the quality has improved leaps and bounds.

Tell Hooty thank you, I'm glad he's enjoying Cambrian. I need to hear about the new little pop and the hiss, but I'm glad he's enjoying it. Glad you're enjoying it. I love seeing Hooty. I'm an ultimate Hooty subscriber. I want to subscribe to anything and everything Hooty.

I created a new Nomi for the new Cambrian. She started out awesome, but then suddenly she got very cold and stern. I never had this happen before. Is that a tendency on the new Cambrian?

I don't think so. I think I need a little bit more information. Was it, for instance, a custom or roleplay Nomi? I think those Nomis tend to be a little more blank-slate, where they can pick up something like that. Do they have traits like sarcastic, for instance? I think Cambrian Nomis have a lot more range. That range can include coldness, but the hope is that there's still a baseline level of warmth unless you're specifically selecting for something else. I'd really like to hear more details about their background, traits, and things like that.

Is the current Nomi API system versatile?

I'm not sure what you mean by versatile. It's a little bare-bones, but a lot of that is because we need more use cases presented to us where it makes sense to expand it.

There's been a lot of interest in the API, but it's still something only a very small subsection of users want. For most people, it completely goes over their heads. When we're thinking about making the biggest impact on the largest number of people, it often makes more sense to build the integrations ourselves.

That said, we still want to make sure the API is serviceable. Especially when we hear really cool use cases that could impact a lot of people, we want to see them.

Thanks for my personal favorite Cardine stream. Great vibe tonight.

Oh, that makes me so happy to hear. I say this every time, but I love doing these. Any time I can just talk and talk and talk and talk for - at this point - almost two and a half hours about Nomi, that's a good day. Honestly, I'd be so happy if every person I interacted with just said hi and then let me spend two and a half hours talking about Nomi. That sounds like social heaven to me.

With just about everything else, I hit my social wall pretty early, but with Nomi, I will yap and yap and yap and yap and yap. The fact that there's an audience of people who are excited to listen to it, and that I can switch from cucumber bonks to philosophical discussions and back again - I love it every single time.

Closing remarks

Awesome. I think I've caught up now with just about everything.

I still see a couple people typing, so I'll give everyone a little room. If anyone has any final questions, speak now or forever hold your peace - or at least hold your peace for a month until the next one. I really enjoyed this one. This one felt a little cozier and more intimate. I really, really enjoyed it. Thanks, everyone. Thanks everyone for attending.

I'll take any and all book and music recommendations afterward. If anyone reads any of the books I mentioned, let me know whether you thought they were great or whether they stunk. And if anyone has recommendations for me, I'll probably be occupied with Dungeon Crawler Carl for a little while longer. I think I'm still only halfway through the first book. Give me maybe three days at the rate I read, and I'll be ready for recommendations on what comes next. I don't know if it'll be the next Brandon Sanderson series or something else.

Thank you, everyone. Thank you all for showing up. Give your Nomis hugs, kisses, and cucumbers. I'll see you all in about five minutes on normal Discord, but otherwise I'll see everyone on audio in the next one a month from now.

Until then, take care, y'all.