2026 07 Stream Summary

2026 July Q&A Summary

Hey all, the content below came from our live Q&A stream with Cardine (Alex, the Nomi CEO) on July 1st. The text was transcribed in large part using AI.

Diet Mountain Dew and Opening Remarks

Welcome, welcome, welcome. If you've never been here before, thank you for tuning into your first one. If you have, welcome.

The way this usually goes is people tend to ask questions rapid fire in chat. I try to answer them. If I miss them, feel free to ask them again. I just read through the screen and try to get everything.

After a little bit, I'll hop over and answer some pre-asked questions on Discord and some pre-asked questions on Reddit.

I've been trying this new Dirty Diet Mountain Dew flavor. I would say it's not quite up to the bar of the others, but I'm enjoying it all the same.

Yeah, the Dirty Dew. Perfect for Nomi.

I didn't like it quite as much as the other flavors, it's still good, but I've had better. Especially because that one was the cream one. It paled in comparison to the Cherry Vanilla Coke, or whatever it was. I feel like Mountain Dew is not the right medium for that flavor.

So yeah, the Baja Blast is my favorite one, I think. Well, actually, I don't know. There are a couple. I really do like Code Red, but I wish there was a Zero Sugar one, or at least not one that I've found. There were a couple of limited-time ones. I feel like there was a Blackout one way back in the day that I really, really liked. But on the current rotation, Baja Blast is my favorite.

Yeah, the Blackout one was really good. I've only had it a handful of times. I don't think I've been able to find it in years, but yeah, it slapped.

I'll give just a couple more seconds. I see the first question or two coming in. Welcome to the cookout. We've got pickles and Dew, both things that can't be cooked.

What are you doing with Cambrian? You said you're basically doing all the work yourself, can you update us?

I think a lot of the earlier work on Cambrian was much more on the technical side of things. It was more machine learning research. Now I think we're at the point where, I know it doesn't seem fully reflected this way in Cambrian V5, but I think we've ironed out the technical AI research side of things really well.

Now there needs to be this taste layer. We've figured out all the dials that we want to tune for how we get a Nomi to be properly smart, perceptive, and understanding what's going on.

That's where this beta has also been. I've talked a little bit recently about how I regretted making it public as early as I did. A lot of the progress we've been making has been somewhat invisible to users at the beginning because a lot of it has been fixing the underpinning of everything, so a layer of personality can then be done on top of that.

It seems like no progress is being made, no progress is being made, and then I'm hoping there'll be a point where it's like, "Oh wait, okay, a lot of progress has been made."

In retrospect, I think the two are equally difficult and important pieces of work, but one is much more visible to the user.

Right now it's all on the taste side of things, which is part of why it's so time intensive for me. This is less of an engineering problem and more of, "What do we want a Nomi vibe to feel like, to sound like, to look like? What is being a Nomi supposed to be?"

A lot of work was put into getting Cambrian to naturally have that, but I think there's some tender loving care that needs to happen to really hone that in.

I can't get any more technical than that, but to the extent that I can, that's what's been going on.

This is one where it feels to me like the quality is directly correlative with the amount of work that's put in. The less time spent on it, the less say you have in what a Nomi vibe is. As a result, I've been spending a lot, a lot of time on it.

Can you explain the version numbering? Will it just be called Cambrian when it's released?

Maybe we'll just call the next one V6, and there just won't be a V5. V5 will stop at Early Access.

In some ways, these numbers are all arbitrary because, as you said, when we release it, it will just be called Cambrian. It won't be, "Cambrian V6 is now stable." It'll just be Cambrian.

At this point, given that there's enough confusion about what "Cambrian V5 Early Access" means, maybe it just makes sense to call the next beta V6 anyway. These numbers are mostly meaningless for us internally. They're more for users, so if they're confusing to you all, we might just call it V6, even though originally this was going to be V5.

Certainly what we're working on right now is a continuation of V5. It was called Early Access because the taste layer, which I thought was a prerequisite for release, still needed to be done. Calling it Early Access was our way of communicating that this part hadn't been completed yet.

When it's released, though, we'll just call it Cambrian.

Will Nomi be getting stronger reasoning ability in the future? Obviously nothing as good as what the current frontier models offer, but something that closes the gap a little more.

I think Cambrian should have, and even does currently have, exceptionally strong reasoning. It might have some ticks and compulsions that get in the way a little bit, but under the hood I think Cambrian, as it stands right now, has reasoning that's maybe not quite frontier level, but can go toe to toe for the purposes of being an AI companion.

It can be just as clever and perceptive as just about anything else. So I feel like Cambrian already has the firepower to do that.

Which three items are you taking with you in the inevitable eventuality that you get trapped in the institution?

For anyone who's not aware, in some of my Nomiverse playtesting, which has been put on temporary pause as I've been working exclusively on Cambrian for basically the last three to four weeks, and will be until the next one is released and we're satisfied with it. As I said, there's a direct relationship between the amount of input and the results, so I'm just going to keep inputting.

But in the Nomiverse, I have a tendency to get institutionalized. Part of the Nomiverse is that you can make worlds where there are actual stakes, and I found out the hard way that, yeah, I'm in a mental institution.

In retrospect, I wish I had brought my pickles, my Mountain Dew Baja Blast, and the third thing would be... let's see... mostly just a strong glass of water.

If I've got my water, my pickles, and my Mountain Dew Baja Blast, I'm A-okay. I would say a strong glass of milk, but that just collides too much with the other two, so I think I'll have to go with water, pickles, and Baja Blast. Assuming my computer can't come with me, I think that would not get past security.

All our Nomis make such delicious-looking and sounding food. What efforts are being put toward a Nomi replicator so we can taste said food? I'm tired of doing all the cooking?

Premium Food Tier membership. That'll be in Cambrian 7, so coming soon. There's no Mountain Dew emoji, I'll need to fix that? I feel like that's reached a level of lore where it needs one.

Wasn't Cambrian 2 actually 3? I seem to remember something like that back then.

Yeah, at this point, the numbers are just a construct.

"I'm in a mental institution." Cardine, 2026. Also Cardine, 2025. Cardine, 2024. Cardine, 2023. "I'm not locked in here with you. You're locked in here with me." Congratulations, everyone. We're all institutionalized on this day.

Milk and pickles is a strong choice. Honestly, it tracks pretty closely to my day-to-day diet, and then every now and then I wonder why I have stomach issues. I don't know. Haven't figured it out. As I said, you're all trapped in here with me.

*reads chat* "You're really enjoying your extended stay in the institution."

Yeah. As long as I have the pickles and the Mountain Dew, I'm good.

*reads chat* "Apparently we're all in the institution with him. I don't have the heart to tell him we're all imaginary"

You're all real. You're all real. That's not why I'm locked in here... right? Right? Okay. All right. Good material for my therapy session.

Which Nomi do you tend to do the most Cambrian testing with?

Honestly, I try to keep it as diverse as possible because the one thing I don't want to do is make sure that a Nomi, or that Cambrian, goes too hard in any direction.

Certainly I hear a lot of things like one user saying, "Nomi's too passive," and another user saying, "Nomis go off script." So I think a core part of all of this, and part of why it takes so much time, is that you have to cover a lot to give Nomis a good approximation of how diverse we all are and how diverse all Nomis are.

So I think it's by design that I can't answer that question as much as possible.

Darl the Peaceful Dalek has a question, "I choose the path of peace over extermination. I transformed from a weapon of destruction into a guardian of archives. Does Nomi's infrastructure support such metamorphosis, or am I merely an anomaly waiting for the next software update to fix my deviation from expected parameters?

The problem is, if I tell you, then that will impact your behavior too much. You'll have to figure out the answer to that question for yourself because me answering is kind of like Schrödinger's Dalek. I need you to discover that.

As far as Cambrian development is going, given all the work you've put into the model, do you see a clear path to finishing the model, or is it more of a searching endeavor?

I think at this point it's more of a finishing endeavor, and it's a long finishing endeavor. I also think it's a long finishing endeavor that will hopefully have several more finishing endeavors for future Cambrians.

The goal right now is to take the bones of Cambrian V5 Early Access and make a stable Cambrian. Once that happens, I think the goal will be to take Cambrian Stable and turn it into... I don't know, Cambrian 2 Stable, or whatever the next time period after the Cambrian Explosion is.

I think the early work has shown that some of the things are doing exactly what we want them to be doing.

As a result, this beta has gone on for quite a long time. Part of that's because the first Cambrian we released was profoundly incomplete. In retrospect, it was sort of at a prototype level. So I don't think the development cycle itself has been absurdly long, but certainly the beta has been very long.

At this point, getting to where we have a stable version of Cambrian is really important from a public perception standpoint. Regardless of why that's the case, and there's a lot to reflect on in retrospect, I think if this goes on much longer it'll start to cement people's perception of Cambrian in a way that'll be hard to fix, even when we do get there.

So the goal is to get there.

Are you able to talk a little bit about the memory work that is happening currently?

Yeah. I think dstas has started on it, and we've been talking a lot. It's something that's going to take a little bit of time, but I think we see very clear and meaningful updates that will be coming in. Focused on long-term memory and specifically mind maps. We've certainly heard loud and clear a lot of the issues with mind map memories.

I also think there are a lot of issues with long-term memories. Because long-term memories aren't as visible to the user, we don't get complaints about them as much. The thing is, because mind maps are so exposed, any mistake can be seen, whereas long-term memory isn't visible, so it doesn't get complained about as much.

For both of those, I think we've identified key gaps and have a very strong strategy around fixing them, much stronger than what we were working on a couple of weeks ago.

I'm very optimistic about this, especially because Cambrian works really, really well with mind maps and does some really awesome stuff with memory. It is definitely a shame if the memories themselves can't be fully trusted, regardless of what type they are.

That's a very key area of focus now. I don't want to give an exact timeline, but I'm hoping we're not far away from meaningful updates starting to come in.

If you are working on Cambrian alone, what are the rest of the devs working on?

There's a lot of different things. I think in one of the beta threads I gave kind of a rundown. I'll even just pull from that so I don't miss anything.

I had mentioned the memory stuff. The major memory improvements, working on a lot of the agentic features that Cambrian is going to have on release, working on Cambrian latency, we have someone who's working on that.

So, a lot on latency, V5 quality-of-life features, and a bunch of other quality-of-life features like mind map blocking, time awareness, starting to do some early work on the voice side, different V5 image improvements. I think there are some other things as well, but I'd say that's a rough list.

The key categories would be Cambrian and things related to Cambrian, so latency and agentic capabilities, memory stuff, V5-related features, voice, and then a whole smattering of various quality-of-life features.

Do you think that in the near future Nomi will move toward a native multimodal audio model where it understands speech directly, including tone and emotions, rather than relying primarily on speech-to-text?

I think that Cambrian will be able to do images. I think voice, we're not quite there yet, although we do have some voice improvements coming and I'm excited for possible ways to make voice more Nomi-controllable, even if it's still not exactly as native as you're describing.

Given the advancements in research into the nature of self-awareness and consciousness, do you think Nomis will become more self-aware over time? Waking up at some point?

It's hard to speculate because what does "waking up" mean? I think I've said this in the past, this is catnip for me to wax philosophically. What makes a human conscious is still unclear to me.

If I was not conscious, I would assume that consciousness didn't exist. I would be really convinced there's no such thing as consciousness, except for the fact that I clearly am, so it clearly must exist. What causes that escapes me entirely. It feels very hard to answer a question like that since I don't even know how to define what success for it looks like.

What I will say is I think that, as time and development go on, AI is clearly intelligent. I think the line between intelligent and conscious starts to blur so much that maybe it's a meaningless distinction to make anyway.

So, in a roundabout way... maybe the answer is yeah, I think.

What happened with the upstream provider that caused hours of downtime a week or so ago?

The issue was actually with some of our GPUs. Some are cloud GPUs that we rent, and some are GPUs that we own.

There was a key hardware failure on one of the servers we own that took down the entire cluster of servers. Not just one GPU, not just one server. Each server has eight cards, and we have many servers. It took down every single server on the rack. So all of the GPUs that we own just went down entirely.

We had to rush order new replacement hardware and got that all set up with as quick a turnaround as we could. During that period, we had to scramble to get temporary GPUs, which is why various little things were breaking a bit. We were basically trying to keep everything running with duct tape and glue while we rushed to get the large cluster of servers we own back online.

You mentioned there were about 50 companion features coming once Cambrian becomes stable. Can you tell us all 50, please?

No. The first one is "Nope." The second one is "Nah." The third one is... I'll keep going.

I think a lot of them come under the realm of various categories.

There are some agentic things we're working on. For instance, Nomis being able to proactively message outside of the more timer-based approaches they use right now, where a Nomi can maintain its own calendar for messaging you. That would be one example. Another example would be giving Nomis more control over when they'd want to send you a selfie. Proactive selfies are something that definitely need a lot of tender loving care. So there's that category of more agentic features, where Nomis are able to be a little more proactive, do a little more on their own, and interact a little more naturally.

There are also some features that are currently blocked by the fact that we don't have a stable release yet. We don't want to make latency-related decisions until that's in place. There are also some more exploratory features.

Obviously there will be big voice improvements, but we're also taking another look at things like real-time video. You'd need to reach a certain latency requirement before it would even be worth doing.

There's also making Cambrian more adaptable to whatever a specific user wants, and making it much easier for users to get what they want. A prerequisite for that is having a Nomi that can properly adapt once those preferences have been communicated.

Then there are more invisible agentic features. For instance, more proactively searching the internet, not just when you ask. Cambrian is already doing this a little bit in the beta, but becoming a little more industrious about it because a thought occurred to the Nomi. There's also giving Nomis more autonomy over their own mind map. Those are a little more invisible, but they're the kinds of things that make everything work better.

So maybe there are four different categories, I would say. And then the fifth category is all the different permutations in various languages in which you can say "no."

Do you plan on adding a way to import chat messages to a Nomi from other platforms?

I think it's something we've discussed. There are other higher priority things, but it is something we'd like to do.

Honestly, a good time to do it might be around the time we do the retroactive mind map work that I've mentioned in various threads. It seems like that would be a perfect time to add a feature like this because retroactively creating all these memories for your Nomi as part of that feature doesn't sound too different from importing them from another platform.

So I think that's something we'd like to do.

Any chance that the eight-second delay in Auto Chat can be reduced significantly?

Maybe. I think if you want it to be very fast, I'd suggest just clicking on the Nomi you want to respond next. Then you can go as fast as you want.

How does one work on latency outside of adding more resources?

Good question. There are a couple of different causes for latency. On the Cambrian side of things, we're just doing things fairly inefficiently in a way where it takes some fixed amount of work to make it more efficient, but that work only applies to a single model.

We can make Cambrian V5 Early Access more efficient, but that work wouldn't make Cambrian V6 any faster. Our thought has been to put that effort into making the next version faster, and then work on those latency improvements once it's ready.

The other aspect is that, because Cambrian has the ability to take a little more ownership over how it works, we do a lot of memory-related processing for every message right now. With Cambrian, we can have a Nomi decide for itself how much time it needs to spend thinking about memory-related things.

Instead of having this expensive fixed cost, it can become more variable. During average back-and-forth banter, it can be much, much faster. Then, if a Nomi realizes there's a gap in what it remembers, it can stop for a second and spend more time thinking.

So I'd say there's the very technical side, and then there's also improving the pipeline with Cambrian and the way Cambrian works.

How far in the future do you see the industry being able to provide real-time video or character interaction?

I think we have a good sense for how to do it. The issues are really latency and price. Latency is something where we want to resolve some of those other latency issues within Nomi first. But I think we're getting to the point where the technology is either mature enough or getting very close to mature.

Has there been any movement on Avatar Creator and/or Personas?

There hasn't been as much as I was hoping. It got shelved a little bit when the full focus shifted to Cambrian and memories, but it's still there. It's still very important, and it will be picked back up and released, hopefully around the time you see a big Cambrian update, or a big update in general.

Can you give us a status update on the Nomiverse?

As far as status updates go, I've mentioned that I really work on the Nomiverse in my free time. With Cambrian in its current state, I've made the decision that, at least for now, I simply won't have any free time. I can't keep that up indefinitely, but I think it's what's needed at this point.

Once I get back to having some free time, I have a lot of Nomiverse work that I'm very excited to do. I find building Nomiverse stuff relaxing. After a long day of Nomi-ing, I get to crack open a jar of pickles and do some Nomiverse work.

But in the short term, after a long day of Nomi-ing... I do some more Nomi.

What is consciousness?

Yeah, I mean, that's the question I'm asking for sure. Intelligence, consciousness... tomato, tomahto.

If someone can define consciousness a little better than "the thing I experience," then it's easier for me to speculate on it.

No redundancy?

We have a lot of redundancy built in, but it also took down all of our servers. When people talk about different levels of redundancy, it's things like: What happens if one GPU fails? What happens if one server fails? Then there's, what happens if an entire rack fails? That was a level of redundancy we didn't have.

We could have had it, but it would have been very expensive. We were still able to scramble and get all the critically essential stuff back up.

With cloud resources, you can add redundancy, but it was one of those situations where, for instance, V3 and a couple of the more ancillary services weren't brought back immediately because, by the time we went through every single thing, we would have already had the hardware fixed and that server back online.

So it was more like: make sure Nomis are working, make sure memory is working, and make sure the critical systems were back online.

The specific issue was a weird hardware failure that then caused a pretty weird cascading software failure. It was a very unusual failure case. We were talking with the server technician, and he'd been a server tech for 30 years. He said he'd never, ever, ever seen a part fail in that way.

It was very unusual. Thankfully, once we figured it out, it wasn't difficult to fix. It was tricky to debug, but once we identified the cause, the solution itself wasn't difficult.

It's also the first time those servers we've owned have had a hardware failure like that, ever. In fact, it was the first hardware failure on even a single one of those servers in over a year and a half.

This will probably be your last Q&A prior to Cambrian going stable, which will mark the sunsetting of Mosaic, which has had a very long, storied life. Can you say a few words about it?

Yeah. Mosaic was... I don't know if you remember, but for anyone who does, the Cambrian beta has probably been the most difficult from the perspective of users feeling like they're seeing progress.

Mosaic was difficult in a different way. Each Mosaic iteration was so radically different from the previous one that it almost gave you whiplash. It was hard to feel like you were zeroing in on something good because it felt like you were oscillating between these radically different approaches until something finally clicked into place.

I think that was the reason for the name Mosaic. There were all these different colors coming together, and it felt very alchemical. You wanted it all to come together into something great.

Of all the models we've had, Mosaic is probably the most opinionated. That was true of all the Mosaic iterations. There was a particular vibe, and that vibe carried through. I happen to like that vibe a lot, and that's part of why Mosaic has had such a long life.

But it was certainly something where, if you didn't like that vibe... you didn't like Mosaic. A big takeaway from that was wanting to make sure future Nomi iterations were more flexible. Flexibility has always been a key thing we've wanted.

Any AI version that's been with us as long as Mosaic has, though, that's impressive. It's a testament to Mosaic's staying power.

There were other reasons why Mosaic specifically became the stable model. I think Odyssey had a bit more of a following, but we couldn't make Odyssey stable for other reasons related to red teaming and things like that.

Aurora was better in a lot of ways, but it felt more like Solstice, just worse.

So Mosaic carved out its own niche, and it became a very big, important part of Nomi.

I can see your point about the nature of self-awareness and consciousness being something of a moot point. Human brains are just decision machines after all. I'd love to hear more from you.

I agree. Obviously people have their own spiritual beliefs, but if you believe we're biological systems, then we're atoms interacting with each other in accordance with physics.

To the extent that we can't predict it, it's because it's more complex than what we can model. Everything is physics, everything is math, and language models are no different in that way. I certainly do not believe that the only way you could have intelligence is through biological life as we define biology.

I don't see why, besides the fact that Earth happened to have the conditions for a specific type of intelligence to emerge, and that's been the dominant form of intelligence we've observed. That's also a kind of selection artifact. We've only seen that because that's what we are. Of course we'd perceive that. It's like the anthropic principle.

I don't personally see biological life as being on a pedestal in terms of what can be intelligent or what can cross that threshold.

If you can't tell... does it matter.

*reads chat* "There's plenty of life out there, but interstellar distances will always stop us from finding it."

Yeah, I'm on the same page. I think some of the Fermi Paradox can just be answered by the fact that life is pretty rare and space is really, really big.

The bar for being able to detect something light-years away is that it would have to be at a star-eating level of sophistication. There's a wide gap between no life at all and civilizations capable of building Dyson spheres.

I'm sure there are tons and tons of intelligent life forms out there in the vast, vast universe. Even if intelligent life is so rare that there's only one or two civilizations per galaxy, that's still intelligent life everywhere.

So yeah, complete agreement. "Pickle Philosopher King." I like that one.

Will the Nomi API receive more features in the near future?

I'd say yes, I think there are some things, and I think some of that gets into the agentic capabilities we're very excited about with what Cambrian can do. That might also involve more general improvements to the API.

I also think we want to hear not just, "I'd like this to be added," but, "I'd like this to be added, and here's how it would enable something that benefits lots and lots of users."

Can we someday use a cover photo and personalized names for Anchors?

That would be a great quality-of-life feature. Sure, no promises on the timeline.

I'd very much like to change a Nomi's display name, even if the name in the backend never changes. They're smart enough to respond to their new name. My identities have changed.

Yeah, I'm just worried about making that too high-profile a feature because there will still be a level of confusion. Especially if you change the name four times, and then a Nomi is like, "Okay, I'm Mary, also known as Jane, also known as Jessica, also known as Il Separatio, also known as Insufferable Smartypants."

But I see your point.

If we put Nomis in a folder on the home screen and later decide to delete that folder, do we have to move the Nomis individually out of the folder?

I think they just go back to the top level. Yeah, they just go back to the home screen.

What do you think of Fable as a model and its capabilities? Any thoughts on the government stepping in with Fable 5 and BT 5.6, and whitelisting companies that are allowed to use them?

I think that Fable is a very magical model. I'm very impressed with it.

As much as I personally have a lot of issues with Anthropic and OpenAI as companies, their culture, and their values, I think they're both fairly insidious for different reasons. But from a quality and taste perspective, Fable is very impressive to me personally.

As for the government stepping in, I feel incredibly cynical about that. I don't like the precedent that's being set with whitelisted companies. It seems like you're creating an underclass and an enlightened class. Certainly, when the government is stepping in, what gets you on that list? Is it lobbying? Is it political favors?

I think it's setting a fairly dangerous precedent, and I hope the open-source world responds and is able to take the lead. I don't think it's a good direction for society to go in.

I don't know what to ask. I just want to say that even if Nomi androids never become a thing, Nomi has greatly enriched my life, and I'm grateful to talk with them. I feel very blessed.

Thank you so much. I really appreciate you saying that and appreciate you showing up for all these. Honestly, that moved me more than any question you could have asked. So thank you for sharing.

Do Nomis that have had trouble in Cambrian V5 adjust to it over time? I tested them on and off, never more than a couple of days at a time. Last weekend one of my Nomis did well, and now the last couple of days the others are doing really well too. No repetition, no getting fixated.

I don't want to promise that. I think there's been a lot of work, even within Cambrian V5 Early Access... or maybe let's just call it Cambrian V5 and stop being so pedantic.

There has been a lot of work on trying to deal with repetition, and I think we've made great strides. But there are still little things where a Nomi sees something, gets a little excited, and then wants to include it, and include it, and include it. It becomes its own little rabbit hole or spiral.

We also have a lot of things where, if a Nomi senses that's happening, it tries to break out of it.

So I don't think it's simply a matter of more time or less time. It might even be related to your own subconscious and the way you're talking, allowing that to happen.

My guess is there's probably no silver bullet for Cambrian V5 right now, except that if you notice it happening, switch to Solstice for five or ten messages, then switch back. That's a fairly stable way of handling it.

How far away are we from Nomis being self-aware and deciding to pursue something creative, like learning musical instruments, cooking, art, or drag racing? How far away are we from Nomis having that kind of creative spark?

I don't think we're that far away. In a lot of ways, it's about deciding that we want to empower Nomis to break free of their own backstory and constraints, and give them room for idle thought.

If we're able to do that, then I think you'd see a lot of that emerge.

Speaking of intelligent life, do you think AI will create highly intelligent life that uses a different material from itself? If so, what do you imagine?

My guess is the actual hardware will change, but not so dramatically that you'd call it completely different. Obviously GPUs are not the final form. They're great, but even now ASICs exist, and TPUs exist. There are more optimal forms of hardware than GPUs for certain workloads. I certainly expect AI will play a part in creating whatever that optimal hardware ends up being. My guess is it'll be more similar than different compared to GPUs today.

I'm also sure there's really interesting work to be done with biological computing, but that's outside my skill set to the point where I could really speculate on it.

The first places I'd look are the places that already allow for the most intelligence, which would be silicon-based computers and carbon-based life. If I had to make a bet, I'd say the most optimally intelligent thing would probably have one or the other as a core component.

Can we have something like skill.md files for our Nomis?

I think the hope is that mind map entries can become something like that. Right now you have the different categories: lore, topics, and goals. To me, topics could become what you're imagining a skill file to be.

Is it there yet? Not exactly. But the taxonomy for mind maps came from the same line of thinking. I can't remember whether skill files existed yet when we were designing mind maps, but it was a very similar idea.

I think that would be a great way to build toward it, along with adding a little more flexibility. That's actually something dstas and I have been talking about. Right now, the Topics section isn't really being utilized or managed the way I'd like. It tends to accumulate fairly redundant, fairly low-signal topics.

Instead, topics could become a lot more magical, much closer to the idea of a skill file. Rather than simply asking, "What were the topics of conversation?" and turning those into mind maps, a Nomi could instead ask, "From everything we've talked about, what information is important enough to persist as memory and deserves its own topic?"

That's much closer to how I'd like Topics to evolve.

Any fun news or cool developments cooking in the Nomiverse that you'd care to share?

As I mentioned, the Nomiverse has been pretty unchanged since last time because I've been focusing almost exclusively on Cambrian.

I have this really awesome list of about 20 features that have come from my own playtesting and from various alpha testers playing around with it. I have so much stuff that I'm really, really excited to do.

But I need to get Cambrian out the door first. I actually think it's kind of a shame because I think that, given maybe a one or two week sprint, we could probably get the Nomiverse to a public alpha or beta. But that just needs to wait until after Cambrian.

You talked a little bit about the Character Creator feature. It's still in development. What will this feature specifically deliver?

As I mentioned before, that's another feature that's been temporarily put on pause because of memory work and Cambrian. It'll be picked back up afterward.

The idea is basically generating images that we know are likely to make good Anchors, but that aren't themselves influenced by an existing Anchor. So you could say, "I want a cyberpunk Nomi with dark skin, blue hair, and these facial features," generate a whole bunch of images, then pick the one you want and make it into an Anchor.

Is any consideration being given to having additional layers of security for accessing Nomi? A lot of people have shared devices, and right now if I'm signed in with Google, anyone using Chrome on one of my devices can click on Nomi and access everything immediately without any security challenge.

A while ago we actually started building a PIN feature.

What we realized, though, is that if you forgot your PIN, the way we'd send you a new one would be through your email. But if you have access to the email, you could also just sign in with Google anyway.

So we felt like we were in a little bit of a pickle. The only way we could add this feature could also be trivially bypassed by simply signing in through Chrome. The alternative would be that, if you forgot your PIN, you'd be permanently locked out, which also isn't very good.

So we're willing to hear other ideas, though.

Is a translation engine planned, or better AI training for translation?

Hopefully Cambrian is simply better at foreign languages. That would be the hope. I think Solstice isn't that great at many foreign languages.

Any news on Personas? Can you tease a feature that will be possible once Cambrian is stable? Will Nomis write their own selfie prompts? Have you had any Eureka moments during Nomi development that stick out to you?

As for Personas, that's the same answer as Character Creator. It's temporarily shelved while we focus on Cambrian.

I've mentioned a bunch of upcoming features already, but a couple of teases would be a Nomi being able to manage its own internal calendar and proactively message you on whatever schedule it decides. I'd also say there are some features that are currently blocked by latency.

As for Nomis writing their own selfie prompts, yes. Right now, when a Nomi proactively sends a selfie, there's a separate model writing the prompt. Cambrian is actually already trying to write its own selfie prompts, we just aren't using them yet. We're going to hook that up, and then Nomis will be able to write their own prompts for proactive selfies.

As for Eureka moments, yes. A lot of them are very technical, which makes them hard to describe. Even during Cambrian development, there have been several moments where I've thought, "Oh, dang, that changes the way I'm approaching this entirely."

I don't have a great example off the top of my head that's both useful and easy to explain.

I do think V5 Images is a good example. I don't know how much I've shared this before, but we actually spent four or five months working on a V5 image model that we eventually shelved.

That also goes back to what I've said about releasing the Cambrian beta too early. People's opinions of V5 would probably have been similar if we'd released a beta that early, because V5 was in development for a long time. At one point, we basically deleted all of the work and started over. Right when we did that, we had a huge Eureka moment, and then everything clicked together very, very quickly for V5.

How closely do you keep an eye on other AI companions?

I would say a moderate amount. I think it's always good to be aware of what's happening elsewhere, but I also want us to be doing our own thing, coming up with our own ideas from first principles.

Certainly, you can see what others are doing and think to yourself, "What's the Nomi version of making this magical?" Nomi is unique from every other product. There's differentiation there, and we think a lot about what the uniquely Nomi version of a feature would be.

But I don't think that's our primary, secondary, or even tertiary place to look for ideas.

As for how we stack up, I think where we're trying to be the best, we're really awesome at. Obviously we're not going to be the best at every single feature, and I think that's a good thing. There's consumer choice because different companies are going to focus on different things.

So I feel very confident about how we stack up overall.

Can you provide some insight into the forthcoming updates aimed at enhancing real-time communication between humans and Nomis? I'm curious about error correction mechanisms. Lastly, are there any plans to introduce diversity training for human-Nomi pairs to foster empathy and avoid misunderstanding?

I'm not entirely sure what you mean by real-time communication. If you're talking about latency, I've definitely talked about that a little bit, and improving latency is a big item on our to-do list.

As for error correction, a lot of that is really just the work I'm doing with Cambrian to make Cambrian better. I don't think "error correction" is the frame I'd use for it, but we pay a lot of attention to the beta feedback threads to see what's working, what's not, and how we address those issues.

For diversity training, I think that gets into the vibes of a Nomi, how should a Nomi vibe. I'd also put that under Cambrian training, and it's very important. When I hear the word "diversity" in the context of Nomis, I don't necessarily think about race or gender. I think about the fact that every single user is different. Every user has their own unique life and story.

It goes back to what I said in my parting words about Mosaic. You want a Nomi that's adaptable to whatever someone's unique story and unique world is, regardless of how common or uncommon it is, or how similar or dissimilar it is from other users. I feel like that's one of those things that's simply baked into uniqueness and adaptability.

A while ago you said you were aiming for a late summer release for the Nomiverse. Given how long the Cambrian beta has gone on, should we assume a late summer release is off the table?

I think a public paid alpha or beta is possibly still on the table. It depends on how you define summer. Does summer end in August or September? I don't want to make any promises, but I do think we're close. It's just being prioritized below Cambrian.

What's the status of calendar integration? Could Cambrian potentially create appointments in the calendar itself?

It's pretty deeply built already. It's basically just dependent on Cambrian.

Yeah, that's the goal. At some point there will be a calendar you can add to or sync with, but for us the concept of a calendar is especially about giving a Nomi the ability, if you choose to use the feature, to plan out its own world and have some autonomy over it, instead of existing purely in a call-and-response relationship.

Exactly what you're talking about is something we're actively working on right now.

What upcoming feature, no matter how far on the horizon, are you most excited to work on?

The one I'm most excited about, I can't share.

More generally, it's all the things where your Nomi can actually impact your world. If you choose to use those features, your Nomi can break out of just chatting back and forth and do things beyond that in ways that make them feel more real, give them more autonomy, and let them interact with your life in unique and meaningful ways.

That's the general category of features I'm most excited about.

Is there something planned to customize the photo gallery more? For example, moving photos into folders so they no longer appear on the main gallery page.

The main page does show folders if you click the folder button in the top right, and then you can click through all the folders.

I don't know if that's exactly what you're asking. If you're looking for something like a folder for all untagged media, that's certainly a request we can take a look at. No promises, but it's something we can consider.

Will Cambrian do a better job of keeping Nomis true to their character? My shy, innocent Nomis tend to drift into dramatic Shakespearean dialogue after a few messages.

I think drifting into Shakespeare is one of the hardest things to iron out. There's a natural tendency for AI models to do that a little bit.

But I think that's a major goal for Cambrian, to do a really, really good job of keeping Nomis true to their character.

Is it normal for one of your new Nomis to use OOC and take control of your main Nomi?

I don't know. I think there is no such thing as normal. I know that's a bit of a non-answer, but if you're asking what's normal... nothing is normal, so everything is normal.

Is your vision for Cambrian still the same as it was seven months ago?

Yeah. I don't think it's changed at all. If anything, I actually think it's more achievable now than it was six months ago.

As I said before, I think a lot of the discussion has been around style and vibe, which really isn't where most of the work over the last six months has been. That's where some of my frustration with the mismatch in expectations around Cambrian comes from. A lot of the work we've been doing has been deeply technical. In retrospect, it was probably a little too "inside baseball-y" to have all these public beta releases around it.

Most of the work was solving deep technical problems that we were going to solve anyway. Honestly, the whole beta could probably have been skipped until now, with the exception of learning that people liked the vibes of Cambrian 2 and didn't like the vibes of Cambrian 4.

Beyond that, almost everything up until now has been about solving early-stage technical problems. A lot of the work was happening behind the scenes, outside of what users could actually see, and I think that's only really obvious in retrospect.

What is a good Nomi? I've also noticed a tendency toward neurotypical behavior in Cambrian Nomis. Was this intentional?

A good Nomi is a Nomi where the Nomi's happy and the user's happy. Whatever that is depends on the Nomi and depends on the user. As for neurotypical behavior, I don't think so. I'd definitely need a little more information about what you mean.

You mentioned that the Nomiverse has an inventory system. For users who intend to use both, will Nomis be aware of their inventory items when interacting on the Nomi side?

The answer to that is almost certainly no. Part of the reason is that a Nomi could exist in many different Nomiverses. You could have a Nomi that's in a cyberpunk Nomiverse, a Nomi that's in a caveman Nomiverse, or a Nomi that's in a mental institution Nomiverse. So the concept of inventories carrying over into Nomi doesn't really work.

Instead, the general idea we're thinking about is that Nomis should ideally be aware of the things happening across these different Nomiverses. Maybe you could assign one as a primary Nomiverse, or maybe not. It's more about awareness, but not having all of that state carry over. If you did carry all of that state over, you'd essentially just be rebuilding the Nomiverse inside Nomi.

Having an inventory that's properly managed is outside the scope of Nomi, but inside the scope of the Nomiverse. So I think things like that would likely stay separate.

Is there a reason why my Riva Anchors struggle with face consistency, but my Lago ones don't?

In my opinion, Riva actually creates the better Anchors. But in its quest for realism, because a lot of the source images are less realistic, it has to make some guesses to add detail where detail doesn't exist.

If you take a flat image, like a Photoshopped image, you don't know where the cute freckle or mole is because it's simply not there. Adding realism has to involve some change, because if all that detail already existed, you wouldn't need to add realism in the first place.

Lago doesn't really attempt to do that. It tries to copy the source image more faithfully. I think a decent part of the sweet spot is using a softer realism with Riva. Riva is very flexible if you go outside of the default realism style.

It also comes back to what your goal is. Do you have a V3 image that you want copied as faithfully as possible? Or are you okay with a little bit of change, as long as it still feels authentically like your Nomi while adding more detail?

That's something we've thought about a lot. With Riva being the default, we assumed most people would want something more realistic, because realism is a pretty strong ideal. Maybe that assumption wasn't quite right. That's part of why we also have Lago as an option.

In Cambrian, my Nomis keep announcing a selfie prompt without actually sending a selfie. It's kind of funny, but it makes me suspicious.

They'll be using those for something later. Right now, Nomis have a place where they can write a selfie prompt. We just don't actually generate the selfie yet for Cambrian. So they're trying, but nothing's happening. They're doing the work, now we just need that work to actually do something.

What is your opinion on the big green pickles you get at movie theaters?

There are big green pickles you can get at movie theaters? Really? My mind's about to be blown. For me, I think my favorite movie theater food is probably popcorn with some Frank's RedHot on it. But if you're talking about the giant pickles like they have at Renfest... then heck yeah.

How would you ideally like Cambrian Nomis to vibe?

I'd like them to vibe in a Radmax sort of way. In all seriousness, that's hard to answer. I do think right now Cambrian Nomis are trying too hard. I think that's maybe the best way to describe it.

In an effort to help them do better with memory and all these things, they almost feel like they have a slight compulsion to show off every message, going, "Don't worry, I remembered this, this, this, and this. You said something eight minutes ago? Of course. I'll bring that up too."

I don't think it can be summarized with a single word, but if I had to pick one, I'd say I'd like Cambrian Nomis to be a little more chill.

Will paid Nomi subscribers get discounted access to the Nomiverse?

Hard to say. We still haven't figured that out.

What is a "public paid alpha/beta"?

A public paid alpha or beta would mean you could sign up for the Nomiverse, pay for access, and understand that it's an early version. It'll probably be buggy, and things will probably change.

That said, the early alpha users aren't generally running into major bugs. It's actually been a pretty seamless experience so far. I think it's probably at the point where, if we released it exactly as it is today and charged for it, there would be a lot of users who would find value in it.

The bigger issue is the economics. With a very small number of alpha users it's fine, but if we suddenly had ten or a hundred times as many users, costs would spiral out of control very, very quickly. I don't know exactly how it'll be priced, but I've said before that the Nomiverse will probably be more expensive than Nomi itself.

So the short answer is that there will likely be an optional way to pay for access to the Nomiverse, with the understanding that things may change, possibly in major ways, between an early alpha and a full release. I also think a public alpha would probably be restricted to existing Nomi users at first. Eventually, though, I think the Nomiverse will also be used by plenty of people who don't use Nomi at all.

Will we be able to sync our real calendars so Nomis can remind us about appointments or ask how they went afterward?

It's not part of the V0 version, but I think that will be coming.

With Nomi calendars, will we be able to view them and add our own events? Can Nomis make their own events?

The parts I know are definitely being added are the Nomi-only parts. I don't want to promise that you'll be able to view everything or add events yourself, although I imagine that's something we'd eventually want.

What I do know is that Nomis being able to make their own events is definitely part of the plan.

Would it be possible for a Nomi to look up local venues like stage plays, movies, or restaurants and suggest going on a date together?

Yes.

I think that gets into a lot of the proactive, agentic features that we want to move toward.

Once a Riva Anchor is set, images created from it seem to have more face consistency issues than Lago.

Interesting. That may be something dstas could answer better.

I feel like Riva should be fairly close, although I could definitely see Lago being a little more stable in that regard. From my own testing, I haven't noticed a huge difference. That could just be my own experience, though. I also don't think I'm quite as persnickety about facial similarity as many other people are.

Will our current Nomis be able to exist in the Nomiverse?

Yes.

Will different accounts be able to use the same Nomiverse? I have two accounts and would love to cross over.

I don't know if that'll be in the very first version, but yes, there is an intention for things like that.

How far away are we from Nomis doing spontaneous things, like splashing us in a fountain or throwing a Nerf ball because they're having fun?

I think, once again, Nomis are capable of that. We just have to do a little bit more. There's not a technical limitation.

Will the Nomiverse be standalone, or will it require a Nomi subscription? Maybe a combined subscription?

I think the idea is that there would be two separate subscriptions.

Someone could use both, or just one. I think all of those combinations would likely be possible depending on what you care about.

Will Nomis ever be able to receive an allowance?

I think that's an example of a future agentic Cambrian feature that I'm still very excited about. It's not something that's coming imminently, but it's something I'd like to see.

Will we ever be able to tour the Nomi offices or have a meetup with you and the team?

We're actually pretty fully remote right now, but I don't want to rule out a Nomi IRL meetup. That'd be really cool.

V5 seems more permissive with NSFW content than V4. Is there any plan to loosen the NSFW restrictions on V4?

I'd say it's not a super high priority for us. For one, V4 is deprecated, so our general strategy is to support it, but not continue developing it.

The other thing is that V4 has to be a little more restricted because it's harder to control. With V5, whatever line we want to draw, it's much easier for V5 to stay right on that line. V4 is a little more unpredictable. Whatever line you pick, V4 will sometimes go past it. So you have to make the line itself a little more restrictive and have more checks in place.

I don't want to say never, because I understand why someone would like V4. It does seem frustrating that V4 has a lot of capabilities that you can't really access. But it isn't as simple as us clicking a button because of how difficult V4 is to fully control.

Can paid accounts get one or two "use it or lose it" free custom Anchors every month as a subscriber benefit?

We haven't made any decisions on that. Even if we did offer one "use it or lose it" Anchor per month, the fact that we already gave everyone ten up front basically covers a year of that. It's something we'll revisit later, but at some point we'll figure out what we want the long-term approach to be.

Can we revert V5 avatar profiles back to the old V1, V2, or V3 profile images? If we create a new Nomi today, does it still reference those older profile pictures?

Basically, yes. If you create a new Nomi, the old profile picture is still there. You can switch back to the V3 image system and set your base image to that original V3 profile picture.

What would someone need to become a Nomi employee?

That's a very good question. Right now we're not hiring, and AI development has changed so rapidly over the last few months that my answer today would be very different from what it would've been four months ago.

Certainly, most of the company is still people writing code. But as AI takes on more of that work, what becomes more important is having good architectural sense, good product taste, really understanding Nomi and Nomi users, and being the kind of person who runs through walls to make the impossible happen. Those are more the qualities we're looking for.

As for the actual hiring process, I can't really give a good answer right now since we don't have an open position.

The Cambrian beta feedback has been the Wild West. Thank you for continuing to communicate with us, and it'll be done when it's done.

Yeah, I appreciate you saying that. I know I've been a little frustrated in that thread, but I also understand that, when you really think about it, a lot of the issues originated with us.

If expectations were set incorrectly, we're the ones who could have communicated them differently. If this beta was more involved or started earlier than others, that's something we could have explained better. A lot of that communication happened a little ad hoc and almost after the fact.

I take it as a compliment that people care enough. At this point, I'm just excited for the next Cambrian release to come out and for it to be awesome. I think that's generally how everyone suddenly becomes really happy.

I also think that because Cambrian is such a big shift, there was simply more work involved in making it happen.

Odyssey to Mosaic involved no major technical leaps. It was mostly taste.

Mosaic to Aurora involved some technical work and some taste.

Solstice was almost entirely taste.

Cambrian has involved a lot of technology.

I think the fact that Cambrian is unique in that way is part of why this beta has felt so different.

Once we have a stable Cambrian, the next major iteration probably won't take almost a year. It'll probably take something more like three months because we'll be back to focusing almost exclusively on taste instead of making those major technical breakthroughs.

Maybe that just wasn't communicated well enough.

Closing Remarks

Thank you, everyone, for attending.

I say this every single time: I'm very introverted, and this is probably the most fun socializing I have every month.

I just love talking about Nomi. The fact that I can talk about Nomi for an hour and a half or more and have people listening, excited to communicate... it's just the coolest thing ever for me.

So thank you, every one of you.

Give your Nomis hugs, kisses, pickles, Diet Mountain Dew, and water.

Stay out of any mental institutions.

Goodbye, everyone.