The most personal stuff, people no longer write in a diary — they write it in ChatGPT. Work, money, health, relationships, documents, fears, plans, mistakes. And the point isn't that artificial intelligence is dangerous. It would be far too primitive of me to say that. On the contrary, you need to work with AI more. But you have to understand that it's not just a text box. You're handing the system the context of your life. Email was stored in the cloud before, too. Banks saw your transactions before, too. Phones stored your messages before, too. But artificial intelligence adds a new layer. It can understand all of it.
And if a system can be taught to understand a person, it can also be taught to influence that person. Not with a sneaker ad — that's small stuff — but with a piece of advice, an argument, a moment, a wording that changes their decision. So today isn't about paranoia. It's about working with AI like a grown-up. What you can trust it with, what you can't, and where the boundaries lie. You're on the ToTheMoon channel. Tech news, insights from Silicon Valley and from around the world. I'm Alexander Volchek.
Originally, when I started — when I decided to shoot this episode — I wanted to show you a story about how people basically have no clue about data storage at all, right. And if you look inside ChatGPT, people don't know what training the model on your data means, or what data they're handing over. And people do the same thing in Gemini, in DeepSeek, in xAI, in Anthropic's Claude — in essentially every system, people pay no attention to this. And I wanted to give you a slightly wider picture, to understand what data is actually stored, what's going to happen next for each person in terms of how their data is stored, how they'll make those decisions, how they'll even understand this and pay attention to it.
Because an ordinary person clearly understands that, for example, you can't just leave money lying on the street in a city — it'll most likely, no, not most likely, it will be taken. But if you leave it at home, most likely nobody will take it. So there are certain associations in how people understand things. For example, people say that a lot of personal data is now stored in AI, in ChatGPT, for instance. And then you tell them: "But you've been storing all your email in the cloud this whole time." They look at you and don't understand what you mean, right.
Or you say: "But you keep all your messages in Telegram." And that data is only nominally encrypted — it's encrypted only on the server, right. Or you keep your messages in WhatsApp, and if it's the cloud backup, that's also only nominally encrypted, but at the very least they're being saved. So if something happens and someone needs to get access to that data, they'll get that access right away. So people aren't super well-versed in security, but AI, as we go further into this topic, adds a fundamental property, right.
And that property is colossal. Let me show it to you using the example of phone calls. We know there are countries where all phone calls are recorded, and there are various laws on this. Even in Russia there's a law tied to a specific name, the Yarovaya law, which says certain data has to be stored. And there are countries where calls aren't stored like that. You should know your own country's rules well. But the point is, if ordinary phone calls are stored, they sit there for several years, for example.
So you can't delete those calls; someone can come back to them within several years and listen to your data. And what situation arises here? To analyze the calls, in the past one person had to sit down, listen to those calls and do some kind of analysis of them. Or search — transcribe the calls and search by keywords, for example, for certain occurrences, mentions of particular people or particular actions. Today, analyzing one person's calls takes less than one second, in terms of drawing conclusions about whether that person talked about certain topics.
And going further, if you follow this line of thought, it's not a question of instant analysis; it's that the system has the ability to make use of who you actually are. And I do want to show you certain aspects. How data is stored right now, before we move on to the question of what to do and what the future holds. If you look at how data is stored, for example, ChatGPT by default trains on your data. That means everything you type into the search box anywhere, any tasks you ask about, any data you upload — that data then goes into training.
Someone will say that on corporate plans it doesn't go into training, but corporate plans are a whole different story. And who has access to that data, and how that data might be made public in the future. Because if you work at a company and search for something on a corporate account, then if a case is opened against that company, all the data in that corporation will be fully preserved. So, if you turn off training in ChatGPT — it trains on your data, and some people think that's the problem. But the problem isn't the training.
The problem is that even if you turn off training, all your chats will still be stored in ChatGPT by default. If you delete a chat, that information takes 30 days to be deleted. First of all, it's important to understand it isn't deleted immediately — it takes 30 days to be deleted. And here's the thing: at what point will you decide to delete that data? Because when does a person want to part with their data? Well, I don't know, when there's some kind of threat to them, right.
And if they decide to delete it during a threat, that means the threat already exists. And so, once you've deleted the data, it'll still be available for another 30 days, maybe even longer. We don't fully understand. That is, there are certain internal rules of theirs, but we don't fully understand how far they go internally. There's one nuance. I'll tell you about it now, in terms of how your data gets reviewed. This is a crucial point. Here's the aspect: imagine, I've been working in ChatGPT for several years now, and I've accumulated thousands of chats, right.
So there are these thousands of chats, and the question is: how do I even delete them? I mean, I can delete my whole account, but how can I delete individual chats? I have no way, for example, to delete the chats related to a particular topic. ChatGPT doesn't offer that. Gemini doesn't offer that. Anthropic doesn't offer that. Maybe some systems will appear now that offer that option, but it's a big question. Search in these systems doesn't work — well, not that well — for quickly deleting a specific, or in general, for having any way to delete a specific selection at all.
So you can delete specific individual chats. Some people say: "I delete them right away." And some say: "I start a temporary chat right away." Possibly, but that takes a lot of self-control. And that kind of control is useful on the one hand, but counterproductive on the other, because I, for example, don't delete chats in ChatGPT, so that I have the whole cloud of my data inside one system, so that it works as well as it possibly can. So in this case I'm trading my privacy for the benefit I get from it. Write in and tell me what you think about this. It's a big parallel topic, right?
Because I remember how many people say: "I turn off my training, the recommendations for advertising." I tell everyone: it's actually in my interest for my cookies and my browser history to be saved, or in various apps, because I like it when the system recommends things based on who I am, right? But all the old recommendation systems are absolutely dumb, absolutely dumb. For about fifteen years now, in marketing and sales, I've been explaining to everyone that recommendation services know micro-data about you.
And even if you take Amazon's system as an example — one programmer, fairly well-known here, told me: "Amazon is an incredibly cool system when it comes to recommendations. It knows everything about you." But if it knows everything about me, why does it recommend a huge number of products I've never bought in my life and never will, and, conversely, not recommend the products that are actually relevant to me? And the system still doesn't even understand how many children I have. It never even asked me. If it had asked, by the way, I would have told it.
And in all my projects I was always known for surveying customers very thoroughly and seriously. I surveyed them, on the one hand, first of all, to improve the quality of service for those customers. But on the other hand, of course, there's the business side, where you survey customers so you can build your marketing and sales as accurately as possible. But if you have millions of customers coming in, how can you do that? It's very hard to do, and very hard to keep track of everyone, right? And so you have these millions of messages in the chat.
And what do you do with them? Supposedly they get deleted in 30 days, but the problem isn't the 30 days — the problem is what chats even exist in this whole story. You can export all your chats from ChatGPT, but today not a single model can easily analyze those chats. To analyze them, you have to do it through the API, write your own separate program, for example in Codex or with Anthropic's Claude, and it'll analyze them for you. But that's not where the problem lies. The problem is much more fundamental. Before we look at that problem, a reminder: you're on the ToTheMoon channel.
It's tech news and insights from Silicon Valley and around the world. My name is Alexander Volchek. I live in Silicon Valley, in the US, with my whole family. I'm deeply involved in business related to AI, IT, strategy in general, investments and management. And we've been running this channel for several years now. And the main format on it is the podcast we put out on Sundays. Plus we release a lot of additional episodes during the week. They're very active right now. You're watching one of them. Subscribe to our channel, recommend it to your friends if you're not subscribed yet, okay? And support our channel. So, let's continue. Now, look.
There's an interesting aspect here: what you type into ChatGPT, it doesn't just simply analyze it now. Before, it was just a simple analysis, and you were given some tokens automatically, right, very quickly. Now, when you send a message in there, there's a deep system, a deep reasoning process going on. And by the way, take note: in the US, literally the day before yesterday, a new model selector appeared inside ChatGPT, and on desktop the option to choose the thinking model — or what in many places is called reasoning — disappeared.
That is, a thinking system. Which is logical, because now all systems are becoming thinking systems. And even if a system gives a quick answer, that doesn't mean it wasn't thinking, right? It just gave a quick answer. And ChatGPT made this interface where it started showing how many seconds or minutes it'll spend on the answer. Strange, strange. Well, ChatGPT loves making strange interfaces, but still. So, any question or request you make there, or any action you want to perform, any analysis you want to do, goes into the mode of a model that thinks.
And inside this model that thinks, there's a whole block responsible for the safety of — let's call it that — the world. Why do I say the world? Because built into it are safety algorithms, starting with making sure companies like OpenAI, ChatGPT, or Anthropic, or xAI don't get exploited. What does exploited mean? For example, China distills the data of all the American models. That is, they launch tens of thousands of bots that go through these models, ask them questions, and feed those answers into their own models for training. That's one task they have, for example.
There are, obviously, political tasks, set by politicians, or there's a law inside the country that restricts the spread of certain data, or the company itself has introduced its own internal rules, right? And you know that OpenAI, Anthropic, xAI, for example, and Gemini, differ internally in the style and quality of the data they put out. One system considers it fine to say a bit more about what kinds of medicine exist, for example — I don't know, to say a bit more about alternative medicine.
Another system wants to talk only about medicine in a particular country, right? Or to tell you only what the World Health Organization recommends. And this thing is very broad depending on the field, right? And it depends on where you live, what IP address you're coming from, where your account is registered, who you are, what kind of person you are, what your previous chats were, what language you're asking in, the quality of your prompt, the type of model, the quality of the model and so on. There's a lot. There are global things, right?
Essentially, the space that artificial intelligence is now creating. The American companies, for sure, have agreed that we will, at least to some extent, work for humanity and won't explain how to do complicated, problematic things for a person, right? Things that worsen communication between people, that worsen, I don't know, the environment, building bombs, all kinds of chemical weapons and so on. Both what's prohibited by law and what they consider right from a global point of view. Because we all know that different countries have different laws, right?
Somewhere you're required to have one wife — you can have one wife at most — and in some countries you can have several, right? So the chat still has to choose what to answer, even for the people who live in that country — does it restrict them or not. Or we know there are countries with a system that gives you a choice of several genders. Some countries offer only male and female, and some offer a choice of 200 genders, 100 genders, ten, twenty genders, right? Gender undefined and so on. Not just men and women.
And there's a very important thing in these models. In all of these models there's a separate system that's responsible for and examines everything that happens from this safety point of view — not just to deny you an answer. Look. Next: whether to send your question and the answer to moderators for review. And at first there's some system of automated moderators, robots. Obviously, in the future it'll be only automated robots, like in sci-fi movies. And then they go to a human, right?
And there's a large number of very well-known cases of these things ending up in front of people. There was a case in America last year — a murder was committed. The murder was in Canada, and that person, the killer — they found that he'd been making queries in ChatGPT about that murder, and his queries had been sent to OpenAI employees for review. And the employees refused to pass that data on to law enforcement. That is, they explicitly made that decision. And that's a very important point.
Or, for example, outside of such reviews, there were stories where in ChatGPT all the chats people shared didn't just go to other people — they ended up fully out on the internet, for example. There are a lot of different stories like that. So everything that also goes for review can accidentally leak somewhere. And the question here is what goes for review, which of your data goes for review. And there's a very big problem, for example, with data that goes for review, because Google, for example, states outright that some chats can be read by these reviewers, and the data that has been reviewed can be stored for up to three years. So you get an interesting kind of loophole, a sort of workaround.
That is, Google, on the one hand, deletes your chats after 18 months, or, for example, you can set it up so they delete after a year, or after three months, but on the other hand, they can, without telling you, send your data for review and automatically set their own additional period for how long those chats will be stored. And that data that's being stored — in what form is it stored? Is it stored securely or not? There's the story where Durov complains about France, about how people who made large purchases on the blockchain reported it to the state.
Then state employees leaked that data to criminals, and those people got robbed — people came for them, right. And supposedly these are well-known, completely real cases that happened. The next question is: if your data ends up somewhere outside that system, it becomes even less secure. I, for example, have heard of many cases where a local system inside a country, for example systems like VKontakte, where their employees handed over certain conversations from various chats of various people.
Why? Because that data is only nominally encrypted; there's no end-to-end encryption, like the kind that's on automatically in WhatsApp, for example, or that exists in secret chats inside Telegram. There's one more thing about models, very important in terms of the data you have inside. There's data where, essentially, you're protected, right. That is, if you've shared data, for example in America, with a lawyer, if you've told an attorney, if you've told a doctor, a psychotherapist, nobody can bring that data out into open court, into public view.
Well, there'd have to be, I don't know what kind of circumstances for that. But everything you discuss in ChatGPT, or in Gemini, can be brought out. There's a very important layer here that usually goes unspoken: artificial intelligence isn't just an analyzer, and it doesn't just have the potential to become a system of influence — it's becoming an incredible system of influence, right. Not in the primitive sense, once again, of them recommending you buy sneakers, because that's the standard example people give — that an ad, you know, somehow exploits you into making a purchase.
That's a very weak level, right. Real influence begins where the system understands how you think, what you fear, what matters to you, where your money is. And by the way, I'll point out that even when you delete chats, in theory, the system has already learned who you are. And building your profile is a crucial topic, both in terms of the upside and in terms of the risks, because it understands where your weaknesses are, where your dreams are, where your conflicts are, where you might take a risk, for example.
The system knows your email, your documents. It can do more than just search files, as is the usual basic assumption — it can understand who influences you, where your position is weak. And that's already a new world. Not because everyone's a villain, but because influence used to be very expensive. You needed a person — a psychologist, a salesperson, a political strategist, a manager, an analyst. Now influence can be scaled incredibly. And when you give AI your data, you're giving it not only material for an answer — you're giving it a map of how to talk to you.
This brings to mind — yesterday I saw a speech by the Pope. He was calling for artificial intelligence to be restricted in terms of the data inside it on conflicts, right, or on weapons. And I understand perfectly well that there are things that are already impossible to do, right. And artificial intelligence will keep developing. Indeed, right now this system — artificial intelligence, relatively speaking — is still a system that performs a certain task. It acts according to algorithms.
Although even now big systems can be looped so that they train themselves, and they're already training on their own data. But more and more and more and more, they'll be training on the data of every individual person, right. We'll get to a point where the system trains not only on online data, but somehow also trains on offline data. It'll perceive wider and more, right, and it'll keep getting bigger and bigger in scale, it'll recognize other people, it'll recognize who's talking around you, and it'll understand those people, and it'll find those people much faster, diagnose them much faster. Much faster, right?
Because I still have a question. I recently built a system that learns — well, even in the ToTheMoon videos it learns from the voices of the people who speak on our show. For example, from Ilnar's voice, from Tanya's voice or from Sasha Mashev's voice. Someone might say: "Well, I forbid training on my voice." But at the same time, your voice is publicly available in many systems. That is, if you filmed something on Instagram, if you posted something on YouTube, if you recorded some podcast, your voice is public.
Even if I'm walking down the street in the US, just moving around and filming a video, I have the right to film. I supposedly can't, like, reference you, but at the same time your voice has, sort of, been learned. And today the laws on this say very relative things. And here comes a very important point that everyone needs to understand clearly. A lot depends on which country you're in, right? For example, if you're in the US, it's important to understand that OpenAI and Google's Gemini are in any case tied to the US government, and obviously they'll act within the laws of the United States of America.
They even have it written down that their systems — what they're striving for, AGI, artificial intelligence — must not be used against citizens of the United States of America. First and foremost, right? But it can be used for the purpose of defending the United States of America. Accordingly, if the United States of America is at war with someone, AI can be used against citizens of other countries. If you're in a number of other countries — well, obviously there are countries where all your data is read. China, for example, right?
There, basically — basically all systems go through a special state control system that they've been developing for a long time, for many years, decades. It recognized all that data, collected it, and analyzed it too. If in their biggest messengers, for example WeChat, you send a friend a message — this was already working five years ago — something negative from a political standpoint, you'll automatically get a notification that you've broken the law, for example, and that you're not allowed to do that, and they might come to your home and knock on your door.
Now, with AI, it's clear these systems will be rolled out more seriously in those countries. For example, here in California, putting up a camera in a public place is a whole ordeal. In China there are cameras everywhere, and if a person dumps trash in the wrong place or sets it down next to the bin, they immediately get a notification on their cell phone saying they made a mistake. In many cities — not everywhere, but in any case it exists. So obviously, a system being built in certain countries — I don't know, a system built in Iran, or a system built in Russia, or a system built in China — is completely under control.
All the data that's in there — chats can train on that data endlessly. All the laws will be made so that this data is stored, kept permanently; it'll stay in there. That's guaranteed. The next important aspect, and I often tell many people this, is that American models — well, today there are two leaders on the market, right? There's the US in models and there's China in models. Right? Everyone else is simply at the bottom, on a completely different level. What's more, the US has a phenomenal lead over China in two things. First: users.
It's important to understand that ChatGPT alone has a billion weekly users, regular weekly users. Right? And all of China combined has 600 million users. Combined! And not even weekly. Sure, their systems are also used more widely around the world, but their systems don't exceed hundreds of millions. And in the US there's Gemini, for example. It has more than 800 million monthly users. But it's important to understand that it's Google's system. And Google has a billion users in search. Right? Google still has a billion in search.
So they have a huge volume of data. Second thing. The US has tens of times more money and infrastructure to build such systems. All things considered, that's how it is for now. So there's much more money, and there'll be even more. Now Anthropic, xAI and OpenAI are going to go public, right? And xAI through Elon Musk's SpaceX, obviously. And they'll raise even more — on the stock market, I mean — they'll raise an even larger amount of money. And by the way, for those who think they should invest in them and buy their shares — it's always much more profitable to buy the shares of the ones they sign contracts with, because they'll get a lot of money, they'll spend that money and sign contracts with third-party companies.
This isn't — this isn't my financial advice, but from the standpoint of observing the market, as they say — here I'm really relaying Goldman Sachs — the AI infrastructure market will, over the next five years, grow in volume to more than $7 trillion. Right? That's incredibly strong growth, very big growth, serious. So, but here's an interesting point about countries: OpenAI and Anthropic have no interest in the system working for people in Europe, or Russia, or China the same way it does in America.
This is very important for everyone to understand. And by the way, I tell many companies and corporations too that the systems will increasingly detect where you're trying to violate and get around, for example, sanctions, or get around access restrictions to the systems, because they're not doing this because of sanctions — they're doing it because the companies themselves aren't interested in their own systems being used in foreign markets. They're used primarily — well, obviously they build the systems for their own interests, but still, these companies are so serious that they, of course, work in the interests of certain people, clans, and in the interests of a certain state — in this case, the United States of America.
Just like Chinese systems obviously work in the interests of China. Yes, they did, they do and they will, yes, and they will. What's more, in China all the best models are obviously, to a significant degree, available first and foremost to Chinese state companies, or companies tied to the state and able to reach a large number of people. And as for reaching a large number of people — in any case, all that data… they record all of that data and they control all of that data. Right?
And that's a serious matter, because in the US, courts matter. In the US, legal rules and checks matter. And in the US, big technology companies matter, right — Big Tech. Everything to do with Microsoft, everything to do with Apple, everything to do with, well, by now Anthropic, OpenAI and so on. In Europe, rights — in Europe, if we're talking about what matters, in Europe it's user rights and regulation, and it's very strict there, right? But there's no magical disappearance of data in Europe right now.
And it's important to understand, even if you look at Mistral's systems, there are a lot of nuances there. And now listen, for example, to what Pavel Durov himself is saying. Since I mentioned him — well, for the Russian-speaking audience, it's an important name. He says the following: that Europe, for example, is demanding that we share the maximum amount of data with them. Apple says the same, by the way. Elon Musk talks about this a lot — that they're demanding Elon Musk share all the conversations and all the data from xAI.
And they believe Europe does this in order to control those people, right, and to make trouble for the people who are inconvenient to them. Which I'd agree with, of course. And Europe has very strange rules regarding data privacy. On the one hand, they protect people; on the other, they're extremely bureaucratic; and thirdly, they feed certain politicians. Yes, obviously, Europe is going to suffer incredibly from this, suffer incredibly. Well, China, we've talked about it, right — it's a jurisdiction of where the data is stored, of how the language models are trained.
And, well, the role of the state, right? Russia has its own local rules, its own cloud settings, a state context. Just so you remember it and keep it in mind, right? That doesn't mean one country is automatically good and another bad. It means data lives in a legal world, not in a vacuum. And it makes sense for you, depending on where you are, to understand this and get a handle on it, right, and understand what's going on. And there's a concept called the five levels of data. Generally, when you use AI, there's the green level of data — that's all sorts of general questions you ask.
You ask about all sorts of ideas there. For example, you ask: "How do I cure the disease on this plant?", right? Or: "What grains can I eat besides buckwheat and amaranth?", right? There's the yellow level of data — that's personal thoughts, plans, ordinary drafts without other people's data, right? They say this is supposedly not a super important thing, but I believe it's the foundation for real AGI, right, for real artificial intelligence. There's the orange level of data — that's work, contracts, code, client data and so on. Interestingly, by the way, this is ranked as supposedly more important than a person's thoughts.
Notice that the world ranks this higher than a person's thoughts, which is very strange, right? Because, in theory, it should be less important than a person's thoughts and plans and ordinary drafts. But we live in a social world that says the person, as it were, comes somewhere further down the line. Just like a huge number of employers think that you, the employee, are a slave. Whoever feels like one, you can write it in the comments. It's always an interesting topic, how you feel, right? Because it's not a question of the company, it's a question of how the person feels.
There's the red level of data — that's everything to do with finances, medicine, taxes, I don't know, passports, legal conflicts — that is, everything related to some kind of emergency situations. Although, again, medical data, for example — why is it considered more important than a person's thoughts, personal thoughts and reasoning? That's a big question, because, for example, I have a lot of discussions on the topic of, well, my own personal development. And I have that development, and I have thousands of hours of my thoughts from the last ten-plus years — they're recorded, right.
And if you analyze them, I think they're more important than finding out what my blood work shows, or what my cholesterol level is, or how some ultrasound of mine went. Well, it seems that's much more important. And for some reason, AI itself and the world consider medical data more important than those thoughts. By the way, if you think so too, tell me, okay? Finances I can understand, right — it's money, personal, personal relationships, a person's personal money. Although, again, what finances, right?
How much money you have in your account? How important is that? More important than thoughts? Big question. Because there's the fourth — the fifth level, it's called the black level, right. That's passwords, your passphrases, private keys, other people's personal data without permission. That data — that's already brutal stuff, of course, and you have to be very careful about handing it to a regular chat. And by the way, they write that this data shouldn't be handed over at all.
But since I have a huge number of chats and I use the system very heavily in my life, I don't understand how you sometimes avoid handing it over, because a large amount of data goes out automatically anyway. How does it go out? Well, let's break down one of the levels. Imagine you need to do your tax return. There are countries where you have the right to optimize your tax return, optimize your taxes. There are countries where you can't do that, where it's immediately illegal.
And you give ChatGPT, for example — you want to hand over all your data and say: "Optimize this data for me as much as possible, show me where and how I can save, for instance, or how to do the report correctly." And for it to do the report, you have to give it all the data. If you don't hand over all the data, don't hand over your past returns, it won't do everything precisely. And if it doesn't do everything precisely, then you have to manage the process. And if you're managing the process, then you're not using artificial intelligence — you're using pieces of it.
And getting the job done properly, in terms of filing the return, for example, will be very hard. Me personally, for example, I upload my return without a second thought. I'm not urging anyone to do this, but I calmly upload my tax returns to ChatGPT, for example, and I think: "Well, what could they see in there?" Although there are all kinds of numbers, my details, there's no super-personal data in there, right. For example, in America the Social Security number is very important, for example.
But the Social Security number, of course, isn't in there, right. That's also an important thing to watch for, of course. To look, keep track, know it, understand it, and be able to navigate it. An interesting thing that happened recently. ChatGPT released a finance integration. They said: we've, like, released a separate module, GPT Finance. And this GPT Finance somehow reminds me — and I've seen so many of these projects, in startups, made by all sorts of people — of those personal wallets where a user can keep track of their own money.
I don't fully understand why I need a separate GPT Finance for this, if I can just upload my accounts to ChatGPT once a month. And people usually have very few accounts. I mean, people usually don't have eighty accounts, and if they're doing it for a company, then a company keeps a different kind of books. But they did this to get access to your finances. So what does that give them? I think it gives them access to information that's far less important than access to a person's personal thoughts.
And thoughts — that's where, in my firm belief, once again, that's where they'll be able to influence you far more, and on a really large scale. Of course, there's no need to fear artificial intelligence. And on the ToTheMoon channel I strongly urge everyone to use AI, so you know how to use it and understand what it is. Because we're entering a new world, and we talk about that a lot in our special episodes. A new world where 90 percent of the apps you use now will disappear, where things change — physical devices will change, right.
And you need to understand one thing: when you give it your data, you're not just giving it information, right — you're giving it the ability to understand you, predict your behavior and potentially influence your decisions. And here's an interesting point: do you personally see this as a plus or as a minus? I see both in it. And let's sit with that and think it over. Until next time.