We're on ToTheMoon. And we have another issue of technology sites worldwide from the Silicon Valley.
I mean, the most interesting thing is that even on demonstrations like that, it's supposed to be all over it, everything's out, still, uh, hallucinations.
I ended up with 10 books not selling any. I feel like I've been on the authors' websites, that they just don't exist. You know, these books?
There'll be glucs and there'll be, uh, mistakes, and there'll be hallucinations.
They gave it to me for the first time, they gave it to me for the first time. They didn't. That's interesting, too.
Today, I think that is a very interesting subject that is about finding, uh, and developing a new search. And the fact that the OpenAI wants to release GPT Search, or more specifically, it has already released, has opened it to a certain group of people. It's interesting, interesting story, because half a week ago I was in touch with, uh, uh, literally, half a week ago, I think we were in touch with partners, there, with one. And my partner says I'm not using Google as a search anymore, but a lot of ChatGPT. And we talked about it, because I also use ChatGPT very much as a search that he's got a very big turn on the 4o exit, and he's got a lot of power to look for and give the right parameters. There are different problems, some glue, stories, but the search is working. So, we talked about you going to Google to, like, see some pictures or something else. And, in fact, we are now experiencing this redistribution of the use of time. And this one, with this subject, I want to discuss, because it seemed like ChatGPT, he'd, uh, be integrated with other applications. Remember when the first plagins came up, you were basically like ChatGPT, opening up some kind of plagin, I don't know, Booking, and you wanted to find something. Then everyone realized that this topic was totally unlived, unworking. A man goes to Booking anyway if he wants to use Booking. And, in fact, what's going on in OpenAI now, they're on the basis of their different decisions. Well, I am deeply convinced once again that there will be complete solutions, completely different, up to my own, there, disk managers, weather forecasts and everything. I mean my own annexes. And this new search they want, uh, to start, you know, all people. The question is when he's gonna be on all people? It's clear that there are competitors who do, you know, there are key competitors, like Google search, and where AI has already been introduced, or Bing, where AI is being implemented, still OpenAI, right? Operators who compete like Perplexity, like You.com. And Brave Search is another company, yes, Brave Search, probably from a large man who competes in search.
But I think with my two hundred plus million OpenAI users, it's going to be a real move.
That's very interesting. I wonder what's going to happen, yes, to the companies you've listed. And we've got Genspark on the bench again. A start-up that, uh, picked up 60 million in an evaluation of sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-sixty-s ago. In fact, the same thing Perplexity is in the view of this one of the-- some new developments that are. And then the OpenAI comes and says we'll do the search. And like those companies, they're not really interesting anymore, but when that giant comes in. But the most embarrassing thing in this story is that the conceptual issue of hallucinations has never been resolved. Uh, I looked at this story, but then I went, I checked. If you open a demonstration that is right on the news website, on the OpenAI website, it's just a man who's been asking to find summer festivals in August, well, there, in some U.S. area. And the first one that comes out is made with a mistake, because it's Appalachian Summer Fest. And if we go on the site, it starts on June 29th, ends on July 20. And in the extradition, uh, that goes to the GPT chat, it says it's the end of July and 16 August, unless I'm wrong. I mean, the most interesting thing is that even on demonstrations like that, it's supposed to be checked out, everything's out, and there's still hallucinations. It's like, remember, Bart's first demonstration was at Google's, where actual mistakes were in the extradition. It's a story. And the most embarrassing thing is that conceptual history--
Wait, but it's a search. Look, it's a search. So if we take a search, if we lead a Google or a habitual browser, we usually get a totally unrelevant answer. Now, at the moment, the answers are very strange. And basically it's the same thing, the same search they're giving, and then they-- they even talk about what they're doing right in the description that Search GPT is for search purposes only. and not used for training, in basic models, there and so on. It's clear he's supposed to be more cool.
But it's not hallucination, I think. No, you think it's hallucination? I don't think it's hallucination.
You can googling...
You can. No, what they are, what they represent, you can, yeah, they just said you could open a separate dialogue window, it's their new introduction, open separate sites and go on. Look at these sites. It's not enough they're still planning these sites there to be some visualization, but there's different parts to add, and so on. You're right from the point of view, it's probably that the quality, uh, of this is gonna be very specific from the beginning, yes, the quality of this search there and everything. But I don't think it's a problem, what they're giving. Well, basically, you're talking about the next thing you're doing, you know, looking for a super smart and super cool guy. Oh, well, I'm, like, getting out of what's now, looking for Google, which works, like, a fucking crust, and to find something, we need to open up a lot of references. And it's not that it's the right options. And I feel very calm about, like, that, uh, while I'm sometimes annoyed with hallucinations in, uh, different models, it's really annoying when you're doing something special, and he just keeps dumb. But it's important to understand that this tool just didn't exist.
Well, for me, the problem is you don't know what you can trust and what you can't trust. Yeah, I mean, if you're looking for Google, you're gonna blow up on the websites and you're gonna get some information. And here you have a gift. I think she's a good one, but I don't know. I mean, this paragraph makes me embarrassed.
I'm on the same page because I've been looking for one book. I wanted to know if there were any books on this subject, and I had an experiment. Asked Google, ChatGPT. ChatGPT gave me a list of ten books. I'm saying, where can you read them, buy them, something else. I ended up with 10 books not selling any. It feels like I even went to the authors' websites, that they just don't exist. You know, these books? And in Google, yeah, in Google, I was looking for, uh, he didn't find those books. So he got some words on the same side, gave me. But these books, which are right with the author and with the name of me, have been issued by ChatGPT, they don't exist.
So it's a good step forward anyway, right? Anyway, it's a market movement, uh, probably happening. But conceptually, the problem with hallucinations is not solved. And if, for example, you're in touch with GPT chat when you're just there, you know, some kind of specification or something, you can handle it, keep your head in your head, that there might be some inaccuracy. And when we're talking about some factual search, it's like everybody would want it to be--
You'll have a reference, you'll have a reference, you can open it peacefully. And in fact, even in the GPT chat when you're looking, it's always a question of prom and descriptions you're doing. If you're looking for it, you'll ask him to give him a link. I, for example, often ask for references. You're moving to a link or you're telling me to check the information, it's true or not. From the point of view of the data, the particular reference is made. It's your question, your lines, your details there. It's clear that most people like you won't be able to look. But once again, there will be glucs and there will be mistakes, and there will be hallucinations.
But I just see that we were just having a Wednesday conversation with the bud-- well, with the guys, there, with partners, how much progress has changed in a year, just a year. I mean, as you can see, the other quality of the system's work. When you ask for information, you're graphic, you're graphic, you're gonna get files, pictures, you're gonna analyze the information on different articles. I've probably stopped 70 percent, uh, online reading. And there was an interesting comment on Techranche late last night.
Last night, TechCrunch had a search article, and there was a comment that, in fact, what was going on, if a man stopped coming in. And that's what you're saying right now, the quality of this, well, here in-- okay, okay, that's a separate, separate problem, and it's gonna be. But what's going on? I see myself, I'm not even stopping on the site. I mean, when I'm conditionally given you extras in Google, you see a little short text. If this text is from the SEO perspective, well, the optimization of the search is bad, you don't know what this page is, you open it, you start looking at it. You're being searched now, and you're just saying, "Do me the summit, this and this block, check the extra information." They-- it's gonna work hard. You stop coming on the site. If you stop going on these sites in the field, the concept itself changes, getting information, sharing time and advertising. Pay attention, will you? And there was a discussion about if people stop their places, you're different from being there at their conversion, and there, some of the offers they give you, you're just not gonna be that. Do, right? You're a little different, different, different, different. Again, I'm probably talking to my bell. I'm very busy using even current search tools. I like this news very much because I don't like it now in the search for trash that's in it. And I like more, for example, that I'm at the beginning of ageneral AI, giving out information, and then I'm second-hand searching, rather than me at the beginning, as much as I'm looking for, this is a huge amount of extradition, extradition. Let's see what they're gonna let out, yeah, what they're gonna do.
And when.
Yeah, and when they do. And they have a whole series of different things that should come out. But progress, I think, is very huge. And Bing now has an update this week to look inside his own. In models, by the way, interesting, you said hallucination. Google Gemini in the updates has a separate unit to describe what they are like fighting hallucinations, launching, writing separate, separate technological designs there to check. Because, of course, when you ask to read the book or watch the movies, and you-- and you ask to give me the IMDb ratings, and you get a fictional movie--
Very strange feeling.
...which does not exist.
Yeah, that's very strange--
Yeah, it's a very strange feeling that's a little bit.
And that's an interesting design, by the way, that many other sites block, block the search for networks. For example, on TechCrunch, you can't give the news to TechCrunch, he'll say I can't go in there and I can't analyze it. And that's very, I think it's hard to make a decision on your website for AI. And I think I'd open most of the sites, but what's the problem? What if you have TechCrunch analysis, why do you need TechCrunch? You'll sit inside and analyze.
Well, that's when you hit a little futurism now, if you're missing a commercial, there, with a search for a place to go, some kind of revenue share with neuronets, that if you open your website, then actually, They share with you some part of their proceeds, uh, that's the right to use your information.
I mean, no, but it's working, it's also there, and the other way around, the story. I have a friend who has a big CMS system. So it's a system that allows websites to develop, and it feeds the models to everyone, and it's written as special as possible, there's some boxes, codes, text to get the models to know about it, all of it. Technical documentation. Because he says, "I am deeply convinced that if I'm inside, at least I can be offered. If someone asks me how to do something on my CMS system, she'll at least write the code. And if I'm not there, she's just not gonna offer me." I mean, on the one hand, we think we should pay the kind of money to that, TechCrunch for giving the information. On the other hand, many systems should pay the money back to OpenAI to give them away. I think that because if you're rightly built and you're being referred to, you're just gonna stop being invoked, you're gonna stop coming.
Well, if you have at some point in time you say, "Well, listen, I'll give you a turf, I'll be, uh, I'll be watching others." By the way, X AI, X, well, Twitter, ex-girlfriend, has launched automatic training on all absolute posts inside. And yesterday, too, the article came out. Yesterday, the story was that how to set aside this-- automatic training in all places. And autopsy-- and post training, by the way, is a very interesting thing. Now, the US policy is taking a huge amount of perturbations, and there's one of the things that erased very much information on the Internet and that's how many websites removed it. And everyone says, "All right, we have X, so all the information is kept there." Well, first of all, how much it's been kept there, the question, but there's a lot of information there.
I mean, if this is social networking training, it's a very interesting mechanism, too.
I'll give you a minute, a minute of pessimist. I don't want to upset you, but I don't know where this is going, do I? So, as a search-- so let's just, like, just... I mean, I mean, I mean, I mean, I mean, business is gonna be big when all the agents are, yeah, all the parties are more and less happy, right? I mean, for example, the main health problem of the US, which is like a patient, there's a hospital and there's another insurance company. It's like a tidy-sized triangle of love, which, well, more or less people who can decide, they make billions of dollars, yeah, build new companies, make billions of dollars. And people who even have bright ideas, but they don't understand how this triangle is built, they can't solve this problem. Now that we're talking about searching, there's actually a love triangle, too. I mean, it's working out, those who want to move. Right, searching for the strongest intent. Then eat like, then eat, it turns out a company that makes a search, that's a search engine, and there's a user. And we need all of them to be happy, right? I mean, you can't just let the user be happy because then long term won't work. I understand Google has a basic model that they've come up with an auction, which has allowed them to form this triangle in two- to two-and-two years. And that's why it's such a successful multitrillion company. And I don't understand-- and that is, it's like we're doing-- with generative AI, we're in a situation where, like, when users like, the user experience is getting better, but it's a user experience. experience is better with generative AI in search of such more research, a, information. But I didn't hear anyone doing a decision to buy something by using Perplexity. I've never heard of anyone. So, as in some sort of vertical, now, the generative AI search is gaining from a routine search, but not all. And I haven't seen it, surprisingly, that I haven't seen any analysts from any high-ranking publication, which is how generative AI search wins and how it's not. So, it's, uh, it's, but it's actually getting better for the user as a whole. Let's put it on, let's take it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it, let's get it. Next, when we're talking about the search system, well, any LLM call, it's pretty expensive. And I'm kind of, and it's amazing that I don't, so the young economy here is breaking into search systems. But more than I wonder what I'm very surprised about, that's what Ilnar said, that's what I, uh, I don't have to do, like I think, all these generative AI systems don't do. enough to, uh, let me know how reliable the results are. I mean, I don't think I have any product solutions that build some trust between the user and the system. I mean, the user might be able to accept that yes, not always the exact results, but I would like to see it as some kind of... understand that within each of these systems, it has a percentage of how much it is. She thinks she's sure about the result. And so far, it has not been integrated into any product. I mean, this information is inside, but it doesn't seem to be a litter. And the third, we all see that Condé Nast and, uh, Forbes, they're suing Perplexity, and they're judging them, and there's enough, and, uh, the ultra-vital law is coming. It was therefore recently also news that Anthropic violated the rules and rabbits of content that was prohibited, which, as the right holders, banned the rabbit for AI systems.
And the other day, the news was that Runway ML, I'm sorry, we went a little into the video, but...
Sorry, we went a little in the video, but Runway ML is, in general, a company worth four billion and that generates videos. Turns out they took the operas, took the co-piranated films, and they were trained on the dateset. So it wasn't accidentally there, and they took it out on purpose. And it turns out that the system, she knows as much as what--- that's, that's, if she's told to generate like a video of a particular opuber style, she'll do it. Now they're sort of fixing it so the system doesn't know. But when this scandal happened, the system knew directly, that is, that is, that I think it's obvious that all these systems, they somehow evaporated the content on the Internet. free. But no one has come close to solving the problem with right holders, except OpenAI, maybe. And the OpenAI, too, there's an article that they have no-- that they have no way that they're low-margin business, that they're-- they're losing five billion a year. Yeah, well, if they make three and a half, they lose, there, five or even more, that's every year. So, the is not yet visible as to how the system is being handled-- as a decision--- what is the final solution of the system with right holders. I mean, we might be, if there's no incredible technological break with LLM in the coming months, but it's probably not gonna be, because even if it's, it's still more powerful. Models, they'll just be much more expensive, companies will have to figure out how they are, and to solve the situation, how they are, and, to optimize their own youth economy and how they will solve problems with right holders.
I mean, I think that, like the quality of small, fast models, we're already on the platoon and we need to take a temporary fee, and that's the question of how to get the young manifold down and how to do it. The right holders were happy.
Look, if you take a little bit of a little bit of a story, it's a little bit of a story. Because if we remember Amazon, Amazon is the company of the nineties where it was, in the ninety-fifth or in which year Amazon came. And yet it was a little bit of a loss until just recently. Yeah, she's been changing. So she's been in for years, over 30 years, probably a lost company or, over 20 years, she's been lost. She's been a long time, she's been a long time late. I mean, it's a few billion dollars worth of dolls at a serious operator, it's probably not some super-dum. Well, I think so. Once again, from the perspective of the young manifold, uh, there's more to be focused, um, really, on the solution... That's the lot of these decisions we've been discussing, there, hallucinations to the law-makers. Which is interesting, by the way, when you put it in ChatGPT, you can put in a book.
I'm very, very much writing my instructions, I'm finishing them up, finishing them up, and so on. And when I write in the instructions that every request I give is rechecked on the Internet, please, separately and so on, ChatGPT doesn't do it. And I think there's a strong restriction on it. They could actually be very easy to put a "Trust this information" on the top of the button right now. All users would have a button, yeah, and you kind of click a button, and she's just, like, her, you know, at least a man. Because I understand that a lot of people don't understand how to check, know how to start, make a request, just don't understand. Well, it doesn't understand how to communicate.
A huge number of people think it's a hundred percent truth.
What kind of problems could be hallucinations.
You know, these books.
Of course!
Not enough.
Well, a lot of people don't know at all. Yeah.
It can't be that way.
A huge number of people don't even know that there's a ChatGPT. Yeah, yeah. And we're basically, first of all, talking about a small number of people, you know, who use the system now and all the time. I think that of the 200 million people there, ChatGPT, how many of them are really active users, right?
Or how many Google users really knew there was something on top of it. Or Zuckerberg has now released the updating of Llama 3.1. We probably won't be able to talk about him today, but still. And he says, "This is integrated, there, in WhatsApp, and then on Facebook, into all our tools." And then, uh, I flew out of Canada yesterday with a-- a Polish-- with my wife, and Pauline says, "I didn't know whatsApp was looking for." And she never found him. I don't know why I have a look at whatsApp, and she doesn't have a vision, does she? I'm in WhatsApp, there's a massager available, I can use IA on the road when you're flying. She didn't find it or know-- she didn't even know it was there. Yeah, so people are still, I think it's gonna take a while to get used to it.
There's a big subject I wanted to say. Two, two topics. The first topic is that ChatGPT's retention seems to be better than most of the Consymer products I know. I mean, if we don't take social media, it looks like, uh, ChatGPT's retention is better than 9 percent in the 9th, right there.
I'll explain to the users. Retention is a user return. Retention is a reuse. Use, use, use, continuation, sale when the human being continues to use the system.
Like loyalty, huh?
More and more, and more.
Yeah. And, yes, there's this, in a rude way, looking at, like, the 12th month, like, one year plus one day, how many users still pay. If that figure is more than 30-five percent Too, it's generally very good. It's like a very good business. This is generally for enterprise. It's a very small indicator. There's an expectation that I don't know, eighty-nine percent will be paid. But consumers are more like this.
Mm-hmm.
Mobile, yes. The consumers are getting less and less globally. And that's why, uh, but 35 percent is a very cool figure. They have, like, a--
Well, I'm also gonna notice that it's too early to study cohorts. I think they, for example, all users can take, which was a year ago. If you take the users who came six months ago, they're probably even more active than those who were, for example, or
who were returning.
Yeah, I agree. Yeah. I mean, it's not even completely evaluated. That's the first. That's-- and that's the second. I think that there's a little bit of a Cookerberg, that they're having such a big PR change in Meta right now, and a little, um, statements from Zuckerberg, say they might not be very broad. They're being replicated, but he said he wanted me to, he said he wanted him to have the most popular AI consumer in the world. It's Eaah. Well, he says he wants Meta AI to use more people than, uh, ChatGPT. I think it's not about coinization, but it's about the number of users. And I think his job is to make more than 100 million, at least monthly active users before the end of this year. Meta AI, but I think he even targets 100 million daily active users. And, uh, it's the biggest one, it's really nice. So this is a new model that came out of Meta by four hundred and five billion parameters, this model is better than GPT-4o, I understand, in most of the baccalaureate. The majority of the experts agree that Zuckerberg has no problem letting everyone use it free, but just to sow Sam Altman and try to take his growth. That's the real thing--
Well, no one will consider the revenue from this model separately. Everyone will consider it an intruder in his softs. Right?
Yeah. He's, uh, he's got a really good story because he can show some food skies in six months, probably and make test, and, uh, test-like, commercial integration. Because all the world's biggest advertisers, they're sitting, they're on Facebook, and he's the best sympathetic to figure out how to put it in. And I'm probably probably not gonna get that in even a stock price because six months later, his products will be brought back by a fascinating strategy for the way the advertisers use the system. And, uh, what I think it's important to add is that for OpenAI it would be a significant blow, because most of their revenue comes from, uh, Concuemer signatures. It's very, very dangerous.
I think...
Sas, listen, but still, it's so different tools from the point of view. So if Meta learns to do it, then of course they should be number one at the cost, capitalization, all over it. If they learn to move, uh, to their own, for example, the massagers of a man or there on social media, in Instagram, in Facebook, anywhere, they'll be missing, they'll be making inquiries through their system. Just so far, it's clear that people don't make a lot of super-requests through their system. And the OpenAI, uh, created a thing that, well, as it was a new entry point, it was a new entry point, it was a reallocation of the user time at all. But it's gonna be very interesting in the part, uh. That's interesting, by the way, about Zuckerberg, about his PR strategy. I saw a couple days ago, there, anon that he'd be doing with the NVIDIA head, and it made me a little bit of arrogant because I think they didn't. Well, I mean, he's really, they're gonna have another story right now. Maybe-- for end-users, it's good. I don't know how good this is for business.
I don't know how good it is for business, for startups.
Who cares?
But it's, it's very interesting because they, they don't, they don't put themselves like... I mean, Snapchat, I'm very much pro-resistant with TikTok. Evan Spigel, CEO Snapchat, always says, " Here, we would certainly support if TikTok were banished ". Zuckerberg in his last interview just a few days ago said it was like a higher grade than his pay-pay, that is, well, that's a little higher than him, and he doesn't want to comment. But he clearly took the lead in getting the generative AI, taking the ChatGPT. I think we all should understand that it is likely that it will be possible because, well, there are a lot of emerging markets: India, Indonesia, Viet Nam, Thailand, Brazil and so on, where free generative AI solutions thrive. And Ia's access to these users is already through whatsApp Wu Zuckerberg. So 100 million ma-- so if someone-- so I don't have any doubt that 100 million MAU is a target for Meta AI, he'll do it by the end of this year.
In the world, he-- he can certainly do that, given his few-- how many systems there? Millions of users, huh? And of course, he has a clear opportunity to do it around the world. I don't think the U.S. is likely to be. He-- they say, uh-- they gave me the statistic this week, by the way, for the first time Zuckerberg said that there were many users in the USApp? 100 million users, if I'm not mistaken. No one remembers you, I'm not. They first gave it to them, they gave it to them for the first time. They didn't. It's also interesting that when someone first says something, that is, that's, uh, PR strategy, some sort of thing they
do.
And you have to understand, yes, that Meta, for example, has Threads product that I think, well, I don't know any person who knows who's using.
They're a rock-- and they've announced that they've had a year since the launch. They say they have 150 million monthly active users, 175 or 100, 100, 150. But it's a huge number. And it's growing. Here, uh, I'm just saying, I'm gonna want to convince you that if anything, the fact-act of the cheque, because it's important that it's growing. And I'm saying that, compared to the first quarter, and I think it's like a new nipple as much as the generative AI now. And, of course, 100 million monthly active users if he delivers Meta AI-
Zuckerberg really said, Sas, he really said two weeks ago that we had a lot of fucking followers in Threads. And I'm just saying there's probably some countries, we don't see the segmentation, which countries are using, yes, and who's sitting there and why.
Well, I've got a note of 130 million in February, 150 in April, and now more than 175 million active lasers.
Yeah, yeah, yeah.
Thank you.
They didn't fall. Well, that's interesting, interesting parameters, yes, very interesting parameters. I, in fact, when I saw 100 million active users in WhatsApp, someone says that in the US WhatsApp is not-- it's 100 million in the US, 100 million in the US. When I saw someone saying that the U.S. doesn't use whatsApp, you're like, "Wait, 100 million. Really? "
Somehow, they really like text messages. I don't know why in Russia, on the contrary, no text reads.
SMS, yes.
Yeah.
Another, yes, another market, yes, another market. But 100 million is still a lot. So, this is huge parameters, huge p-- from different sides, of course, everyone comes from different sides. Google on its part, Meta on his side, OpenAI, of course, does something different, but they still have a friend of Microsoft. I don't want Microsoft to swallow them, and then the company will die from a development perspective, I think. Well, maybe I'm a bad bet-- maybe I'm, uh...
Cut it out.
Well, tonight, we'll leave a good one.
Today was a good one, uh, graduation. Uh, by the way, today didn't come up with us, so...
Ilon.
...the self-imposed, yes, man, I think Elon Musk. Although, by the way, this week of interest, this week is really, uh, showing a huge ring-- it's summer, guys, late July, like, a quiet time, but we see a very strong, direct incitation of all. leaders, major leaders in AI. All the releases are giving some updates, saying that there's gonna be a bunch of anons, so there's a lot going on. Interesting, by the way, the subject is still in AI, Tesla, in the cars, I think they're gonna be doing the autopsy thing very seriously right now. I mean, they're the first ones. That's who's trying to use and use. I wouldn't let go, and, uh, hi-i, they're there, uh, they're gonna be getting money there, too. Here. Tesla, by the way, now, I think yesterday or the day before yesterday they're planning to move five billion more into AI. Such, of course, serious, serious parameters. But we did, a week ago, who didn't see, we had a, uh, a roll of 28 companies that gave more than $100 million. It's worth looking at him, studying to understand, and where the money is at AI. Guys, thanks a lot. Uh, to the Moon. The technological news on the world from the Silicon Valley. Although I'm alone in the Silicon Valley now because everyone's rolling around, so the world is. Uh, we're gonna reunite together again soon and we're gonna have to write a live release at some point. All-- you have a good evening, and I'm having a morning, you know, a good time for our audience, uh-uh. Bye, everyone!