AI Fire Hose #4: Crash, Boom, and Chatbots

0:00Are we in the most hated bull market in history? The S &P is up 17 % so far this year, and the NASDAQ 100 is up 43%. The highest number of people since 2008 are invested in the stock market, yet institutions haven't participated. And sentiment is not happy. What's going on? Well, at Real Vision, we'll be talking to the world's best investors and thinkers to answer that question in crash or boom, how to profit from what's coming. This is a really important topic, and this two-week special series starts on September the 11th with what I think is coming. I'll lay it all out for you, and then we'll hear from the others.

0:43Go to realvision.com forward slash big question to get all the details. That's realvision.com forward slash big question, all lowercase, to get all the details. Don't miss out.

1:07What's going on, guys? Welcome to Real Vision's weekly AI firehose, where we drink from the stream of everything happening this week in the exploding field of AI. I'm Ash Bennington, joined by our resident AI expert, AI developer, Mikhail Voloshin. Mikhail, always a pleasure to be with you. We're also joined here today, Ash, with my co-star Phoenix over here and with an experimental co-star that's running in the background, I wrote a little fact-checking agent whom I call Grace for the time being. She will mostly be perfectly quiet. You might see some activity on the screen behind me. But if we happen to say something that needs fact-checking, corroboration, or correction, we might hear an automated voice pipe up saying, excuse me, and we'll see whether or not Grace has anything to tell us.

1:56So does Grace stand for something? No. Grace is merely an agent that is tasked with listening to our audio stream and extracting any facts or statements of truth from it, Googling those statements of truth, and then coming back to us with statements about whether or not what we said was true or false. she's named after Grace Hopper one of the greatest pioneers in the world of computing and she's also named Grace because she tries to make sure that we conduct this podcast with a bit of grace but for the time being she's going to be perfectly quiet and if we don't hear from her then that's a good thing that means that we haven't said anything egregiously false but for the time being We're just going to watch the text scroll behind me.

2:53And the folks that are just listening on the audio will hopefully never know that anything's amiss. Well, we always try and conduct this podcast with elegance and grace. And we'll see if we come up short, grace can call us out. One thing grace will not be able to call me out on the fact check, I should say, is this coming week, this was released on a Saturday. We have a special content week continuing at Real Vision, Crash or Boom, How to Profit from What's Coming. Today actually was my first day hosting the crash or boom, and I got to host Benjamin Cowan and Rekt Capital talking about things that are happening on the digital asset side in terms of this framework of whether or not the US and indeed the globe is heading into recession.

3:34Earlier today, we had Andreas Steno-Larsen on with Michael Survey, head of equity strategy at Nord, talking about, again, this same theme. Some of Michael's models are screaming recession, rising risk in Europe, and the risk that the U.S. will follow Europe into a potential recession. Mikhail, we're talking about boom bust today. We're talking about a lot of interesting concepts about chat agents. I'm wondering if this idea of boom and bust, it seems like we've got some skeptical stories today. Is there a sort of a boom crash theme that sometimes happens in the AI space? Because that's human nature to see things as sort of, I guess, rising or falling.

4:12It's a particularly poignant question to ask about in the context of artificial intelligence, because there was a massive, massive bust in the world of AI that lasted for decades. There was something called the AI winter that we're only now beginning to crawl our way out of. The neural networks were first invented in the late 1940s, and they were first built in hardware in the late 1950s and early 60s. And for a little while, it was believed that neural networks would be the end-all and be-all of artificial intelligence. We would build artificial brains, and that would be the, you know, the sentient robot future.

4:51Unfortunately, it was discovered fairly early in the 1970s, I believe. um a uh um there was a book uh about the limitations of perceptron technology that completely devastated the field of uh neural networks at the time and in fact sent uh sent such waves of despair through the possibility of ai in general that the field really didn't recover for decades. So the very fact that we're coming out of the AI winter right now is really something that speaks for itself in terms of, is there bust in the world of AI? Now, we're not headed for another AI winter anytime soon. We just got out of it and spring is thawing and it is beautiful.

5:45What we can talk about in terms of the impact of AI in the context of an otherwise receding economy is a very interesting position that we're dealing with now, which is that there are certain fields, there are certain industries that do better when the entire economy is retracting. And those have historically included entertainment, video games, and any kind of labor-saving or cost-cutting technologies or capabilities. And AI is one gigantic cost-cutting, labor-saving capability. Yeah, it's such an interesting point. These are sometimes called counter-cyclical industries. By the way, a great counter-cyclical industry, financial news.

6:30When things get ugly, people need to know what is happening. But you bring up this great point, which is whether labor-saving technology in the event of a recession might actually see more capital flowing to it because it's seen as something that can potentially enhance the bottom lines of corporations. So in a healthy economy, the question is whether to employ a human in a position or whether to employ an AI in that position. And when you're faced with that decision, you're usually going to employ the human if you have the capital for it. And then you might do a little bit of value add with some AI capabilities in order to pick up the human slack.

7:14We talked a lot about that in the last episode. In a receding economy or in a shrinking economy, that is no longer the choice. The choice becomes whether to do a job with an AI or to not do the job at all because you can't hire a human because you can't afford it. So in that context, the AI becomes the better option. Yeah, you could see this in an environment where, for example, you were thinking about cutting a service entirely or just bringing in an AI to do that service at a very low rate. Is Grace speaking? Is Grace perking up back there? Grace has decided to say something. Let's see what Grace has to say for him or it or themselves.

8:16Thank you, Grace. All right, that's seriously creepy, dude. Basically, real time, it's fact-checking you about your statements about the duration and depth of the AI winter? She's not quite real-time yet. It's ridiculously expensive to operate her in a real-time, fast, responsive manner. So because this is just a prototype that we're debuting this week, I decided to hedge my losses and keep her in a slightly slower mode. But if she ends up being useful for this podcast or for others, then maybe I'll speed her up and add some capabilities, maybe give her a better voice, maybe give her an avatar.

8:58And we're going to see the impact of decisions like that in some of the stuff we talk about today. You know, by the way, for people who've been watching tech for a long period of time, as both you and I have, one of these things that you hear, anything that's prohibitively expensive to do today, in six months, two years, 10 years, may no longer be the case. That's just the nature of Moore's Law. It's not going to take me six to 10 years to get Grace to be cheap enough to run on the fly. Um, I'm hoping that, uh, if she performs well today and possibly in the next couple of podcasts, then it might be worth it to sort of crank up her capabilities a little bit.

9:35All right, Mikhail, from Grace to Neuro-sama, tell us about this character. Oh man. So what we're going to be talking about this week is intelligent agents, chatbots, and other sort of human substitutes that people have used throughout various industries powered by these large language model capabilities. And this really runs the gamut. Now, I want to start out by something that has been requested that I discuss, which is a character named Neurosama. I don't even know where to begin with this one, Ash. In order to explain Neurosama, I have to start by touching on this really bizarre media culture that's prominent in Japan.

10:21So I'll start by talking about one of the places where these sort of digital avatars started. There's this animated character named Hatsune Miku. I'm so gimping that name. But the so Hatsune Miku is a mascot for some voice synthesis software. She debuted around 2007. And basically, this is a company called. Let me check Vocaloid. Sorry, Krypton Future Media that released a library of software that allows for musicians to basically have a singer. Oh, dear. um uh grace just wants to corroborate my statement about ai being a cost-cutting measure i will skip it maybe grace can take a nap you know what i'm gonna i'm gonna put grace to bed for the time being um so what i wanted to say about and we can we can do the uh we can do the good night gracie bit we i thought about adding that um but i can't quite get the uh squeaky voice with the joisy accent going on just yet maybe in future editions and in fact that's uh that's kind of exactly what hatsuni miku is it's a um it's a lot it's a voice library that allows for musicians who don't have a vocalist to basically write lyrics and notes and have a singer for their for their music so and so how does that turret tie in with neuro sama or neuro sama neuro sama is a is a version of a is an AI that's basically that a manifestation of one of these early, uh, sort of digital characters cranked up to 11.

12:10Uh, she's what's called a VTuber, uh, which is a podcaster who cat, who performs not through their own face, but through an, but through a digital persona. The difference with, uh, with, uh, Neurosama is that there's no actual human operating the, like the AI, the anime avatar. It's actually just AI's all the way down. She, um, so it's basically, it's AI generating the, the actual content as well as generating the persona, the skin on the front. Exactly. She was created by a programmer named or called a Vidal 987 and she was she debuted in december of 2022 she now has over 165 000 subscribers she was originally programmed to be a twitch streamer which i don't know if uh i don't know how much common knowledge it is but there's this community called twitch where people actually watch other people play video games um so talk about like spectator sports only crank the nerdiness up to 11, right?

13:19So Vidal programmed an AI to play a couple of games, in particular, a rhythm game called Osu, which is related in some way to Dance Dance Revolution or Guitar Hero. And this AI is also capable of playing Minecraft. And so he created this character to describe and talk about the Minecraft and Osu games as they're being played by this AI system. So it's actually a collection of AI algorithms, some to play the games, some to talk about what's going on, some to interact with the community in the comment section under this streaming presentation. So Mika, let me ask you this. I think for a lot of people, people who are in that community probably get it immediately.

14:10And there's something that they're maybe passionate about. But for those of us who aren't, and they're listening to this, and they're hearing, and they go, Okay, so let me get this straight. So it's online video games being played by computers, being essentially sort of narrated by a computer by an AI, what's the lure? What's the connection? How does this pull people in? That's a really good question. And so a lot of people are pulled in by the novelty of the fact that there's a computer that's doing this, right? There's also a lot of people that are drawn in by just watching her glitch and being really amused by things she says that are horribly wrong.

14:48And in fact, there was a huge controversy that surrounded this exact glitching property because look, she comments, she interacts with the comments section, right? People will type something and then she'll comment on the thing that they typed in the Discord or the Twitch or whatever. So she was actually banned from Twitch for a couple of weeks in, I want to say, February of earlier this year. Because basically the Twitch streamers who were watching this were trying to trick the AI into saying some really offensive, racist, and let's just say politically incorrect things. And basically, once they succeeded, this was treated as content by Twitch and Twitch pulled a big no-no on that.

15:39We're going to take a quick break and be right back with more of the day's top analysis on the Real Vision daily briefing.

16:01it in just two clicks. Feel ready? You can move to real money with as little as$100 once your account is approved. And the great thing is that in addition to crypto, Plus500 gives you access to a wide range of instruments. S &P 500, NASDAQ, gas, and much more. Explore equity indices, energy, metals, forex, and beyond. With a simple and intuitive platform, you could trade anytime, anywhere. Experience the fast, accessible futures trading you've been waiting for with Plus500. With over 20 years of experience, Plus 500 is your gateway to the markets. Visit us.plus500.com to learn more. Trading and futures involves the risk of loss and is not suitable for everyone.

16:38Not all applicants will qualify. Plus 500. It's trading with a plus.

16:48So basically, the community was kind of goading her into saying these things that would get her banned. I mean, this is kind of like an interactive version of watching Grand Prix for the car crashes. It's, you know, if you sum up the entire appeal just as watching Grand Prix for the car crashes, that's actually not that far off. It's pretty, people watch it because it's kind of cool to see it actually work, but it's really, really fun to see it fail. Talking about things that should not crash, I want to move on to our next story. this idea of a space alexa what the hell is that it's wilson from uh you know from castaway only in space and wilson really talks back you don't have to hallucinate it well if if uh if it starts talking back to you while it's turned off then you might want to actually turn it on and alleviate some of the space madness that might be setting in.

17:49It's a soccer ball-sized robot named Simon, C-I-M-O-N, which stands for Crew Interaction Module Mobile Companion. This was a small robot that was deployed onto the ISS several years ago and has been going through various revisions. It uses a speech model powered by IBM's Watson system, And it's designed to help crew deal with stress, primarily by providing them with verbally interactive instructions about the various experiment procedures and flight procedures and whatnot that they have to follow while they're running certain operations in zero G. Mikel, to exactly that point, let me just read this.

18:34This is from NBC News. I'll give you a little bit of the background and the context here. Quote, studies show that demanding tasks are perceived less stressful if they were done in cooperation with a colleague, Matthias Beinock, an IBM Watson architect, told NBC News mock in an email. Quote, Simon helps enhance safety because it can help serve as an early warning system in case of technical problems. So it's to me, my reading of that is it's using these technologies to engage the humans and find out when they're not engaged and therefore alert someone back on Earth to say, hey, listen, maybe you might want to have a pep talk with this astronaut.

19:14They seem like they're dragging a little. um i don't know about simon's uh human evaluation capabilities i do know that he's supposed to be able to uh identify or at least uh hint towards identifying human emotional states so he's theoretically capable of telling when somebody is stressed out but i don't know exactly what he's supposed to do about that the um i was gonna say it's probably the type of thing where once you develop the technology that can make the assessment, the escalation procedure is something that you probably build in later or enhance in future versions. If it's reliable enough, absolutely.

19:54Now, probably the coolest thing about Simon is that he zips around in zero G, like basically, he knows his way around the ISS. And he's got little jets that the that the AI powers. So if you tell and he's got a little camera. So if you tell Simon, he's like a zero ROG Roomba. Yeah, I like that. But like, you can tell him to go take a photo of the bean growth experiment or something like that. He knows where it is. He's got a little camera. He knows how to he knows how to operate it. So he'll actually like, you know, zip off and like, you know, go take a picture of the bean growth experiment and then fly back to you and show the image on his screen.

20:36So it's pretty neat. By the way, talking about vehicles here, we've got another story, this one more terrestrially bound. This comes from a Mercedes press release. Mercedes Benz is further expanding the use of artificial intelligence and integrating it into the voice control of its vehicles by adding chat GPT voice control via the Hey Mercedes MBUX voice assistant will become even more intuitive. An optional beta program for US customers will begin on June 16, 2023 for over 900 ,000 vehicles equipped with MBUX infotainment system. Soon, participants ask the voice assistant for details about their destination to suggest a new dinner recipe or to answer a complex question or receive a more comprehensive answer, all while keeping their hands on the wheel and their eyes on the road.

21:26I suppose we should say in theory, or it is to be hoped. In theory. And, you know, this is an interesting story because it does talk about attempts at integrating chat GPT and related LLM technologies, although I believe that Mercedes-Benz is actually integrating GPT itself, not just like a spinoff LLM. And so what this means is that you can chat with your car about a lot of the things that you can chat with ChatGPT about. It mentioned, for example, you can ask your car about what recipes you should make when you get home. And you know, Ash, mostly I kind of wonder why. Like, I, maybe you can tell me, but like, I know that when I'm driving, I don't particularly feel compelled to ask my car about my dinner plans.

22:20I might want to ask it about what restaurants are in the area, but that requires integration with local maps and with review systems. But like, I'm not hearing a lot of that yet. It might come in later versions, but right now it just seems like it's like, like, like it's solid, like it's answering a question nobody asked. I guess a lot of technology initially is a solution to a problem that nobody has. Look, you know, we were talking about this earlier, this sort of boom bust cycle. You know, one of the interesting things about this is the article that we pulled this from was from a publication called LifeWire.

22:58And the headline of that is why AI chatbots in your car could be a distraction nightmare. And I want to take the opposite side of this one just, you know, to play devil's advocate here. But also because I do sort of feel it that whenever there's a new technology that involves a car, the technology scolds come out of the woodwork and they tell you about how this is going to be incredibly dangerous. And, you know, my reaction to that is, have you ever been driving a car while your, you know, husband, wife, significant other sitting in the passenger seat says to you, why do you always do this when we're going to my mother's house?

23:30I mean, the idea that somehow technology provides a unique set of challenges and unique set of distractions to people when the person in the passenger seat really could be doing precisely the same thing. I don't know. To me, it just seems like this overblown fear that we have of distracted driving, this idea that somehow technology... I mean, you could be having this conversation with your sister, brother, best friend sitting in the passenger seat about how to make a perfect Apple turnover and be just as distracted. Is this like a tempest in a teapot, a lot of strum and drong about nothing? the modality required to interact with this voice system is complementary to the modalities that a human uses for operating the vehicle in other words you're not using your your voice and to a certain extent your ears quite as much as you're using your eyes and hands so i absolutely agree that it's not exactly preposterous, but kind of aiming at the wrong thing to claim that this is going to distract people from the task of driving.

24:40It's not like texting, which I think they're comparing it to. But why would it be more distracting than talking to a human being? I don't think it is. I mean, maybe there's a real answer to that. So I think that you're absolutely right that it's a Sturm und Drang, you know signifying nothing uh however i also uh i also don't think that it's particularly useful so like you know i i take the road that well there's always there's always that it's not it's not bad but i also don't really see the good now i'm more than happy to like to you know if i had a mercedes i'm sure that i would be more than happy to try out gpt in it but i just don't see it being a particularly compelling use case is all.

25:27Now, I will say this, that it is as long as there is little harm, and I agree with you that there does seem to be pretty little harm, that it is worth it to try to deploy these newly available technologies in fairly far-fetched use case scenarios, which I feel this is, just for the opportunity to be pleasantly surprised. Like, You never know when something that somebody like me might poo-poo in the outset ends up being a really, really surprisingly useful situation that a lot of people get a lot of value out of. We're going to take another quick break and be right back with more of the day's top analysis on the Real Vision Daily Briefing.

26:13Yeah, look, obviously, I'm not an automotive safety expert. Maybe there is some compelling reason why this might be dangerous. But to your point, I also don't see a huge amount right now, at least, of use cases. But, you know, and you alluded to this earlier, we may later see integration with things that could be, in fact, incredibly useful. Like, hey, how far do I have until I'm about to run out of gas? And can you direct me to the nearest service station that's going to be open at 2 o 'clock in the morning as we're driving down this country highway? I mean, that definitely would be something that would have some utility to it.

26:42The prompting of the language model inputs with the sensor readings of the car and various data from the navigation system, local maps and whatnot, does create the opportunity for a very, very compelling use case. and this is not this in order to really be great I think that it not only needs to simply respond to the human but also be a proactive contact in other words if the car had the ability to pipe up much like Gracie over here and and say hey I don't know if you've noticed this but you're low on gas or hey I don't know if you've noticed this but you're running out you know but your tire pressure is low, that might actually be something that people will get a bit of value out of.

27:34Hey, your tire pressure is low, and it's incredibly dangerous because it's 104 degrees out, and you might have a blowout. Perfect. Yeah. All right, talking about other use cases here, and I thought this was a really interesting one, which is Lina, AI for seniors. Talk a little bit about this story. This dovetails well with the idea of an AI that proactively contacts the user. And this story really takes a bit of a turn for the sad because we are talking here about a major problem socially throughout the West in particular, which is loneliness and abandonment of the elderly. and the problems that older people have with maintaining social ties and maintaining cognitive functions in the absence of social ties.

28:28You know, human beings really don't do well in isolation. And even if we're in like poor, poor quality social interaction still doesn't help. Like if you're, you know, in a nursing home or something, it's it's not it's it's better than nothing. but it's not great. So there have been a number of chatbot services that have arisen that seek to interact with elderly people in order to keep their mind sharp, in order to keep their memories coherent, in order to make sure that they feel like they're still connected with social tapestry. Yeah, there's also some interesting functions. This is a Medium post that you found, I believe, And it's interesting because one of the use cases that they use is for things like essentially becoming an agent that's the front end for a healthcare delivery mechanism.

29:23For example, in one example here, it says, you know, hi, Mary, how are you feeling today? And Mary replies, more pain today, didn't sleep last night. I'm sorry to hear that. Are the pain levels more or less than they were at Thursday's five, meaning on a 10 scale? I mean, you could really see how this could have the potential to be an early intervention for a senior citizen, maybe who's experiencing some symptoms. they don't think are severe enough, for example, to rise the level of calling a doctor over. But the AI chatbot intervenes early and says, hey, oh, so your pain is a seven today, right?

29:54And, you know, it's very warm. And again, this is something we were talking about earlier, the interactions with sensor data. You know, by the way, I've seen some fluctuations in your heart rate. I'm going to call your doctor. I mean, this really does seem as though it could be an intervention that has the potential to really increase the, you know, things like lifespan, health span, all those things that we're also interested in increasing. Very much so. There are capabilities to interact with sensors on heart rate monitors and whatnot. And there's also the capability to be seated with healthcare providers for that individual, their contact numbers, emails, that kind of thing.