In this bonus episode, our host Deep Dhillon and fellow Xyonixian Carsten Tusk discuss the recent release of OpenAI’s ChatGPT, which is widely considered the best artificial intelligence chatbot ever released to date. They discuss what this chatbot could mean for the general public, from combining and summarizing existing information as an assisted authoring tool to the usefulness of ChatGPT as a “personal genius friend”, or reference tool. They then dive into what problems ChatGPT and other large language, image, audio and multi-modal models can potentially solve in the future.
Dig deep and learn more in our articles on ChatGPT3 and Large Language Models:
Deep: Hi there. I'm Deep Dhillon, welcome to your AI injection, the podcast where we discuss state-of-the-art techniques and artificial intelligence with a focus on how these capabilities are used to transform organizations, making them more efficient, impactful, and successful.
Carsten: Hmm. With all this autogenerated generic knowledge that we're throwing out there and that we're amplifying with these AI tools. Yeah. How do we find the, the individualized content that is actually important that matters?
Deep: I would ask a different question. Does it matter if the non individualized content passes the test where, I don't know, X number of people love it. Think it's amazing, then doesn't.
Carsten: It does because it's not about whether you love it or whether it, it appeals to the masses. It matters whether it's unique. Is it new? Is it something worth you spending your time on?
Deep: I think that me personally, if I read something and I don't know the author, I feel like the goal of an educational system should be able to get us to the point where we can reason about.
And process it regardless of the source and Absolutely. Absolutely. But let's say, you know, so in this case, if the source is the AI mind, if you will, And it's good. And I can reason about it cuz I feel pretty confident that I can read something. Like you could, you could take Tucker Carlson's name off of something and put it in my feet and I'm pretty sure I'm gonna be like, that's dribbling crap and I'm outta here.
And similarly, you know, you could take somebody, some brilliant person, Plato, Einstein, whoever, and I'm gonna be like, okay, this is, you know, I can trap the, the thinking and the, the layout. I'm okay being confused about whether it's an AI or a human, if the reasoning and everything is there. The part that I find more concerning.
And I think it will get solved is the utter like lack of trying to like connect the things. It says back to citations. Like there's no citations, there's no referencing. I wonder if that is even possible. I mean, you, we could definitely do it. After the facts. Like we can take
Carsten: these things. Yeah, you can take the snippets and Google for them.
Yeah. Find, find somebody that, that said something similar to something like that. Yeah.
Deep: Yeah. And somebody will do that. And I think that's actually a good idea. First somebody to make the ultimate reference generator. Uh, cuz I think that stuff's gonna become more important so we can track these back. And in, in that context.
It makes sense what you're saying. Cause in that context we wanna track it back to a human. But what's the difference between the AI saying something. The AI saying something and a human standing behind what it said, like stamping it and, uh, and a human using the AI as a tool to say something that they don't acknowledge the AI for.
No, I mean,
Carsten: it, it's totally fine. And even if it would go and it would, uh, assemble articles out of let's say 30, 40 different articles and put them together in a meaningful way and it's all good, meaningful information, I wouldn't, I wouldn't say anything against it. Right. The question is how much of it is just repetition and replication of something that is old or something is already out there, and how much will it kind of like further modern the field of finding like new worthwhile things that I should spend my time with reading?
Deep: I don't, I don't know, like the, the thing that's been most impressive to me as I've been playing with chat g b. It really excels at high level thinking. And that's like some of the hardest thinking to do, like distilling thing, like the, the top three things to, and I, and I've been testing it in like really specific domains of our clients.
Like I was down in plastic surgery land and I was over in, uh, deep in machine learning and ai.
Carsten: Is that really true? High level thinking is, is one of the most difficult thinking to do, or is it just like maybe for me, for me, Of abstraction that everybody that doesn't actually know what they're talking about is blabbering about.
Cause very often in higher level thinking, people are talking about generalized concepts that are so vague that when it comes down to it, they would have no clue what to do and actually how to realize something like that. I'll give your typical business strategic talk that is somewhere up 10,000 feet of the clouds and they all repeating, repeating what the other people are saying.
Without saying anything. It's talk. But that's not, that's
Deep: that's bad strategy. That's not what I'm talking about. I'm talking about higher level summarizations geeks like you and me would listen to and say, yeah, that, that. That makes sense. I'm talking about passing it through your own personal filter.
What did I ask it yesterday? I, I took a, you know, a problem that, you know, that we're facing. I said something like, how can I use a large language model to generate high quality summaries? Of large conversations and how can I, uh, assess efficacy on those generated summaries. And at first time, it kind of just, I, I don't usually like it when it just spits out a, a paragraph of stuff.
So then I say retry, but give me an explicit list with details and the use of examples. And I looked at that and I'm like, that's a high level proposal. It's not exactly what I would write, but it's pretty much a high level proposal. And I looked at it and I'm like, nothing in there is wrong. And this isn't subject matter that I don't know.
I mean, I know, I, I know if it's saying something stupid. Yeah, yeah, yeah, yeah, yeah. I'm looking at it going, huh, okay. And then if there's something that I'm like, I need a little more detail on, then I grab like, It'll list like seven, seven steps to do whatever. I'll grab number three and I'm like, and I'll rephrase that as a question and I'll, again, I always seem to have to ask it to add, uh, examples and detail and then it tear, tears it apart a little bit.
I mean, it's just something, I don't know, man. We're in a different place than we were a week ago. , we're in a different place. I mean, there's a reason everyone in tech is obsessed with this thing because last Wednesday my feed starts getting all this stuff in it and I'm like, whatever. Thursday my feed starts having, you know, people making like really bold claims that I'm like, who is this idiot?
Like, why are they saying that Google's irrelevant and blah, blah, blah, blah, blah, and I . And then by Friday, by Friday, it's like people whose opinions matter. To me and the world saying the same crazy things. And then I'm like, I gotta play with this. So then I start playing with it on, I figured 10 minutes, I'll be outta here.
I'll be able to fill it in with my, you know, like what I've made up on large language models in my head before, but I feel like there's like a tipping point. You know, like if you think pre-web. When we, we had the internet. We had Darnet, we had email, we had all, you know, Archie at Mosaic
Carsten: as a search engine.
Deep: Before Mosaic. Before Mosaic, like I'm talking 1992. We had all of the, the raw ingredients and then an HTML browser. Mosaic comes along. And just kind of stitches it together a little
Carsten: bit. Well, it, it's all about, it's all, it's all about discovery. It's all about finding things that are relevant to you at that moment in time.
And that's how the whole search was born. Right. Well, that,
Deep: that little ux tweak, which I would argue like that's kind of what the, the browser was the UX tweak.
Carsten: Yeah, I guess it definitely was a ux, HTML was a ux, like it was a way of, um, it was a publishing system. Really?
Deep: Yeah. But it exposed the graph in a way that was navigable for the first time.
Carsten: that, we had BBS systems, so that's kinda like the precursor to it, right? Where everybody had like their own little, um, content management system that they would give you access to and you could play with it and read it. Yeah. Yeah, exactly.
Deep: So I would, I would argue it was like, it was something.
Intellectually you would think of as fairly small, but it was like a tipping point. Like as soon as everybody saw that, I mean, I remember sitting around in a computer lab in, in grad school and, and there was like five of us overlooking somebody's shoulder and they were like, yeah, this just came out yesterday.
And there were, there was literally 80 websites on the internet at that, um, at that moment. And we were looking at it going like, oh my God. Like, but the good, the good thing the next 30 years of my life just flashed before me. I'm like, I gotta change everything. This is what I'm working on. Like, Right. I feel like that something just flipped and, and you know, we've been working in machine learning and AI forever,
Carsten: but what does it, let's come back to, what does it lead to?
I mean, great. We have this thing. It's good at like, combining, summarizing existing information. I, I still would say it. There's no reasoning behind it. It does not come up with like novel and new concepts. It just doesn't, right. Um, What does it lead to? I mean, on the one hand, yeah, content explosion. Right now, everybody that knows how to use this will use it as an assisted authoring tool and will just get a lot more content, I think.
I think the problem with the internet today is there's already too much content and the real challenge is you have only seen so many hours in the day. What should you? And you had a perfect example earlier because you said, well, you know, last week I saw my feeds explode and at first I didn't pay attention to it.
And then, you know, there's more and more on the same topic and uh, then you're like, well, now I have to look what this is. Right? And, and of course, yeah, if something goes viral, that's how we discover content that many other people, you know, are discovering. But how much of it gets lost, right? Even if I browse through my Apple News in the morning, it's kind of like, I could do this for three hours because every day there's so much freaking content.
What I would like to know is what should I be reading? And at that point, it becomes literally dangerous, right? Because. Now, not only there's no objective way to do that. You're at the, at the, um, mercy of the recommendation engines, but technically bias you.
Deep: Yes. But that's where I think the conversational piece is kind of interesting.
So like yesterday, I just did an experiment. Once I started getting infatuated with this thing, I'm like, okay, every single question I have in my head, I'm gonna go to this thing. So the first, it's early morning. I'm like in the office, it's kind of dark. I'm trying to get moving, so I just ask it like, Hey, you know, what music should I listen to right now?
And I gave it a little bit more info, like I gave it some genres. It's dark , it's dark, I'm in a somber mood. I just need to get a little bit of motivation. And I gave it like, I didn't even give it specific band names, I just gave it a couple of genres. I kid you not, it came back with a totally obs to me, at least obscure French male, electro pop, singer.
And I never told this thing that I've been obsessed with, like weird French lounge music for the last, you know, year uhhuh. I don't think it even knows. I don't know how the hell it figured that out. It gives me this guy and I'm thinking, what a random thing. I figured it'd be like the Neil Diamond of of France.
I put it in, I'm like, oh my God, this is so good. And it was the perfect thing to be listening to. Right? That I don't know. I, I'm just gonna run some experiments for the next couple weeks. I'm asking this thing
Carsten: everyth. Like for the longest time, I, I'm a strong believer that the dream pool is limited, right?
Mm-hmm. , you see that, you see people that look like other people, you know, we are not all that as individual as we think we are, right? And so if you really can take. The experiences of 8 billion people, you'll find a lot that are like you probably mm-hmm. , right? Mm-hmm. and this music example is something like that.
Like, kind of like in this particular scenario and situation, what should you be listening
Deep: to? I know, but I'm, I'm so used to, you know, the 30 years of machine learning that we've been doing. I have a mental model of like, oh, I have to give it a bunch of history and give it a bunch of examples. I'm not used to having, you know, a three prompt, a three chat response turnaround and getting something like that.
Have data, have a hypothesis on some high value insights that if extracted automatically could transform your business. Not sure how to proceed. Bounce your ideas off one of our data scientists with a free consult. Reach firstname.lastname@example.org. You'll talk to an expert, not a salesperson.
And the other thing I did yesterday is that, you know, I'm,
Carsten: I'm in a, it's kinda like, it's kinda like a mixture of the, of one shot learning and feature importance. Yeah, it's literally like your one shot is like your most important feature. So what else is out there that has this one thing as its most important feature?
Deep: feels like brain augmentation on some, like yesterday, I'm, I'm in a conf uh, a meeting with a client and I, I'm about to like, say something and then I'm like, oh, wait, I'm on chat GBT experiment time. So I'm gonna ask chat gbt first. I ask it during the. Yeah, during the meeting I ask it something, it gives me something.
I'm like, oh my God, I didn't think of that. And then I like assimilated into what I said and like that is, that is nothing that I've ever done. So it's
Carsten: like your personal little pocket
Deep: advisor now. It's your personal genius friend. Yeah. Like it's, it's crazy. Like what does that mean when we all have a little personal genius friend and I, I get, well, I really get the dangers of it saying something stupid, which is why I'm, you know, I'm you YouTube,
Carsten: but, but it's not so different, different from Google or a search engine in that way.
Right. Because you, how often would you do that? You sit in a meeting and somebody talks about something you don't know. You Google it real quick and you say, you know, but now it's hard to, to digest the content because you got like 10 search results. Maybe at first glance you can kind of figure out what it is about, but you don't have time to read the articles.
Yeah. And so this thing goes one step
Deep: further. You get a five minute experience and put it into a 15 second
Carsten: experience. Yeah. And this thing, like I said, this thing goes one step further and. It read the articles for you and it gives you like, you know, the cliff notes. Pretty much the
Deep: part that really blows me away is just the, the trajectory of this stuff, right?
Like, like all we do is machine learning. We're constantly doing machine learning, and I just feel like every day there's some new like, massive model. That's crazy. I wasn't even thinking about GBT a week ago. I was still digesting, you know, like Dolly and some of the other stuff, right? That, that's come out and, and now.
Like, but you, but you also have to, these guys are already saying G PT four is around the corner and it's, and like the word on the street is, it's multimodal. It's, it's gonna be a one and a half trillion parameter model. It's gonna be texted to audio, to video to, I
Carsten: was gonna say the, the last thing I've seen is like video generation, right?
They're working on video generation was,
Deep: yeah. But the open AI guys, like the next model, I don't know if this is true, we'll find out when it comes out, but that it's, it's multimodal. And, uh, and it's gonna do videos. So it's gonna be like, Hey, surreal is seen of, you know, giraffe walking down the street in, you know, in 18th century, you know, Paris or something, and it, and gimme 15 seconds worth of boom.
There you go. But, but
Carsten: I guess we have to ask ourselves. It's all nice and it's definitely like worth the marketing or the sensational hype. It's so was Dolly. It's kind of fascinating. It's fun to play with. What you really have to ask yourself though is like, what problems does it.
Deep: I think we're just at the very beginning of unpacking that, but I think certainly with respect to anyone who is authoring content, it does not make sense to me to not at least learn from this thing, like to leverage it.
That's if you think about if
Carsten: you, if you are authoring content, For the sake of authoring content. I agree. If you're authoring content because you really think you have something to say, you should write it yourself.
Deep: Yes. I'm not saying you're not writing itself when you're consulting this thing. Like, like think of it as your genius buddy.
And you're working on your research project and you're just like asking, you know, like you're just telling 'em what you're up to. I think of it more like that. I don't think of it as like, oh, I'm just gonna like hit this thing. It spits out a bunch of stuff. I'm gonna
Carsten: publish it. Like, like research is a good example, right?
It this, this thing can like write articles about existing things. It could never publish a new research paper. I
Deep: mean, people publish survey articles all the time, right?
Carsten: Right. But then back to the regurgitating things that are already.
Deep: Yeah, no, I mean, you're right about that. Like there's a few things that it excels at and there's a few things that it clearly doesn't excel at.
I mean, one of the things that it excels at is this like, it's like the best list generator I've ever seen in my life. , like, it makes the best lists of, of anything I've seen. Um, the, the other thing is it makes, and, and that includes like, lists of questions to ask. And as soon as you start realizing that you can use it to help you figure out the questions to ask, that's like next level.
Interaction, you know, and then you can ask it the question once you've got the listing questions, and you can start torquing them. I feel like I'm not gonna really be in an ideation session again without consulting it. I feel like I'm never gonna send a proposal out again without consulting it.
Carsten: I, I, it sounds like it's gonna be a great writing tool.
Yeah. Um, it sounds like it's writing, thinking, writing is, it's going be,
Deep: if it's done right.
Carsten: You hope so? You would hope so. . You hope so. Um, if it's really as good as you say, it might actually find its place, like you said, it might find its place next to.
Deep: Yeah, and I, if I was at Google, I don't know everyone there should be figuring out how do we integrate this into Google if they aren't already.
I'm sure they have been for a few
Carsten: years. Google, in a way, has something like that integrated in its prompt, like depending on what you Google for, if it's really a question or something, you know, the first thing that pops up in Google, which is very often just the answer to your question. And then the search results come.
Yeah. So they do have that in a less, like, you know, we write an article about it, but we try to figure out what you mean and what you are looking, what you wanna solve.
Deep: But they don't really have anything that, like Google's a, uh, ultimately is like a reference lookup. They're like directing you to places and
Carsten: the, and the calculator.
And they, they do stuff like, um, you know, events are covered like the FIFA World Cup. If you're asking what other stores, there's definitely like all the one box thing and all you can, you can ask stupid questions like converting Celsius to far. Tough. Right. Get the answer right away.
Deep: So, yeah, I mean, I think they're gonna continue to release features that are based on this P capability.
I mean, I don't think Google's going away. I mean, I think Google has every capability of, of matching what we see here. And, and
Carsten: even I see this as, I see this as complimentary. It's, it's, this is a trajectory
Deep: Yeah. That, that, that global AI community is on. And you know, this kind of multimodal reasoning is gonna continue to go.
Carsten: And it's content gen, it's, it's not even reasoning, it's just content generation. Well,
Deep: I think content generation is a demeaning term for what it's actually doing here. The content generation implies that you're just writing shit for the sake of writing shit if you're authoring something.
Carsten: So that's what Dolly does, right?
It's generates shit for the sake of generating shit, for example. Okay. Well,
Deep: I don't know. I haven't spent as much time with Dolly as I have with this. It generates
Carsten: images based on your.
Deep: Yeah, but that's like, that's also saying that, you know, if you're an artist, you know, and a Picasso that all you did was generate shit for the sake of generat shit.
I don't think that's true. I think what Picasso brought the world was a different way of perceiving reality. Right. And I think at fundamentally differently. I
Carsten: think that's, that's exactly, that's a good example. I think that is exactly what is lost in something like Chad, g p t in Picasso. He had a vision, so, or he would like project in a way.
How he sees the world or how, how he imagines something. He has an intention behind this image, right, that comes from him and he's trying to express it. In a painting, in church pt, it's a little differently. It's kinda like you have an idea too, but you're not expressing yourself. You're trying to like, It's more like how other people would have presented the idea.
So the idea is there, but not your projection of the idea, if that makes any sense. Right. Your prompt is
Deep: your idea, but that's where you really gotta spend some time playing with Chad gbt. That's where you're interacting with it. You are. The puppeteer of this very powerful puppet. You're steering it, you know, like I'm having it author, a college essay, you know, for of my kids' friends.
Carsten: But it's, it's kinda like, you're just like, you're just like trying to, um, like twiddle with it until you like what it
Deep: wrote. I think you're trying to get it to a starting point that you're gonna go take and do something with. In
Carsten: that case, you are the reviewer, you're not the author. Like I said, you're like the editor of the newspaper that gives us Yeah, but every editor is underling an idea.
Yes. And then reads the article and then says, no. I think that's a very, I don't like that section.
Deep: I think, I think you, you play the role of editor until you cut and paste what's in there and tweak it, and now you start writing on top of it and now you're the author. Yeah. So
Carsten: anyway, I mean, definitely super useful.
Deep: I don't know. I'm just fascinated cuz if you look at it,
Carsten: I'm a bit scared about the spam. We're gonna see.
Deep: That's inevitable. It's gotta be spam, but it's already all over the internet as of Wednesday. But how
Carsten: we gonna distinguish the spam from the real stuff now ?
Deep: Well, I think Providence is still important.
Like, you know, so and so works for such and such company and is a real human as opposed to so and so is an AI bot created by so and so who's out doing whatever. Wow. And Providence is gonna be key and being able to like prove Providence.
Carsten: And the same, the same way that we get Google articles, right? It's kinda like basically peer reviewed, like people that link to content.
I guess. I guess the ranking of content doesn't really change. Just the sum.
Deep: I mean, I think there's a, there's gonna be startups that emerge and companies that try to help us navigate, but even just the providence isn't an easy problem with this stuff. You know? It's like, I mean, I, I can see. Like the legal world is going to get taxed with questions.
Somebody's gonna go into, you know, something like chat, g p T, they're gonna have it author lyrics for a song. They're gonna go sing their song, they might have the audio version and render the chords and everything. You know what? We need somebody, somebody else is gonna be like, you ripped off so and so
Carsten: and so and so and so and so.
We need patent G P T.
Deep: Yeah, , but that's where I'm saying like, reference checking capability is gonna be super powerful. Like someone needs to make automatic reference checking so that you can like be an honest, you know, user of this thing. Whereas right now I don't think you can feel safe in just cutting and pasting it into, you know, to your college essay or whatever, or your, your exam or whatever.
You have to manipulate and change it a ton. And even then it's. It's sort of like, did you plagiarize? Like you don't even know. Even the machine doesn't
Carsten: know. You don't know. You don't know whether or not you copied like somebody's stuff verbatim or if it's assembled or if it's like somehow put together or rephrased or paraphrased or who knows?
Deep: I think we're in a new world. It's, I think we're in a new world. That's all for this episode. I'm Deep Dhillon, your host, saying Check back soon for your next ai. I. In the meantime, if you need help injecting AI into your business, reach out to email@example.com. That's x-y-o-n-i-x.com. Whether it's text, audio, video, or other business data, we help all kinds of organizations like yours automatically find and operationalize transformative insights.