We’re seeing so much generative AI art today: text, images, even video created by smart(ish) machines. I can’t get the image out of my mind of “The Creation of Adam” by Michelangelo in which God reaches down to touch Adam and stir him into life.
In this TechFirst, we’re going to chat about generative AI:
- What it can do
- What it means
- How it will change the world
- And how it might change us
Our guest is Alex Cardinal, CEO of Glimpse.ai. They have two AI projects … Article Forge, that generates articles on a topic based on a keyword, and WordAI, which rewrites content uniquely in the defined style. Keep scrolling for the full video and transcript, or check out my story on Forbes here …
(Subscribe to my YouTube channel)
Generative AI: chatting with an AI exec about the future
Subscribe to TechFirst:
Apple Podcasts
Spotify
Google Podcasts
Overcast
Other podcasting platforms
Transcript: How AI that creates will change the world, with the CEO of Glimpse.ai
(This transcript has been lightly edited for length and clarity.)
John Koetsier: We’re seeing so much generative art today — that’s text, that’s images, even video created by AI. It’s getting to the point where I can’t get the image of “The Creation of Adam” by Michelangelo out of my mind, that’s God reaching down to touch Adam and stir him into life. In this TechFirst, we’re going to chat about generative AI, what it can do, what it means, how it’ll change the world, and maybe how it might change us. Our guest is Alex Cardinell. He’s the CEO of Glimpse.ai. They have two major AI projects. One is Article Forge that generates articles based on a topic, a keyword… and WordAi, which will rewrite content uniquely in the same style. Welcome, Alex.
Alex Cardinell: Thank you for having me today.
John Koetsier: Let’s start pretty high level, pretty broad. I mean, probably even a couple years ago I was hearing people talk about it’s the golden age of AI, right? As machine learning and other things started really coming into their own. What are we building here with all this generative AI?
Alex Cardinell: Yeah, definitely. Well, and it’s been over the last couple of years monumental in terms of the impact that’s been made. Just like, you know, a couple years ago when you were thinking like AI that could create content, you were imagining this kind of machine-written gibberish, it looked like, you know, like even your four-year-old kid could probably write something better.
And in what was basically like six months, it went from kind of this novelty that you kind of laugh at to something that’s actually generating really high quality content that humans can use, that can even in many cases show [the] same quality as humans.
And it’s really even just the tip of the iceberg in terms of what’s then possible in the future.
Like, it feels kind of in many ways, like the AI revolution is upon us, but I think we’re even in the next couple of years, yeah, kind of just hitting the beginning.
So yeah, what’s coming in the future? Certainly, like lots and lots of high quality across all those, all the different ways of generating — that’s art, that’s text, everything you mentioned — where right now maybe it can kind of take some ideas you had and put it on paper. But we’re getting to the point where pretty soon it might even be having some ideas of its own.
So it’s a really exciting time. And I think the next like three, four, five years are going to have an impact that’s bigger than really, even people in the AI industry, even bigger than kind of what we’re all anticipating.
John Koetsier: Let’s explore that a little bit because it’s interesting, we keep talking about metaverse lately, and there’s a lot to build there, obviously, but I’ve seen AI just in the past couple weeks that generates video. Literally video. Show me video of a forest, you know. We’ve seen creative diffusion, other things, things like that and you’re making incredible art and images and stuff like that. But literally video… forest, river, mountain, city, that sort of thing. The implications are really, really staggering.
What does it mean to have AI as a creator? I want to ask you, do you ever get kind of mystical about it? Think about the singularity when technological growth goes insanely exponential?
Alex Cardinell: Oh, definitely. I mean, just in terms of what that can even do for society. Like if you imagine, okay, well, right now AI’s able to create this video. I think a lot to what will happen when AI’s at the point where we can direct it towards like how do we cure cancer, for instance. And even in realms like that they’re starting to be, it turns out like a lot of the same AI that’s being used to generate text, it turns out, yeah, it’s great for generating images. It’s great now for generating video. It’s even great for coming up with, like, molecules and drugs that can be used for curing cancer.
And then, yeah, the most exciting part of it all, what you mentioned is when maybe AI is also at the point where it can start writing the code that will make its own AI even better.
And there’s been lots of research. It’s at the early stages right now, but there’s been lots of research in that. And that’s like where the true singularity is, is when it can kind of set itself to improve itself, when it can start to improve itself better than what a human can. And the fact that we’re already making great progress in that… it’s impossible to speculate, like, what society could truly look like in such a situation. But I think in most of our lifetimes we’re going to experience that. And for me personally, I can’t think of anything more exciting than that.
John Koetsier: It’s definitely exciting. But at that point, AI may not decide that curing cancer is on its list of priorities [laughter], or that maintaining this race of, I don’t know, bio humans, it’s really important. We’ll see how that goes. There’s some freaky things to think about there, for sure. Let’s talk about your tech. You make tech that writes, how good is it?
Alex Cardinell: Yeah, so I would say that right now it will get you maybe 90% to 95% of the way there.
So, a big area that we focus on because there’s a lot of other kind of generative AI out there — you mentioned text and video, images — a lot of what we focus on is making the text that we’re writing as factually accurate as possible. That’s actually, surprisingly, you would expect that AI text would be like it kind of looks like a robot wrote it but it’s perfectly accurate.
Like when people imagine that super intelligent machine, I think that’s a lot of what’s conjured up. It’s oftentimes actually the reverse, where AI has done a really, really good job at figuring out how humans speak, how humans write, what persuasive text looks like, how to kind of paint the picture of a story. And then often where it falls flat is knowing what the actual correct fact is, and that’s where a lot of our work is centered around.
And really our goal is to make it so that when you’re writing text, especially like Article Forge where you enter the keyword, get the article back, that you’re able to rely on the accuracy and that it isn’t spreading misinformation. It didn’t get wrong like what your company does, when your company was founded. So I would say still there’s a lot more progress to be made. At this point, it’s generating text that you’d want a human proofreader to look over a little bit. You’d still want a little bit of fact checking to be done, but the fact that you can get it like 90% of the way there is still able to make a really big impact.
John Koetsier: It’s interesting because I’ve been able to get early access to the tool and test it and play with it a little bit. And I just literally, 20 minutes ago, asked it to make an article about Apple’s new iPhone 14 Pro. And so it generated about 500 words on that and tells me that the 14 Pro is splash, water, and dust resistant. It’s a new phone, 6.7 inch ProMotion technologies, touchscreen, 48 megapixel true depth camera, photonic engine.
So it’s getting a lot of details there that are fairly new, right?
What is interesting and actually kind of corroborates what you’re just talking about is it says Apple has released and then a paragraph later says that it will probably increase in price [from prior models]. So it doesn’t correlate those two things, right? That it actually has been released and we know the price now and it found material that said, “Hey, it will probably be more expensive.” So that is quite interesting.
One thing I do — because I’ve done a couple different tests with this — one thing I do kind of comfort myself with because I write [laughing]. That’s one of the things I do. I write, and am I gonna be replaced by a machine? I comfort myself with is it doesn’t have what I would call personality. I haven’t seen humor for instance or other things like that. I assume that’s coming.
Alex Cardinell: Yeah. So I would say that a lot of ways the direction in which writing and becoming a writer I think is going to go is, as a lot of kind of the more mechanical aspects of writing text is being replaced by machines, a lot of it will be the steps that come before and after. So the steps before would be like, for instance, on the research side. That could even be like the call that we’re doing right now that, you know, there’s a lot of work in figuring out what it actually even is that you want to write about.
But then a lot of it too is on that step at the end where maybe you have all the information like kind of on a… you’ve written it all down, you’ve figured out the things you want to talk about, you have like decent pros around it, but then you really want to give it like your touch.
And certainly, that’s something that we want to do a better and better job of. A feature that’s coming out in just a couple months is where you can specify the style, where you could say, like, “I want this to be written kind of witty, or kind of playful, positive, dry,” whatever you want it to be.
But still, the AI is not you, and you have something special that you want to be adding to a piece of content, a reason why your readers want to read what you’re writing versus someone else. And a lot of that can come still at the very end. But still, that’s like an important part of a lot of text.
So, I think in many ways, tools like this can be empowering to writers where I think, I don’t want to speak for all writers, but I know at least for me personally, I think that that research step and that step at the end where you’re giving it your personality are where most of the fun is and the kind of parts in the middle where you’re just trying to get all of your ideas then onto the paper itself. And that’s something that then the machine can, in a similar way to, you know, like when Photoshop was first created, that didn’t replace all artists. In many ways, it empowered them to focus on different things.
John Koetsier: It’s interesting as you’re speaking there, I was remembering… I think it was about three, four years ago, and I think the Associated Press or Reuters were testing technology that was writing news articles. Now, very specific ones and these were sports write-ups, especially for I think high school baseball teams and stuff like that, right? Where a machine could take the box score, take the record of the game, what happened, and actually build a bit of a narrative around that.
And that was doable in the time and actually it kind of even really worked. I mean, you’d kind of buy that it was done… you wouldn’t assume it was done by a machine because there’s very defined rules and principles about how to report on a baseball game or different things like that. This is another order of difficulty, right? Because what you’re doing, you’re just saying, “Give me a random keyword.” You have no idea what the world is going to give you as a keyword, and your AI has to basically say, “Okay, what do I know about this? What can I learn about this? What can I write about this?” That’s quite an evolution just in a few years.
Alex Cardinell: Definitely. Yeah. And the long tail of that is really difficult where, yeah, users can enter and they do between all 4,000 of our customers, they enter the absolute darndest things across from, you know, like what you entered, iPhone, to like why should I fly this airline to Bali? Or what is the return policy for Macy’s? Every single thing you could possibly think of and quite a few things you didn’t realize anyone would ever even want to write about.
And so, yeah, I think a really big way in which AI has really evolved in the last year or two is these kind of like deep learning, kind of machine learning algorithms and models that have come out in the last couple of years have done a really, really good job generalizing where when you were writing those articles around like baseball, as you mentioned, just like very formulaic rules. You knew like if a run was scored in the ninth inning and then the team went from having less runs to more runs that that was the game winning run. You could code that in, you could say that’s what that’s like. And there’s these four ways of describing that, so when that happens, write one of these four sentences.
But when you’re writing about any topic in the world, you need to have something that’s very, very generalizable where, with a lot of the work we do to make things more factually accurate, we’ve kind of also built this almost like search engine for machines called like a knowledge search engine to give the AI we’re writing the right information as it’s writing. But still that then requires this next step of given any information you could possibly imagine in the whole world, how do you turn that into a coherent sentence, paragraph, article that’s able to kind of cover that information and add value to the user?
John Koetsier: Right. Right. And if you’re watching the video of this, by the way, and you wonder why Alex brought up Bali that’s because he’s there. And if you’re wondering why he’s shaky, that’s because he’s holding his laptop. He doesn’t have a proper office where he is. So thank you for that. We’ll see if I can get some AI to stabilize that video.
What gave me a little bit of… I don’t want to say hope, but what was interesting when you mentioned some of the queries or prompts that people are entering into your system, many of those are things that no person would ever write about. And one of the reasons that Reuters or the AP was looking into this technology to do write-ups about high school baseball games is that nobody was going to write about that. So that’s interesting because if nobody was going to write about that anyways, somebody, something writing about it adds something that some people will find useful and that’s a good thing.
I guess the question will be at what level does it become, hey, we don’t need a human to write, you know, give the AI a topic and boom, there you go and it’s amazing. You write it in, I don’t know, Ernest Hemingway style. Write it in, you know, write it like an evening columnist for the New York Times, whatever. When do you think we’ll get there?
Alex Cardinell: I think we’re getting there.
I think I’d say people continually underestimate how long it’s going to take. I think that we’re in the span of just a year or two. And I would agree with you, I think that really where AI can be empowering is in that long tail when there’s like non-consumption with the alternative, where you could not afford to create that content in the first place.
And you can imagine that with like these very obscure topics. You could even imagine that for news where maybe there’s something that happened in your local neighborhood where only 20 people want to read that article and then it doesn’t make sense for a human to write it. You could also imagine, even almost as like an extension of a search engine for instance, where if you’re, I think Google said that like 25% of search queries that people enter are things that nobody has ever searched before.
So, you can imagine that in many ways the next version of that could even be, well, I want something written for me that’s covering a topic that nobody else has ever written about. So, yeah, technology’s moving incredibly fast. I think that we’re in the span of just a year or two from a lot of those being realized.
John Koetsier: That is fascinating. And what comes to mind is that we now have virtual celebrities. We have celebrities with millions of followers, essentially influencers who are constructs of AI, constructs of code, who are not actual human beings. They are… in fact, people, in some cases you can buy a representation of that and people have married that [laughing]. I’m talking Japan here, okay, so it’s a little different but… we see that.
That is really interesting because one could assume perhaps that not only are you developing the base level technology, but you could develop a personality for your AI. You could develop, or somebody could white label or say, “I want to use this, but as a certain personality,” train that personality, give it a certain backstory/background, and develop a writing style that is unique and that could be somebody that people, something that people follow and listen to and look for. Does that sound insane or does that sound probable?
Alex Cardinell: Actually, well, Google just recently released a research paper about this model LaMDA, which had showed really, really groundbreaking results in terms of like a chatbot to the point where I think one of their engineers was convinced that it was sentient and even I think got fired over it.
John Koetsier: And he got fired. Exactly.
Alex Cardinell: Yeah. He got fired over it.
And then actually I think two of the researchers who are working on that then quit the company to create their own startup centered around creating personas that you could chat with. So, I think that that’s coming right around the corner and it might even be in the span of like six to 12 months, you’ll start seeing prototypes surrounding that getting released.
So, now the next step of like animating it, of, you know, having it have like maybe unique text to speech, a lot of it too is getting, right now it’s similar to Article Forge and Text Generation where you’re getting it to like 90% of the way there and then still you need to do a little bit more to get from that 90 to 100. But for both, yeah, it’s…you can see it coming.
John Koetsier: This is an insane world we’re building because we’re generating fake art. Not fake, it’s real, but it’s generated. We’re doing that with images. We’re doing that increasingly with video. We’re doing that with music. And now we’re creating text that is generative as well, AI-generated. The nature of reality is one that is seemingly fast outracing our ability to understand it. What is real?
Alex Cardinell: Our human evolution did not…our biological brains were not prepared for this. So it’s going to be a brave new world for how a society…I mean, I would even argue that our brains are not well adapted even for the current society we’re in right now with like social media and information bubbles and things like that. Like a lot of things were kind of off script, out of distribution for what the human brain was built for. And that’s just going to keep on continuing.
So I think that AI is going to provide a lot, a lot of good — and I’m like the ultimate AI optimist — but it’s also going to lead to some major, major societal changes in large part just because, yeah, our brains and our bodies were, in no point in our history were built to adapt for the future that’s coming for us right now.
John Koetsier: Well, I guess we have to upload ourselves in and [laughter] overclock our brains. Increase our storage capacity, add a few CPUs, and maybe possibly we can handle it. Well, this has been fascinating, Alex. Thank you for taking some time. I know it’s like 11:30 p.m. for you right now, so you’re pretty sharp for that late. But thank you for the time.
Alex Cardinell: Yeah, I appreciate you having me, John. It was a pleasure talking with you.
TechFirst is about smart matter … drones, AI, robots, and other cutting-edge tech
Made it all the way down here? Wow!
The TechFirst with John Koetsier podcast is about tech that is changing the world, including wearable tech, and innovators who are shaping the future. Guests include former Apple CEO John Scully. The head of Facebook gaming. Amazon’s head of robotics. GitHub’s CTO. Twitter’s chief information security officer, and much more. Scientists inventing smart contact lenses. Startup entrepreneurs. Google executives. Former Microsoft CTO Nathan Myhrvold. And much, much more.
Consider supporting TechFirst by becoming a $SMRT stakeholder, connect to my YouTube channel, and subscribe on your podcast platform of choice: