Hello, this is Darren Pulsifer, chief solution, architect of public sector at Intel. And welcome to Embracing Digital Transformation, where we investigate effective change, leveraging people process and technology. On today's episode, exploring generative air use cases with returning guest Dr. Geoffrey Lancaster. Geoffrey, welcome back to the show.
A third time's a charm. Thanks very much. Appreciate it. Hey, we had so much fun talking and we've been talking a lot. I wouldn't say theoretical, but not brass, not brass tacks, not day to day use.
And what can I do today? Let's talk about that. Let's get into. All right. I got a GPT account. I got a cloud to account. I've got a with Bard account and some other ancillary ones too.
I may go into playground API to go generate images or whatever. There's a lot out there. Yeah. So what do I do with it all? Great question. Where do you wanna start? I Yeah, I brought that up. Where do you want to? The world is, like, way open for anything now.
Right. So when. When. Whenever I talk to people about Where do you want to start? Right. The question that I start with is what type of data are you trying to either use and what type of data are you trying to create? So, for instance, you want to start with texting. Go to image.
You can start with text and go to text. Do you want to start with text and go to music? You start with music and go to image. Do you? What is your input and what is your output? And I think if you can pin those two things down, depending on a given use case and then we're going to talk through different use cases, is going to arrange different tools that people are going to use. So you mentioned a couple. You mentioned Chatty Beauty, you mentioned Bard, and those are both really going to be text in, text out. Something like Mid Journey is going to be, let's say text in, image out.
You know, there's going to be variations on that and on daily and on stable diffusion where you can do image in, image out. There are tools that you can do image in video out. There are ways that you can do audio in, audio out, audio in, text out, text in, audio out. And so you know where to keep us kind of from boiling the ocean about ways that people can use this. Let's start by just admitting there's a little bit of creativity that has to go into this. It's not just, oh, I saw this thing.
I want to do that exactly now. But the creativity is around what what am I working with and then what am I trying to accomplish? And based on that, I'm going to select a set of tools that gets me there. Okay. So what you just said is really interesting. I want I want to key in on creativity.
So these are all your what is it? Right brained or creative or left Ranger Creative. Refereeing or creative left brain? Your analytical. Okay. All right.
All those analytical people are going to be left in the dust if they can't be a little bit creative, right? I mean, because like you said, there's so many opportunities, so many possibilities. Yeah, I've already started playing around. So the more creative you are, you can really do some amazing things in in creating things. So is it safe to say that these A.I. generative A.I.
is create is unleashing creativity? And you and I talked a little bit about that previously, that what the human brings to it in many ways is that creative component, which maybe that creativity is about asking the right question. Maybe that creativity is about stitching tools together in a way that the tools themselves aren't meant to be used kind of packing things together. Maybe the creativity is coming from this scenario or the use case that you're actually trying to solve for. So yeah, I do think there's creativity to it. On the right hand side of things, but there's also your left brain, like you actually have to be able to implement that. And so do you have people who can do that? Do you have partners that you can work with to do that? Do you want to try and do it yourself and like really get under the hood and figure out how do the pieces all fit together? And maybe you want to code it yourself.
All of those are possibilities. And I know, you know, we want to kind of get to specifics, but I think it's worth starting there just to recognize that the specifics that we're going to talk about are not the only options. There's people doing amazing, amazing things with these tools because they're bringing that creative mindset to it.
They're bringing a problem to it that they can't solve with Google, that they can't solve with sort of traditional search. And by Google, I mean like Google searching that they can't solve with maybe a traditional analytical approach. So all of these other tools in their toolkit aren't going to work.
And so they get to a point where they say, you know what, I need something that's going to produce something new. Therefore, I need to use Jeremy AI for that. Okay, So this reminds me a little bit of in the nineties about I learned about brainstorming ideas and things like that in group settings and things. And so I've already done this. I can brainstorm with an AI now.
Yeah, which is pretty cool. It gives me a second opinion. Yeah. It doesn't mean it's the whole opinion, but it gives me a second opinion and then my brain went into weird places like couldn't I do group and have an AI in my group brainstorming session? Sure, kind of, because part of the hardest part of brainstorming is coalescing and bringing everything together. What a great opportunity for an AI to listen to a conversation between a group of people brainstorming and prompting and moving, moving them into having a consensus and and coalescing an idea. Well, and I'll I'll go you one further, which is instead of a second opinion, what if you're a I can give you a second, third, fourth, fifth and sixth opinion.
So I actually prompting that to say, okay, you know, as an analytical thinker, give me your opinion as a creative thinker, give me your opinion as a leader. Give me your opinion as a customer. Give me your opinion as a whatever persona you might actually have it generate multiple perspectives for you that's actually better than what just a single other person might be able to do.
So you can augment your team of maybe four or five with a team of of ten or 20 through an AI taking on different personas. I never thought of it, but you're right, and I can be a multiple personalities. You can take on different personalities and. All at the same time too. So what's good about it is and you set something up where you start to say, For any prompt that I give you, I want you to respond.
As you know, this kind of array kaleidoscope of different voices and arrange it into a table or arrange an involvement. Yes, you can do that now. And so is that going to take place? That's right.
Yeah. That's our first big use case. Brainstorm. What? I never thought of it until we started talking. What a great idea. Right. I can use it to bounce ideas off of from different perspectives. So great brainstorming tool and I appreciate that.
It's a market research that branches out in synchrony development. Absolutely. Yeah. Anything you might want to ask multiple, you know, product market set like market validation, things like that. Imagine being able to use that as kind of an early stage of your creative development process, your product development process, where you're essentially asking it. You still might go and do this with humans later on, but you know, giving you a direction, give you additional questions, giving you additional feedback, that's a great place to start with something real.
I think it's funny that you've called us humans. You're already talking like any AI. Jeffrey I'm just I. Talk about humans a lot and I think it's a contrast to the technologies that we're talking about. But I really believe, like you have to keep the human at the core of the decisions that are making. I totally agree.
If you invert that and you put the technology at the core, I think you're going to end up in a place that you don't want to be. You're going to end up solving problems that you don't need to solve. Whereas if you keep the human at the center of that decision making, then ultimately, hopefully you're doing something that somebody actually is great.
All right. We got our first use case brainstorming. Great. One, great one. Next one, let's come up with another one. What about generating content specifically in my day to day work email, for example? Email's a great one. You know, being able to both write an email from scratch for you on a particular topic, being able to expand an email based on maybe a shortened prompt, being able to shorten an email if you're too verbose, being able to spell check or change the voice of an email.
I think anybody who is particular about their writing, you know, if somebody else goes writes for you, maybe you want to convert that over into your own, your own voice, just the way that you speak, the way that you write your own kind of written style, so you can feed in your own style into these tools and say, given this new piece of writing, convert that into being like my style. So I've already noticed a little bit of that with a tool called Grammarly. If people aren't using I've been using Grammarly for about five or six years as I've been working on my my PhD. It's been a life saver because I'm a horrible writer. Graham Grammar, all that stuff. Yeah, I'm maybe I'm still my freshman, an English teacher.
I got a D in that class and she said I would amount to nothing on my last paper and I sent her the first magazine cover that I got the cover on that my article. I sent that to her years later and said, Yeah, no, you're wrong. But she didn't know I had a tech writer helping me. Well, I don't need a tech writer anymore because I have a tool like Grammarly. That is, it learns my reading or my writing style.
It corrects my grammar for me, has the right voice, and they recently added a AI to it. But to augment or. Talk about the content of that. Right. So okay. Anything on how you set up your AI? The more technical writing might still require more human intervention afterwards. So we can't necessarily we've talked a little bit about this.
You know, are you necessarily going to trust the factual content? So it might be in the style of technical writing, but is it going to be accurate to the kind of situation that the technical writing is being written for? Maybe it will, maybe it won't. And so that's where I think it's really important, really, for the human to still intervene at some point, but to also come back and say, Did I get out of this tool what I wanted it to? Because what you don't want to doing is writing an email to somebody that mischaracterizes something that misrepresents something that maybe says that you can do something that you can't do. But it was part of the language model and that then became a component of that email.
So there's still, I think, an editing editorial role for the human to play it. I'm glad you brought that up because I think a lot of people are going to completely trust. Yeah, give it an eye and just send the email. Why? A lot of the tools are smart enough today to say, Are you sure you want to send this right? It doesn't say approve everything and send it says approve. You can read it now and then send.
I agree with you wholeheartedly. We do need to double check because we can't fully trust the content generated by this. Yeah. And then there's going to be other things, you know, as you think about kind of that day to day work. So whether it's scheduling people, you know, that's a day to day task that actually can be really hard because maybe you're aligning schedules. So doing something like that and I can actually do really well.
I saw examples where people were using this to do kind of master scheduling for courses for students in K-12, where, you know, the idea was, here's all of the students, here's all of the classes that they need to take now, arrange them into which class is going to be which period. Our problem. It's a hard problem for human to do, but for AI to take that data in aggregate and then to spit out that kind of overall schedule.
Great problem for it to solve. So logistics, I think, could be an area where people can look and start to say, you know, what? Is there either a more efficient way to do this? Is there a way that maybe it just gets me to my starting point again, 80% of the way there now I the human can go back and say, oh, we can't deliver from Peoria to Detroit. I'd rather have them stop over here first, because the timing, you know, there might be things that as a human, you know.
That we're like the driver has lunch with his wife. Exactly. Or Yeah. And it makes a happier employee. So there's context that the AI might not have, right? That as humans there's there's tacit knowledge, tribal knowledge that we have that isn't written down anywhere. That's right.
But that process, piece of it, you know, and there's another layer to process to where there's still a lot of organizations that are having people fill out forms. Maybe it's a paper form, maybe it's a an online form. Maybe it's just you filling out a PDF, something like that. But to extract the data out of PDFs is another area where you see AI doing really, really well.
And what it does when it does that, it's not doing one PDF, it's doing all of your PDFs. And so to take that and essentially build a knowledge base out of it, which you can then query, you can then ask it questions. So imagine being able to ask your documents questions and having the AI respond in a way that that document would respond. Oh, okay.
So so you brought up another, another interesting use case, which is it's a new especially large language models. It's a new interface. Absolutely. Right. Where it's more natural language for us to ask a question. Because before, if I wanted to query, let's say it did read, it did OCR and all my PDFs, it stored them all in databases, right? In non structured databases, unstructured databases.
For me to write a query to get information out requires me to understand the forms. And relationships of ideas also. Yeah, exactly. But now I can I can explore those forms in a more natural language way like saying, Hey, you know, can you give me an idea of all what's his medical clinic can give me an idea of all of the men that filled out the form over the last year that had skin cancer. Yeah. And that simple question like that where before it would take, you know, some considerable amount of time for data analysts to go and analyze all the data and set up all the queries.
And then a simple query of like select star where, you know. Yeah, so a natural for us. Right? So that's what's the name for that is really a conversational UI.
And so there's different types of user interfaces. This is a conversation with the rotational UI. There it is. Exactly.
And so you know, where that's coming into different industries is anywhere that you've needed to have a conversation, you know, think about tech support, think about customer service, think about sales, think about PCs, even making presentations like communicating an idea, doing explainer videos, like all of those things where it's required somebody to essentially translate an idea into vernacular. That's one area. Or to say, okay, I'm going to take your question, digest it with all of the knowledge that I have and then give you a product back or give you a, you know, address or. Maybe ask you more questions. Or maybe ask more questions. Right. Right. Because what we've had in the past is if I query a system and I get let's say I get 100,000 hits, but I was only really looking for maybe ten, then it may say, you know, I got a lot of information back and Key does does color matter to you? Does a color of hair the color of their eyes matter of a why are you asking? Well, because if you change it to color of eyes or Yeah, then I got a smaller dataset that might be more relevant to you.
That's right. To me, that's that's that unleashes more information in a much faster for me to now make decisions on. So I think that's where we're headed. And it's really important to understand another distinction of conversational UI as well, which is there's something happening behind the scenes when you use some of these large language models, which is called dialog tracking.
So if you were to write a query, you write your SQL query or some other query, each time you do it, you kind of still have to carry forward every other. Text or Yeah. Yeah, with dialog tracking instead of having to say, okay, now you know, can you give me all of the ones who have this and this and this and this? You can just say, Well, what about ones that also have this? And that also implies use everything that we've talked about, everything, learn and bring that to bear.
On my current question and query or whatever. And by doing that, that's the way that humans talk, right? We don't say, you know, we don't accumulate a conversation until we get to. Query resistance to be really long. And it's hard for us to keep all that in our brains. But yeah, we bounce back and forth. And so what the conversational UI does in the dialog state tracking does members, what happens in that back and forth so that you know, the answer at the end can include all those things you talked about.
Okay. Okay, great. All right. Let's, let's move. Let's, let's shift.
Let's shift over to a generation. And specifically, I'm going to talk about business users. Let's talk about generating the presentations because that's how we communicate.
Well, good or indifferent, whatever. But let's talk about that first and then we talk about new modalities of communication. We, a lot of us create PowerPoint presentations. Yeah, that's how we communicate in business, right? And they even teach it in school now, right? The kids give a PowerPoint presentation on, you know, dinosaurs or on the American Revolution, whatever the case may be. I started seeing a little bit of this where people are starting to use general generative AI to produce PowerPoint presentations in a very interactive way.
Where I. Had. Have you seen this as well? Yeah.
So you know, where I think this is going and what you've seen from some of the tools is to build a presentation that does a certain thing. So right now if I go to chat, I can say, you know, give me the bullets of a presentation about a particular topic and it'll go out and it'll pull all of that together. Well, it's not so unreasonable to, to take that and then say, okay, for each of those go out and get a relevant image and a little bit of text and put that onto the slides. So the automation of that kind of bringing everything together is really, I think, the kind of synthetic function of generative AI.
And by synthetic I mean synthesis. So what it's doing is it's taking a lot of disparate data and trying to bring it together in a way that then can communicate that idea effectively where, you know, imagine being able to take a team's channel or a Slack channel or a a thread of some sort and say, take everything in this thread and synthesize it into a few key points or a few key, maybe it slides or a few key, whatever medium that is, that's tremendous because now you've got legislators who can take a piece of legislation and synthesize it into a few key talking points about what it's actually about. Anybody who's ever seen legislation, you know, it's usually a stack, but technical documentation synthesize that technical documentation into a few key points, synthesize anything that requires tremendous cognitive overhead, really, for a human to go through and digest and not just read it, but read it and make sense of it. The AI, the generative, I can do that much, much more efficiently than we can. Okay, so this reminds me of the term software engineers. Yes, it took too long to read.
We didn't write. TLDR Well, yeah. Yeah. So. So I can use generative AI to take a really long email I got from my boss. Say I don't have 5 minutes to read this. Drop it in there. Give me the key point in my.
Suggested action items. Yeah. And then you can also say do this one. Not this one, do this one, not this one. And trigger another series of events in that in that chain.
Now, on the one hand, that's great, but the hope is that it's freeing you up to do other things with that extra time that you're getting. But I know that's a different conversation, right? No, no, no, no, no. You're right. In fact, I was talking to my wife the other day when when I was a CIO, I had two admins. I had a group admin for our IT department and I had a personal Adnan executive that been that was the most productive I've been in my whole life.
She was amazing. I didn't read my emails. I only read the emails that, that she you know, said I we need a decision from informational emails. She would at the beginning of the day and the end of the day.
I got a summary of of the key points that I needed. Right and and only important emails flowed through to me. My phone was monitored by by her and everything. It was, it was wonderful.
I could focus on getting stuff done instead of all the interruptive things that happened during the day. And I was like, Man, I wish I wish I had an executive assistant that was an AI that was in that. It's not there. It's not there yet because it can't learn from me yet.
What I think is important, where my my executive administrator every morning we met, she knew what was important for me that day, what was stressing me out. At the end of the day, we met as well and said, All right, this is what we got accomplished. And it was it was it was wonderful. I see what you're saying. You still. Had that.
It it makes me wonder how would I go about that. Right. And Stephen Wolfram has a really interesting article where he analyzed all of his email over the last 20 years or so.
It's a great, great. Would you. Say. In analysis and data visualization when he was emailing versus when he had his kids, you know, he was emailing later at night and earlier in the morning. And so really interesting kind of behavioral analysis of that. But what I, I could imagine is a world in which you're giving somebody access to your email archive, It's determining for you what does this person find important? What do they reply about what? What's their style of replying? When during the day do they reply to the point that you could start to build up some of the characteristics that you were saying about your ad? Been? Why couldn't you train an item that you absolutely could do that? Yeah, well, and then also I always look at my. Mm hmm. I want to get better.
I'm not great. I can always get better. Right? So wouldn't it be great if it could analyze and say, look, you're not great at responding to emails? You sure? A lot better job and give me and give me tips or you're not great at responding to to this or your answers are too short, which require multiple email exchanges. So what could you do too? There's so much room for improvement there. So that's what I'm looking for. That executive assistant I'm sure.
I'm sure some startups out there say, Yeah, we got it. And let me take. Yeah, your point about something I'm not good at and doing better at it and give you another use case, which is something that a lot of people are not good at is drawing. So a lot of people are terrible at drawing.
You see draw they draw, you know, whatever it is, I am, you know, a little thing like that where there's some really interesting developments is around taking people's hand-drawn doodles and converting that into a really slick, really well-produced graphic, which you could then use. Maybe it's in your PowerPoint, maybe it's it's a marketing image, maybe it's some other way so that that human piece is still starting things off, but you're then tapping into this kind of tremendous graphical resource that then brings you back something that's better than what you, you know, how long would it have taken you to build that in Photoshop plus Illustrator plus Unity, plus all these other tools? Yeah, so long time. I think in graphics there's certainly some interesting things happen happening. What Adobe's doing with Firefly is very interesting with Generative Fill.
You know, I've seen a lot of these cases where maybe people are storyboarding an idea, they're storyboarding a commercial, they're storyboarding, you know, maybe a presentation or something else could be a movie or something. And they kind of have an idea of the different beats that they want. But then they want to put an actual graphic to that. And maybe the generative is the one that wrote the story in the first place.
And you're saying, okay, great. For each of those beats, go out and generate an image. Well, now anybody can start to make a children's picture book or can write their own book or, you know, can. So so I have a I have a question around that then is everything going to become kind of whitewashed or I washed? I guess the right word would be I washed. We just coined word washed. Right. Anyway.
Of what I want. Yeah. Yeah. It's interesting with so I believe most things to be on a pendulum, you know I think there's a lot of course I think what you're seeing with the strikes that in Hollywood is definitely going to change the kind of level of acceptance of some of these tools in the entertainment industry. But from a writing standpoint, from a, you know, character generation standpoint, some of the things what most air doesn't do well right now and where I think people have to start demanding is the breadcrumb trail. So tell me how to read or that email or can I look under behind the scenes and start to understand why something was done the way that it was done? When that happens, then I actually think you're going to see a monetization of kind of human creative behavior. You see this with Grimes, actually, she's monetize her voice so that people can make songs with her voice and to split the revenue.
But what I think is going to happen, I don't think it's going to lead to I think really by I was saying maybe we made like the homogenization of some of this stuff, which is really does everything start to look the same, feel the same effects? Well, that's what I'm worried about, right? Because we've already seen I can see this happening totally happening. Right. I'm using AI to generate my emails. You're using AI to read my emails and respond back. So the eyes are now talking to each other and we're looking at it cursorily and going the apps and apps and and all of a sudden everything's everything's kind of washed out in the details.
But what does that. Say about the value of email? What does that say about the value of maybe the thing you're working on might not have been that important in the first place. Thank you for listening to Embracing Digital Transformation today. If you enjoyed our podcast, give it five stars on your favorite podcasting site or YouTube channel, you can find out more information about embracing digital transformation and embracing digital. Dawg Until next time, go out and do something wonderful.
2023-08-31