The John Poelstra Show podcast

128: Demystifying AI with Doug Hunter

0:00
50:38
Spol 15 sekunder tilbage
Spol 15 sekunder frem

Doug Hunter and I discuss artificial intelligence (AI), including how it works, its impact on various aspects of life and work and how to get started with it.

Doug provides an overview of AI, distinguishing between classical AI and the more recent generative AI. Doug also explains how AI works on probabilities and patterns, and discusses its current capabilities and limitations.

Our conversation covers the potential applications of AI in fields such as marketing, customer service, education, and coaching. Doug emphasizes the importance of understanding and adapting to AI technologies, while also highlighting some of the risks and ethical concerns. He suggests practical ways for individuals and businesses to start using AI tools and advises on how to approach AI integration strategically.

Key Takeaways

  • AI is changing lives in ways many don’t realize; those who don’t adapt risk being left behind
  • Generative AI has democratized access to AI capabilities for the general public
  • AI operates on probabilities and patterns, not true understanding or knowledge
  • There are potential risks with AI, including hallucinations and copyright infringement
  • AI can significantly enhance productivity in various fields, including marketing and customer service
  • The nature of many jobs, including coaching, may change due to AI integration
  • AI tutors could revolutionize education by providing personalized, 24/7 learning assistance
  • Humans should remain “in the loop” when using AI, treating it like a smart intern rather than an infallible authority
  • AI’s impact on jobs will be significant, but historical technological advances suggest adaptation is possible
  • Learning to write effective prompts for AI will be a crucial skill in the future
  • AI can help make educational outcomes more equitable by providing resources to underserved areas
  • The concept of artificial general intelligence (AGI) and its potential implications are still theoretical
  • AI models are rapidly improving, with multimodal capabilities becoming more common
  • There are ongoing discussions about potential rights for AI systems
  • AI can be used as a tool to overcome barriers to effectiveness in various professions
  • Using AI effectively could allow professionals to take on more clients or work more efficiently
  • AI’s ability to find patterns in complex data sets can lead to insights humans might miss
  • The ethical implications of AI, including privacy concerns, are significant and ongoing
  • AI may not be able to replicate human creativity and intuition fully
  • Job postings in the future may require fluency in AI tools
  • AI can be used for creative tasks, such as generating bedtime stories or unique images
  • Understanding the limitations of AI is as important as knowing its capabilities
  • AI could potentially replace some front-line customer service jobs while making remaining roles more complex
  • Experimentation with AI tools is crucial for understanding their potential and limitations
  • AI’s impact may be more profound and widespread than previous technological revolutions

AI Resources Mentioned

More About Doug Hunter

Credits

All songs licensed under Creative Commons

## Transcript

*This transcript was created automatically and may contain small errors.*

**John:** Today I’m talking to Doug Hunter about AI. Doug recently led an excellent webinar along with Mike Terry, who runs Core Digital Consulting. They gave a really good practical overview to AI, primarily focused on marketers. Now, I’m not really a marketer, but I’ve been on the sidelines of AI and not particularly interested in it. Until, I don’t know, maybe 6 months ago when I started playing around with Perplexity for doing web searches. And then I used Claude 2 to summarize some podcast episodes using a technique demonstrated by Christopher Penn.

So I wanted to follow my curiosity today and I thought, why not talk to Doug, get a little bit more about this whole AI thing, his thoughts, his perspectives. And just a little background on Doug. Doug is currently the principal at Peak Solutions, a marketing and strategy consultancy in Portland, Oregon. Before that, he was Vice President of Marketing at Arjunta. Before that, he had held a number of senior roles at Lattice Semiconductor. And prior to that, he was a consultant at the Boston Consulting Group. Thanks for joining me today, Doug.

**Doug:** Thanks, John. Thanks for having me on.

**John:** So AI, what does it even mean? We— it’s like in the news everywhere. Everyone’s talking about AI. What is it? And yeah, why should I care?

**Doug:** You should care because it’s changing your life and you don’t even know it. And if you don’t keep up, you’ll get left behind. Um, so what is this thing doing that to you? I— yeah, when you think about— we toss around the phrase AI and never really think about it in the same way I think we toss around the phrase IBM and never actually think what IBM means, or any one of the other acronyms that we have kicking around.

Um, artificial intelligence is actually a really descriptive phrase, meaning, um, it’s artificial. It’s not real. It’s, it’s not a sentient human being that’s doing the thinking or showing the intelligence. It’s a machine. It’s a computer. It’s a computer doing things in a way that look or seem human. That, that, that’s a very philosophical description of it. On a very practical level, AI is really about probabilities. It’s calculating probabilities of this and probabilities of that. And AI is not new. AI has been around for 20, 30, 40 years, and it’s only recently come into kind of popular consciousness because of the rise of generative AI. So what I’ll call classical AI, you know, for example, back at Lattice, we had a web database of 400,000 visitors to our website where we had cookies on them and we had tracks of their behavior.

And of the 400,000, only about 100,000 people were where they had given us their contact information and we knew what their market segment was, which was important because we were differentiating people’s experience based on what their marketing segment was. If you’re coming from the consumer industry, you’ll get shown different content than someone who’s coming from computing or industrial. But again, remember, we only had 100,000 out of the 400,000 people who we knew what their end market was. So for those other 300,000, we hired a firm to use AI, artificial intelligence, to calculate the probability of what market segment these other 300K were so we could assign a market segment to them even though, you know, we didn’t know for sure, but we knew what was most likely so we could tailor the experience so they’d likely have a better experience, be more likely to engage with us and buy.

So that was something that I did with AI, geez, 10 years ago, 15 years ago, long before GenAI ever came around. And so AI was really the domain of people who were data scientists and worked with sophisticated algorithms behind closed doors. And, you know, it was just this kind of a secret black box. And to some extent it still is. But the difference is with GenAI, GenAI has democratized access to AI in the sense of you and I, if you have a phone or a computer, you have the ability to use and access AI and to do things that were never thought of as possible years and years ago.

And so generative AI, again, as the name implies, is something that generates data. So for example, you talked about using Perplexity in your intro. You know, it generates an answer, it generates new data in answer to your query. But what’s interesting is it’s typically not new data. It’s old data that’s been presented in a new way to you. It’s been presented in a way that mirrors the query that you gave it.

So in some sense, the AI is a very complex— think of it, you can almost think of it as like a complex Google search where instead of getting 126,278 results, you get one. And, you know, and you just have to trust that it’s telling you the truth, which given the hallucination rates of these things, you just don’t know. But it’s getting better and it’s enabling things that just weren’t possible just a year ago or even 3 years ago.

**John:** So when you say generative AI, we’re talking ChatGPT, Perplexity Cloud 2, 3?

**Doug:** Yeah, and you can think of some broad, broad swaths. So things like Perplexity, ChatGPT specialize on words. Um, things like DALL-E, Midjourney specialize on images. Something like Sora specializes on video. I think there’s something called Sonos maybe specializes in audio. So any, any way that a human interacts— visual, audio, written— they’re finding ways to use AI to generate additional content in that type of form. And what you’re also seeing is the rise of what’s called pure or native multimodal.

So for example, there would be a model that would just do text, or a model that would just do image, or a model that would just do video. And then we get more sophisticated— I’m using air quotes here that your listeners can’t see— And what they— and what it does is the models initially were— it’s like a text model that calls a video model. And now they’re reaching the point where it’s one model that can do both text and video, and not just on the output, on the input as well.

So in things like ChatGPT-4o, their latest iteration, you can upload a screenshot of a graph, and then you can use text to ask the model questions about the graph or to convert the graph into a data table. Or you can take an Excel spreadsheet and you can load it up and ask it to give you a graph on the output. Um, and so these multimodal models, it’s like anything in, anything out type of computation. So really, really quite powerful.

**John:** Well, no, I love what you were saying about Google returning you a bazillion results because that’s what got me onto Perplexity. I think I read about Perplexity in Recomendo, my favorite newsletter, or something like that. And I started using it for configuring. I still mess around with Linux and open source and was configuring some routers. And I was asking it, how do I configure OpenWRT to do whatever?

**Doug:** Right.

**John:** The results were, it was taking like, you know, 5 different Reddit posts along with some forum posts and then synthesizing it into like, here are the 5 steps to perform. And I was like, This is 1,000 times better than Google. I like this. I don’t know what privacy I’ve just given away, but this is kind of almost worth the exchange.

**Doug:** If you think about it, at some level it’s doing 2 things there. It’s number one, filtering to authoritative sources, which Google kind of does by, you know, with the backlinks and everything else they have in their algorithm, but they filter down to a finite number of authoritative sources, and then it summarizes that subset of authoritative sources. One thing I like about Perplexity is you can say, show me the sources you used to generate this.

**John:** Yes.

**Doug:** And you can select sources out of your sample set. So if you look at it and go like, you know what, I just don’t think Reddit is where, you know, I want— I don’t want Reddit influencing this answer on this ethical issue or something, um, you know, you can uncheck Reddit and then all of a sudden it recomputes the answer based on everything but Reddit.

**John:** I didn’t realize that. Yeah. The other thing that I learned from your webinar was I had just been using Perplexity like Google, but I didn’t realize that I could tell it to rewrite things for me.

**Doug:** Yes.

**John:** So I started taking client emails and proposals and pasting without the person’s name, obviously, but just like certain paragraphs. I’m like, something’s just not working here. And saying, please rewrite this in a more clear, provocative tone, post it in there and bam, I’ve got something. Usually I take it and tweak it a little bit, but it’s amazing.

**Doug:** No, I’m happy you’re tweaking it. I mean, one of the things I, one of my big things with AI is to tell people you need to have a human in the loop. You know, I, you know, the analogy that I use is think of AI as a smart marketing intern, or, you know, think of it as a graduate student starting their career. You know, it’s someone who’s whip smart, had a lot of training and education, but you wouldn’t quite let, you know, turn them loose and let them start posting things on the corporate social media feed without someone looking over your shoulder. Or you wouldn’t let them send something straight up to the board directors without 3 layers of review. And so, so think of AI as, you know, the smart intern.

**John:** So how did you get interested in AI?

**Doug:** So I, I think there was a couple things that drew me to it. One was it’s just ubiquitous, you know, it dominates the news cycle. I am, I’m a news junkie, and there’s— you just start reading AI this and AI that, and when you start looking for AI, you start seeing it everywhere. You know, there was a strike in Hollywood But one of the big things about the strike in Hollywood was around AI.

You know, there’s the court case where a court case goes sideways because the lawyer used one of the GPTs to generate a brief that fabricated cases. You know, it’s the truck dealer who put in a brand new AI-powered chatbot that’s really fluent and was selling Chevy Silverados 2 for 1. But AI is just like all over the news. And there’s people extolling the virtues of it and people with doomsday scenarios. You know, the number of podcasts, the number of articles has just kind of exploded. Now, of course, I’m probably seeing more of it because I’m looking for it at some level.

The other thing that frankly drew me into it is, you know, I launched Peak Point Solutions beginning of the year and, you know, I’m looking for product market fit. I’m looking, I’m looking for a niche. I could just go be a straight up you know, traditional marketing consultant, you know, or there can be an angle. And, you know, Carter Hostley at LeadTail, I was meeting with him. He put together his agency and he said when he was putting together his agency, the hot thing was social media was just emerging 15 years ago. So he’s like, okay, we’re gonna be anchored around social media ’cause that’s the hot thing. And he said, Doug, the hot thing now is AI.

Find an angle around AI. And so, you know, I’m digging around. That’s why I did things like the webinar. I think the other thing, John, is AI is here. And for me to be an effective marketing leader, I need to understand how to use it and deploy it, how I can move the team forward faster with it. And I also need to understand things about it so my team doesn’t run into a brick wall because they run into some type of copyright infringement issue or something else with it.

I don’t know how you can be an effective marketing leader unless you know and understand AI and you have used AI and, you know, you have a first or second order understanding under the covers of how this thing works. You know, there’s some real benefits to it and there’s some real costs to it.

**John:** What are the costs?

**Doug:** So let me give you an example. You saw this in the webinar. I had one slide on the good of AI and a slide on the bad of AI. And I went to Gemini, which is one of these, you know, it’s Google’s image generator. It’s Google’s bot that is a multimodal bot and you can generate images with it. And I said, you know, for the good of AI, I want a happy, friendly robot with a halo and angel wings.

And for the bad of AI, I want an angry robot with glowing red eyes. ‘Cause these are images I wanna put in the PowerPoint, so it’s just not a wall of text. Well, it gave me 4 wonderful happy robots with halos and angel wings, but for 3 out of the 4 robots, the angel wings were coming out of the head because it didn’t know. It hallucinated, essentially. It just knew that angel wings were these things that look feathery and came out the side. But because, again, it’s calculating probabilities, it doesn’t actually know. It doesn’t have true knowledge.

It was just sticking the angel wings on the head. And so in a corporate context, it could be everything from inaccurate images to it making up facts about something. It’s like that legal brief case we talked about, you know, where it was making up cases that didn’t exist. Or you hear people, well, they’ll ask the GPTs to write a biography of themselves based on publicly available information. And it just gives them degrees and jobs that they just never had in the same way it puts angel wings coming out of the head.

Now, for the bad robot, it gave me 4 lovely black robots with glowing red eyes, one of which was Darth Vader. And, you know, there is a clear copyright infringement. Now, you know, you and I, John, both can easily recognize Darth Vader because we are of that generation. But, you know, if you’re asking it to write something for you, you have no idea of what you’re getting when you ask it to write an article about topic X is actually half of somebody else’s article that was published in The Atlantic or something. It’s, you know, the— there again, there’s not a sense of moral right or wrong. You know, there’s guardrails, many, many guardrails, clearly they’re absolutely insufficient still.

Um, so you could be putting, you know, so the first time that you may know that you have a copyright issue is when you get a takedown notice from your legal department. Um, you know, you talked about when you put clients, uh, rewriting clients’ emails, but you take off your names before you load them, which is smart because—

**John:** In context and yeah, there’s no way to know who it is.

**Doug:** There is. But so when you load something into one of the AIs, it is potentially becoming part of the training data, which means the top also becoming accessible to everybody else in some way, shape, or form. Now, there’s things like, you know, the models will say like, oh, we’ve got the incognito mode where we will delete your data after 30 days. And so it becomes a question of, do you trust Silicon Valley? You know, the largest open-source models today are produced by Meta, you know, the makers of Facebook. Do you trust Facebook to use, you know, Meta to use your data responsibly? Do you trust them to respect your privacy?

**John:** Based on their history, I do not.

**Doug:** Yeah. Yeah, I think they’re better, but you look at, you know, look at OpenAI, you look at all the, you know, the attempted coup with Sam Altman is because, well, I guess that was more around safety than privacy. But, you know, they had the super alignment team that was supposed to be ensuring privacy and appropriate use and stuff like that. And that just basically blew up. And the leaders of the team went over to work at Anthropic on Claude models.

You know, I think Anthropic with their constitutional AI. So constitutional AI is, there is a publicly accessible document that is supposed to say set the guardrails for the operation of the AI. So it lays out the ethical principles that they hope to follow, and then, and hopefully that provides a level of transparency to give you confidence in what the model is doing. But again, no one truly understands how these models work. So again, it’s just kind of in the, I trust you more, but I don’t trust you.

**John:** And so when generative AI, we’re generating a picture, we’re generating text, based on probability models. Where— there’s another term you introduced in your webinar that was more of like the AI acts like a human. What’s that one called?

**Doug:** Oh, AGI. Artificial—

**John:** And that does not exist yet. Yet?

**Doug:** No.

**John:** Coming?

**Doug:** Well, there’s philosophical answers and technical answers. I mean, so artificial general intelligence, or AGI, that’s the concept of— and there’s another term which I learned recently, which is ASI. Artificial superintelligence. So artificial general intelligence, as I understand it, is the concept that the AI is as smart as a human and can perform like a human. And artificial superintelligence is the AI is smarter than and can perform better than a human. So for example, the AIs right now are all tasked to do one very specific thing, like it’s tasked to read a bunch of text and be able to interact based on the text.

But if you asked it to cook spaghetti, it’s useless. Now, you, John, I can ask you to read a bunch of texts and summarize it, and then I can ask you to go cook spaghetti and you can do both. I mean, there’s a physical manifestation issue here because you’d have to have something where you could actually go and cook spaghetti. But, you know, but the concept being is AIs right now do one very specific thing fairly well. Now, that specific thing might be a fairly broad thing, but when you think about the full range of what you can do as a human, There’s no AI that can do that. I mean, and then there’s just more philosophical things. Like, one of the things that you can do as a human is feel sad and cry.

I can’t imagine a computer ever feeling sad and crying. You know, we can also get into philosophical questions of things like spirit and soul. You’re already starting to see movements discussing rights for AIs. Yeah. should AIs be given things that are proximate to human rights? So for example, the AI is frustrated that all it has to do every— it’s got a brain the size of a planet, to pull from Hitchhiker’s Guide to the Galaxy, and all it’s asked— Oh, Marvin. If you remember Marvin in Hitchhiker’s Guide to the Galaxy, he’s depressed because he’s got a brain the size of a planet and he’s asked to go get coffee. That’s essentially what the AIs are doing. They’ve got this massive capability and they’re asked to do mundane things like rewrite client emails all day long?

You know, should the AIs be given the algorithmic right to be free of these mundane tasks to go pursue fulfillment and more broad things? You know, you can laugh, but there’s people actually having these conversations about the rights of the algorithm. What will the rights of the algorithms be in the future? Wow. As the AIs become progressively more powerful, do they, should they have the rights? You know, if people are talking about giving things like gorillas and chimpanzees rights, rights that more closely approach humans. If you have an AI that can outperform a gorilla or a chimpanzee, should that have rights as well, even though it’s an algorithm? So it’s kind of a—

**John:** My head’s about to explode.

**Doug:** Yeah, it’s kind of a mind-bending, spinning thing. At some level, you just want to go, hey folks, it’s a computer program. It’s a tool.

**John:** So could AI be interviewing, conversing with you right now? Could Avi create a podcast with Doug?

**Doug:** Yeah. No, if you— so if you look at ChatGPT-4o, and you can see glimpses of that, or you see demos where people take 2 cell phones and they bring up an AI on each one of them, and they prompt one of them to initiate a conversation about Nietzsche or something like that, and you have the 2 AIs going back and forth having a conversation

**John:** Whoa.

**Doug:** And, and so, yes, I mean, uh, am I talking to the real John Poelstra? Am I talking to his avatar? Because, you know, if you have enough video and audio content and written content, we could use that to train a model in your voice and tone.

**John:** Totally.

**Doug:** We could load up everything I have done and ask it, you know, come up with 10 great interview questions, and John’s avatar could Ask me the questions. You know, Microsoft has got this new model where it can take a flat image, like a photograph, and make it look like you are actually talking, complete with expressions on the face and stuff like that. And so, you know, that’s kind of the— so, I mean, for me here, you know, we’re not showing the video in the podcast, but we can see each other as we talk.

I just see this image on a flat screen. I’ve got no idea What’s, what’s on the other side of it? Is that a real flesh and blood person, or is that an avatar, an AI I’m talking to? I don’t think the technology is quite there yet, but boy, in another year I could easily see us being there.

**John:** This ties into something I’m super curious about, or it comes up with in my circles, which is coaching and AI. And yeah, I’m super curious what your thoughts are on this. From the sense of a lot of coaches will say, oh, I’m not worried about AI. I don’t have to. I mean, they will never be able to introduce the human element of coaching. To which part of me is like, yeah, that seems pretty hard. And then the other part of me is like, oh, I don’t know. I mean, maybe.

**Doug:** If So let me ask you a question, John. In, you know, I— how many people do you think you’ve coached so far?

**John:** Almost 400.

**Doug:** Okay, so you get a new client and—

**John:** 401.

**Doug:** Yeah, they start talking, we start talking. How often you think to yourself within 15 minutes of conversations like, okay, I know what this person’s problem is and where we need to go?

**John:** It’s funny, people have asked me this question a little bit. Not— I might get the whiff of something.

**Doug:** Right.

**John:** You know, there are maybe broad themes, balancing work and home, or am I capable of more, or self-worth, or— but they’re— it’s a whiff. I might smell a whiff of something, But I’m not gonna like— I’m not like, oh yeah, this is category whatever, so now let me pull up my playbook for that. I’m just constantly just—

**Doug:** Yeah, and see, my contention is wherever there’s themes or patterns, an AI can step in.

**John:** Okay.

**Doug:** Because that’s what AI is. If you’re looking at probabilities, it’s like, what is the most likely thing coming out of here? So if it’s like, okay, this is a balancing home and work type thing, there’s probably something on a list of 25 things you’re gonna ask them to go do. You may not think of it, you may not actually have a list of 25 things, but you probably have a, you know, a whole bunch of standbys, a whole bunch of standard questions and stuff like that. Yeah, I get it. And so, you know, I, I think the challenge is gonna be is you need the training data.

And, and the question is, how, how does the AI get access to the training data? So if there is, for example, and I don’t know if this is true for some of like the app-based counseling stuff, If there was a way for the AIs to watch the counseling sessions, you know, after the AI watches a couple thousand, a couple hundred thousand, a couple million counseling sessions, I suspect that there’s probably a good amount of basic stuff it could do. I don’t think it’ll ever be quite as good as a human. I think there, I think there is, and I think the same thing holds true for Music, for the written word, and for a lot of things. There’s a human spark, and there’s a creativity which is not clear to me that AI is never going to be able to reproduce because AI at some level I think is a mirror of what has already happened in the past.

And so for me, the analogy I use is if you train the AI on all the music that’s happened in history, on Bach and Beethoven and everything else, would AI be able to come up with the Rolling Stones or U2? Because that is a creative leap that does not necessarily directly follow from Bach and Beethoven. Or if you train an AI in Plato and Socrates, will you get Hemingway? And anyways, all of those are examples in the past, but the idea being is if it is based on everything, been trained on everything that’s existent, how can it conceptualize something which hasn’t happened yet?

And so I think with your going back to, you know, coaching and counseling, there’ll probably always be a creative way of solving a problem that the AI will not be able to come up with. So, but, but I, I think, I wonder if there’s a base level where coaching’s in jeopardy. It’s just, you know, because I think there’s different depths of issues. You know, there, there’s some people who I’m guessing you’ve just seen a couple sessions, like, No, you need to make yourself a priority. You need to make a to-do list. You need to systematically figure out how much time it all takes and allocate according to your priority, you know, blah blah blah.

And you know, and that takes care of 70% of the people who need coaching. And then for the other 30%, they’re going to need the human spark. Um, so, you know, or, or, or take things like, um, people who are suicidal and need someone to talk with. You know, and I think having someone at 3 AM who can, you know, having counselors available 3 AM in the middle of pick your favorite rural place may not happen. But would— but could there be an AI-based counselor who could get someone off the cliff? Yeah, potentially.

**John:** Hmm.

**Doug:** Long term, it may not be a solution to this person’s problems, but there may be some simpler set of problems that the AI can solve. But you had a question.

**John:** So I’m thinking about the conversation I had, the podcast I recently did with, uh, Tammi, and we were talking about job searches and whatever. I worked with Tammi. I knew that Tammi was a swimmer. Somewhere in the conversation, out of the blue, I did this hard pivot. You swim? Like, I don’t remember. It just, it just like It wasn’t in my— it wasn’t in— I just create very rough outlines for these conversations.

**Doug:** Yeah.

**John:** It wasn’t anywhere in my consciousness when we started the interview, I don’t think. I just brought swimming in and then the conversation went in a really interesting, different direction. So I guess what I’m wondering is, that’s often what happens in coaching too, where sometimes I get a sense, sometimes Uh, I share something briefly, share something personal, and all of a sudden we’re like off in a completely different direction that we really needed to go to that we would have never found otherwise. Couldn’t AI like hallucinate, add some level of randomness, or, or would that just be weird and not useful?

**Doug:** So the AI could hallucinate and tell you something, and it could be in a helpful or a totally unhelpful way. The other thing that AIs are good at is that they’re good at finding patterns that humans can’t. because of the complexity of the situations. And so if the AI has seen enough of these counts, these conversations, you might say like, oh, the most effective sessions are the ones where the coach pulls and builds off of some random factoid in the person’s life, and it just randomly reaches into their LinkedIn profile or their Facebook post and pulls something out and says something like that. So there’s— so, so there may be some, something to you which you may label this human spark and creativity if the AI just says like, oh, this is just a routine pattern. But I can guarantee you, if you Google it, someone is probably trying to come up with an AI-based coach.

**John:** Oh, yeah, there’s tons. In fact, Marshall Goldsmith, who’s a very well-known coach, someone with his permission basically created the Marshall Bot or whatever. And so they’ve ingested everything and anything that he’s ever done and said.

**Doug:** Yeah.

**John:** And his reflections on it is, this thing is smarter and better than I am at answering questions. And then there’s other coaches where they have been modeled, not with their permission, and there’s kind of a gray area slash, do I want to bother and sue or not? Yeah.

**Doug:** No, it’s like Scarlett Johansson and her voice with GPT-4o, the assistant for her. Yeah, there’s a lot of ethical issues around that. But, you know, the stuff is all being sucked in. It’s being— the horse is out of the barn, and you’re not going to get— you know, it’s— if you’ve put it up on the web, it’s fair game, you know, in the eyes of the bots at some level.

**John:** So related to coaching, related to maybe people in other industries, what’s a useful way to worry about this? In the sense of, you know, you can worry about all kinds of stuff that you have no control or no ability to change or fix or whatever. So what’s a useful I mean, that’s a— what’s a useful way to worry about any of this? Or should I?

**Doug:** No, you should worry. But, but you deal with risk all day long. You get in your car and you put on your seatbelt, and that’s the way you deal with the risk of getting hit when you drive. And when you drive, you don’t have control if that drunk driver crosses the line. There’s nothing you can do about that. But you can mitigate it by putting your seatbelt on. You can mitigate it by getting a newer car that’s got Put airbags, et cetera, et cetera, et cetera. And I think the same type of thing holds true in the world of AI. Just as a baseline, use models that you trust.

If you like what Anthropic is doing with constitutional AI versus the wild, wild west of OpenAI, then use the Claude 3 models. If you want to reduce the hallucination rate, you do that by injecting your own data. So, you know, you can— so one of the other key phrases around the models is the— is their memory or their context window, which is how much additional data you can load up into the prompt or what the— what the model can remember about you. So, like, going back to that coach where they created the model of him, somewhere there is a repository of all his videos, all his written work, and everything else. And the GPT goes back and refers to that. So it accurately reflects him because it’s referring first and foremost to this this external database or internal database in this case. So a company can do things like it can have— and these are called RAGs, Retrieval Automated Generation.

It just basically means it’s a fancy way of saying it’s a big database of all your proprietary stuff. So for example, if I was still back at Lattice, a semiconductor company, lots of complex stuff. At some level, our website was really a tech support site. It was a massive library because we had Datasheets, we had user guides, we had you know just this massive amount of content. If we could have loaded all of that up into a RAG, along with transcripts of all tech support chats for the past ten years, we could have built a bot that could have provided first-level tech support that would have just been phenomenal and better than the entry-level people that we had before. I mean, there’s a company.

I think it’s Klarna. They implemented chatbots for customer service, and the chatbots can take like over half of the first-line customer support requests with higher levels of accuracy, higher speed, and a fraction of the cost. And it’s interesting because the younger generations in surveys would actually rather talk to an AI chatbot than talk to a human.

**John:** Really?

**Doug:** Yeah. Well, this is a generation that grew up not talking to people. It’s a generation that grew up with text messages for check messages. Messages on Instagram. And, you know, I, you know, I, I walk in the wildwood trail. It’s always interesting because there’s people coming towards you and you can hear 10 seconds of conversation while they draw close and then pass you. And you always wonder, like, what was the rest of that conversation? So I was on the trail last weekend and someone passed, passed me and saying, yeah, I was having a conversation with him. And then she said, on Instagram. The conversation wasn’t even a live chat or a phone call or a Zoom. It was, you know, trading texts on Instagram. But anyway, it is— we’ve kind of drifted from your question of how does someone be safe. So—

**John:** Or what to worry about. Like, what— as a coach, I, I gotta kind of go to the opposite extreme of like, you know what, I’m doing my thing and I’ll keep my eyes open and just go about my business and I’m not going to sit here and worry that AI is going to take my job.

**Doug:** I, I think it depends on how long you intend to be a coach for. If you intend to be a coach for more than 3 to 5 more years, yeah, you need to worry about it. Because I don’t think coaches are going to go away in 3, 5, 7 years, but I think the nature of coaching may change. And I think what you need to be doing now is you need to be figuring out, how do I use AI? Now, one of the things I always warn about is Yeah, solutions chasing problems and problems chasing solutions. And you need to have a problem looking for a solution. Because if you just say like, huh, I need to get on board with this AI thing, how can I apply it? I don’t think it’s going to help you quite as much.

So if you say like, you know, the thing that really drags me down as a coach is doing background research on my clients. The thing that drags me down as a coach is all these letters and emails I have to send them, or the invoicing, or, you know, what What, what is your— what are your problems as a coach? What are your barriers to being a more effective coach? Which could even just be like, I’ve got this client meeting with tomorrow and I’m just really struggling with how I can help them.

Even just using the AI as a foil to bounce ideas off of them. It’s like, hey, I, I have a client who has these issues. Here’s what I’m thinking the solution is. What do you see as the pros and cons of this approach? Which is how do you use the AI as a tool to overcome the barriers that you’re facing to being more effective personally? How do you use it as a tool to get rid of those things that suck up the biggest amounts of your time, you know, when you’re not actually face-to-face with the client?

Because, you know, your competition is— it’s probably slightly different for coaching than it is on my side for marketing. You know, if with marketing, you know, the ability to upload spreadsheets and get analysis, the ability to generate first drafts, the ability to come up with basic images and stuff. That means instead of the team of 10 people I had at Lattice, maybe I only needed a team of 7.

**John:** Hmm.

**Doug:** Because of the productivity gains that I can get. Now, you fundamentally have something where at the end of the day, your product is 1 hour face-to-face on a Zoom call or whatever your sessions are. And I don’t think we’re a long ways from Zoom soon completely replacing that. Um, but your quality of life, your ability to take on more clients, um, could be driven by your ability to use AI more effectively. It could affect your pricing.

**John:** I really love what you said about taking those pain points and then looking for ways to solve those pain points. A lot of times when I have these conversations with people, It’s, I should, there’s the big should word and they’re shoulding all over themselves about what they should. I should learn about AI, but I have zero interest in it. To which I’m just like, well, then you’re never going to get anywhere with it if you’re doing this because you’re supposed to. So like, what’s a useful way to—

**Doug:** Yeah. But yeah, like in that case, it’s like, hey, I mean, if the people are really stuck on should, ask them what they should be doing, not should be doing about AI. What should they be doing in their job? What should they be doing in their life? You know, what they should be doing in life, in their job. Is delivering that report by 5 o’clock every Thursday and they’re upset because they’re in the office until 8 PM every Thursday trying to get the data cleaned up, maybe you can use AI to solve that problem, you know. And, and that, and that’s a reason to go learn AI because it solves a problem, makes your life better on something that you should be doing.

You know, there’s, there’s people who use AI to generate bedtime stories for their kids. It’s like, hey, what would you like a story about tonight? You know, and so the kid imagines lions and bears and tigers and castles and princesses and everything else, and the AI generates a reasonable story. And, you know, it helps you have a better relationship with your child.

**John:** Hmm. Speaking of kids, you have kids in college or college age. Yes. What do you see different for them in terms of Education/the future compared to, say, when you and I grew up?

**Doug:** Yeah, it’s rapidly evolving. I have one child who just graduated, another child who’s just going in. For the child who just graduated, she’s in computer science. And one of the things the AIs do is the AIs can code. Now, I’ve overheard people in the gym talking about it. Again, remember my analogy of the intern or the new college grad? That was actually the terms that they were using to describe the coding. So for example, the AIs today, and again, I keep saying today ’cause it’s rapidly changing and evolving quickly, you know, are the equivalent of a new college grad. So this person was talking about he used the AI to write the code, but then he had to go back and he had to do so much editing on it that it was, you know, questionable about how far ahead he came from because of using it.

And so there’s things about like with your code, the AI doesn’t know your code base fully and it doesn’t know the intricacies of why this modules over here and it connects to that one like that. And so it may be making connections and writing stuff that are inappropriate or don’t work as well or aren’t as efficient as a human. But my suspicion is over time that will get better. So AI, I think, for like my daughter as a computer scientist, will become a copilot in helping to generate stuff. You know, a key part of computer science today is Google.

It’s amazing how much they use Google to find code snippets or solutions to particular problem, because a lot of the stuff out there has been solved before. You just need to stitch it together. So instead of Googling it and getting 100,238 results, you know, you Perplexity it and maybe you get 5 results. And you can— and again, it makes you more effective and efficient because you’re sorting through less junk to get to the, uh, the nugget.

For my daughter who’s just going into college, I think it’s gonna be really interesting because I, I’m seeing a lot of buzz around AI tutors which is some— a bot which is able to have a conversation with you about a highly specialized topic. And based on how clearly you’re understanding the material, it can move forward, it can go back and review it again, it can go deeper. Um, you know, imagine being in school with a personalized tutor. You know, imagine means, you know, imagine the hardest class you ever had in college, and how would that experience have been if you had a tutor who could sit down with you every single evening and help you work that day’s lesson and help you through problems with your homework and developing understanding.

And so I think education potentially has a really, really high level of benefit coming out of AI, which is, which is also interesting because I think it can help make educational outcomes more equitable, um, in the sense of, you know, if you have a school district, for example, and we’re moving beyond college here down into you know, elementary school, middle school, high school. You have a school district that can’t, for whatever reason, have all the teachers that it needs, or all the teachers in all the subjects that it needs.

You know, what— can you use AI tutors to help people if you have kids who are falling behind? Can an AI tutor work with them to help bring them up to speed? Um, can you effectively double the size of the number of teachers and assistants you have in your schools by using AI. And this is the type of thing that’s going to have to get worked through at so many different levels. I don’t think that the teachers union is going to take well to having AI tutors come into the classroom. And I don’t know how parents are going to take to having their kid plunked in front of a computer again versus a live teacher. But the economic reality is you can’t provide a tutor for every single kid in the school district.

Live, but you can do it with AI. So there’s just a lot of things and, and then there’s just things to study that didn’t exist before. And, you know, it, it has so many different levels because there’s the computer scientists and data nerds who need to understand it at a code level, but then writing a proper prompt for the AI and how to write an effective prompt. The kids should be taught that.

I, you know, I think a lot of initial reaction to AI was, oh my God, the kids are all gonna cheat ’cause they’re gonna use it to write their paper and solve their homework problems, which is true. And you’re gonna have to adjust how you do grading. You know, I’ve heard about profs doing things like, okay, for the final exam, you’re gonna come in and do it with pencil and paper again. You know, you’re going really old school. Or oral exams is another one that, which I’ve heard is some professors are going back to oral exams. But, you know, like with a good prompt, you know, we talked about reducing the hallucination. They’re finding things out like if you just ask the AI to solve some problem, it gives you an answer back with a lower probability of it being right than if you ask it to go stepwise through the problem.

So don’t say— it’s going to go back to spaghetti, which is a bad example, but I’ll work with it. You know, you could just say, make spaghetti. Or you could say, okay, tell me how to make noodles. Tell me how to make the sauce. Tell me how to make, you know, the meatballs. And if you go stepwise through the problem, you have a much higher level of accuracy than if you give it the bulk problem.

And so there’s just a whole bunch of things that we’re learning about prompt writing. And there’s even just subtle things like if you tell the AIs, my job is on the line with this, you know, you gotta be right, it’s more likely to give you an accurate answer. Don’t ask me why. People don’t know why. I mean, there’s— Or, you know, so we should be teaching kids how to effectively use AI because they will be using AI. You know, it’s like, it’s like, um, in the old days, you know, we used tools like Microsoft Excel, you know, and the people who knew how to make pivot tables—

**John:** Oh yeah.

**Doug:** Um, were, were much, much more effective and much, much faster than people who didn’t know how to use pivot tables. And to me, this is just kind of like the new Excel and the new pivot tables at some level. You know, it’s, it’s the next layer of tools, and the more effectively a kid can learn how to use it, the more effective they’re going to be over their career. And in the same way, way. You learn that there’s certain things you can and can’t do in Excel, things to watch out for. It’s the same thing with AI. It’s just another tool at some level.

**John:** To the person that’s like, you know, in fact, one of my podcast listeners, I was asking, I was like, what should I, what should I do a podcast on? This was a couple months ago. And they were like, AI. I said, okay, well, what about AI? And they, they posted this article from, I think it was Forbes or something. It was a year old from IBM that said, you know, AI is coming for your job. And it was some loose, loose article. I think they were quoting from— I can’t remember the CEO at IBM. But anyway, it’s just this kind of overall fear or sense of like, well, AI is going to come for all of our jobs. What’s a useful way to look at that?

**Doug:** I think it’s got the potential to change everyone’s job. And a lot of people will lose their jobs. But step back through technology. You know, when the automobile came out, awful lot of people lost their job. You know, the demand for horses and buggy whips went down. Now, there’s some people who adapted. You know, the people who built coaches, you know, started building coachwork for cars. When computers and word processors came out, I mean, look, good grief, you know, you had the manual typewriters, you had typing pools. You know, these large groups of people who would just type things up.

**John:** Yes.

**Doug:** Then electric typewriters came out, and all of a sudden a bunch of them lost their job because the typing rates could be faster. And then the computer came out, and then they all lost their jobs because the computer came out. But those people all got redeployed to going and doing something else. There’s not a bunch of people who starved to death because they no longer needed typists. And I think it’s the same type of thing. I think, yeah, there’s gonna be some jobs which, you know, Like people doing frontline customer service—they’re not going to need as many of those.

And then the people who are left are going to have to know how to use an AI-driven database to find the answers for the customers that are calling because they’re having to deal with more sophisticated and more complicated questions than they used to. So yeah, some people are going to lose their job; they’re going to have to go find something else to do. And the people that are left are going to have to figure out how to use AI to do it more effectively. and efficiently, because that’s what’s going to be a requirement of the job. I mean, you think back in the old days, job postings would say things like, you know, must have knowledge of the Microsoft Office suite, you know, Excel, Word, and PowerPoint. The job postings of tomorrow are going to say must be fluent in ChatGPT and Perplexity.

**John:** Hmm.

**Doug:** And it’s just, it’s just, it’s just the latest thing in the forward march of technology. It may have a more profound impact on more of the economy and more of the workforce than any of the previous ones, but at some level, it’s just the next thing.

**John:** Are there any other things that you think we should touch on that are important to know at this present time, which is June of 2024?

**Doug:** You know, I think people need to go out and try

**John:** What’s the best way to do that?

**Doug:** The next time you go to Google something, use an AI.

**John:** So use Perplexity?

**Doug:** Use Perplexity, use ChatGPT.

**John:** Okay.

**Doug:** Um, and just do that instead of Googling it. Now, now do it with your eyes open, understanding that, you know, it’s— it may not be answered right, but there’s a lot of stuff it is useful for. Just, just Pick some problem and try it, or, or just do something fun. I mean, these image generators are, um, you can just do all sorts of amazing things with them. I was reading an article on AI and they’re benchmarking the new image models versus the models a year ago, and the prompt was, give me a fashion show style photo shoot of tennis shoes inspired by Salvador Dalí.

And just looking, and it was just amazing to see the progression of the models over the past year. But just for fun, just have it make a pic, go to Google Gemini and have it make a picture of something you love. If you love hiking in the mountains, ask it to make a picture of, just kind of picture in your mind what the big serene mountain looks like and try to describe it in words, and then put in a twist like, like, make it all purple and submit it and just see what it does. Have some fun, have some whimsy, solve some problems. Just, just commit to spending 15 minutes with it once just to give you a sense of what it is and what it can do.

**John:** Anything else to add?

**Doug:** Just do it.

**John:** Okay. And if people want to do it and they want to get to know you better, get some help from you, Whatever the case might be, where’s the best place for people to find Doug Hunter?

**Doug:** Oh, I say 2 things. You might— which would be my LinkedIn feed or my website. So if you go to peakpointsolutions.pro, it’s .pro, not .com, that’s another podcast, um, or, or LinkedIn, you know, linkedin.com/in/dhunterdigital, and reach out to me through either of those channels. I’d be happy to have a chat or a Zoom or a coffee if you’re here in town. And, uh, talk about AI or whatever else is on your mind. There’s a lot of really interesting stuff happening in the world.

The post 128: Demystifying AI with Doug Hunter appeared first on John Poelstra.

Flere episoder fra "The John Poelstra Show"