Latest / The Adults In The AI Room / The Earley AI Podcast with Seth Earley - Episode #52- Human Touch in AI: Nick Usborne on Emotional Intelligence in Marketing

Transcript

The Earley AI Podcast with Seth Earley - Episode #52- Human Touch in AI: Nick Usborne on Emotional Intelligence in Marketing · The Adults In The AI Room

• 0:00
  SPEAKER_04: Well, uh, welcome to the show.
• 0:02
  SPEAKER_04: Good afternoon, good evening.
• 0:03
  SPEAKER_04: My name is Seth Early, and I'm Chris Featherstone.
• 0:06
  SPEAKER_02: It's good to be with you, Seth.
• 0:07
  SPEAKER_04: And uh we're here on the Early AI podcast.
• 0:10
  SPEAKER_04: And uh we're really excited to introduce our guests for today.
• 0:13
  SPEAKER_04: We're going to be talking about some of the misconceptions around generative AI in the context of creativity, in the context of copywriting, and how we can use generative AI most effectively.
• 0:25
  SPEAKER_04: Uh, we're going to look at context-rich prompting.
• 0:29
  SPEAKER_04: Of course, we had uh Nick on uh one of our webinars, but we're gonna have more of a conversation about prompting.
• 0:35
  SPEAKER_04: Uh, I'm sorry, I gave gave away who your uh who our guest was before giving it away.
• 0:41
  SPEAKER_04: But the importance of emotional intelligence when we're doing prompts so that we can be more engaging and be more genuine.
• 0:48
  SPEAKER_04: So our guest uh today is an industry veteran with over four decades of experience in copywriting.
• 0:55
  SPEAKER_04: And he's a very passionate advocate for the use of generative AI in a much more nuanced way than many people are doing this.
• 1:03
  SPEAKER_04: Originally from the UK, Nick has uh been at the forefront of internet writing since the 90s, now focuses on educating others on how to effectively integrate AI with human-centric communication.
• 1:16
  SPEAKER_04: Welcome to the show, Nick Usborn.
• 1:19
  SPEAKER_04: Thank you very much.
• 1:20
  SPEAKER_01: Did I say your name correctly?
• 1:21
  SPEAKER_01: Did I say almost, almost, Osborne.
• 1:23
  SPEAKER_04: Osborne.
• 1:24
  SPEAKER_04: You know, I usually check that ahead of time.
• 1:26
  SPEAKER_04: My focus, my bad.
• 1:27
  SPEAKER_04: All right.
• 1:28
  SPEAKER_04: No worries.
• 1:29
  SPEAKER_04: So a lot of times we uh we get started by talking about the things that people don't quite understand about uh the topic about AI, about generative AI, and what are the most common misconceptions, especially when it comes to you know, thinking about how AI can handle tasks.
• 1:46
  SPEAKER_04: People are talking about, you know, artificial general intelligence.
• 1:49
  SPEAKER_04: I think we're a little bit away from that.
• 1:52
  SPEAKER_04: But you know, what are your thoughts about this and give us your kind of summary of what people miss about prompt development design and engineering and really what those misconceptions are in this phase?
• 2:05
  SPEAKER_01: You you thought you'd start with an easy question then, right?
• 2:07
  SPEAKER_01: Yes, exactly.
• 2:08
  SPEAKER_01: I thought I'd go with a nice softball and easier way into it.
• 2:12
  SPEAKER_04: I know there's about five questions in there.
• 2:14
  SPEAKER_01: I think a lot of people, if you look at some of the data, it is like everyone's tried Chat GPT.
• 2:20
  SPEAKER_01: Everyone's tried it.
• 2:21
  SPEAKER_01: And and I think the majority of those people have tried it like once or twice for fun, curiosity, like what is this thing?
• 2:28
  SPEAKER_01: And and I think a lot of people start off treating it as like another software tool in a sense.
• 2:33
  SPEAKER_01: Um, and and they go in and they just ask a simple question, like, you know, write me a blog post on the general aviation industry.
• 2:41
  SPEAKER_01: And then they look at the output, and again, not so good, not very good, very disappointing.
• 2:45
  SPEAKER_01: What's all the hype?
• 2:45
  SPEAKER_01: What's all the hype about?
• 2:47
  SPEAKER_01: So they kind of throw in just a question and expect like a miracle to come back, and they're disappointed, so then they don't come back again and they tell their colleagues and friends and family that it's all a big, a lot of hype.
• 2:58
  SPEAKER_01: So where I come from is is to is trying to get people to be to change their viewpoint.
• 3:05
  SPEAKER_01: This isn't like a software tool, this isn't Google search, this isn't Microsoft Word.
• 3:09
  SPEAKER_01: You don't just plug in an instruction and get out a miraculous output.
• 3:13
  SPEAKER_01: You have to work with it.
• 3:15
  SPEAKER_01: So you've heard me talk before about the kind of collaborative approach to actually go back and forth.
• 3:20
  SPEAKER_01: And part of that is a mindset thing, I think, of not treating this as a software tool, but almost treating this as a co-intelligence, as a partner, as an ideation partner or research partner.
• 3:31
  SPEAKER_01: So I think when you when you reset your expectations, like like this isn't like old tools like search.
• 3:36
  SPEAKER_01: This is something different.
• 3:37
  SPEAKER_01: And you have to, like with any of these tools, you you have to learn, you have to learn some fundamentals of how to use the tool to get the best output.
• 3:45
  SPEAKER_01: So that's my kind of top-level thing, is when I've talked to people, I ask them what they've done so far, what they've tried so far.
• 3:52
  SPEAKER_01: And usually when the output is disappointing, it's because the input from the user has been disappointing in a sense to the AI.
• 4:01
  SPEAKER_01: Like, come on, human, you can do better than that.
• 4:03
  SPEAKER_01: So, yeah, I guess that's my that's my starting point.
• 4:06
  SPEAKER_02: I guess they don't really understand the inputs to your point that you want a really good outcome.
• 4:13
  SPEAKER_02: They use really limited inputs.
• 4:15
  SPEAKER_02: And you don't want to say that they're lame or that they're, you know, you want to say that but you just use the limited amount of information and context in order for you to drive, you know, this grand outcome that you thought was going to happen, however, it didn't.
• 4:28
  SPEAKER_02: That's disappointing to you.
• 4:29
  SPEAKER_02: But then you when you go back, you know, and look into the question, your question is too vague.
• 4:33
  SPEAKER_02: Right, right.
• 4:34
  SPEAKER_01: Or or or otherwise, it's had ambiguity in there.
• 4:37
  SPEAKER_03: Right.
• 4:37
  SPEAKER_01: Um that that's happened to me.
• 4:39
  SPEAKER_01: Or I remember once I was writing something and I wanted some quotes from established experts to start each chapter or section of this document.
• 4:48
  SPEAKER_01: So I asked ChatGPT, I said, hey, write me some quotes from recognized experts for these each heading section.
• 4:55
  SPEAKER_01: And it did.
• 4:56
  SPEAKER_01: And I and I was about to publish and I thought, hang on.
• 4:59
  SPEAKER_01: So so I copied a couple of these quotes, pasted them into Google and to see if they existed.
• 5:04
  SPEAKER_01: And they didn't.
• 5:05
  SPEAKER_01: GPT had ChatGPT had made them up, and and it was my fault.
• 5:09
  SPEAKER_01: I didn't say find me existing quotes, I just said write me some quotes.
• 5:14
  SPEAKER_01: So it's like, okay, I can do that.
• 5:16
  SPEAKER_01: So it really so so again the the problem was me, not not not not the tool there.
• 5:23
  SPEAKER_01: But uh yeah, is it so so there's different ways of I think becoming a bit more sophisticated with your input.
• 5:30
  SPEAKER_01: One is to be more structured in how you do that, and another is to kind of let go and and and trust in a sense, and I I hesitate to use the term because some people will push back, but but trust in the intelligence of these models.
• 5:46
  SPEAKER_01: So let me start with that, the second one first.
• 5:49
  SPEAKER_01: I I have a good friend and former colleague, Anne Handley, who's very well known in the content marketing world and has written a few books, and and she wrote in in a newsletter of hers a few weeks ago, she's writing a new book right now.
• 6:02
  SPEAKER_01: One of the things she does is she will write a section of a chapter and then she will paste that into, I believe it's Claude 3 she uses.
• 6:10
  SPEAKER_01: She pastes it in and she says, Hey, what do you think?
• 6:14
  SPEAKER_01: Does this flow?
• 6:15
  SPEAKER_01: Am I missing something?
• 6:16
  SPEAKER_01: What do you think?
• 6:17
  SPEAKER_01: So here we have a New York Times best-selling author using Claude 3 as a cointelligence, as a partner.
• 6:25
  SPEAKER_01: So it's not using it as a, she's not seeing it as this dumb software tool.
• 6:29
  SPEAKER_01: She she has sufficient respect, but you know, she's this is this is this is her business, speaking and writing.
• 6:35
  SPEAKER_01: It has to be the best she can do.
• 6:36
  SPEAKER_01: So she's using this model.
• 6:38
  SPEAKER_01: So that's one way of hey, just just change your attitude to this.
• 6:42
  SPEAKER_01: This can be enormously helpful and powerful to you.
• 6:44
  SPEAKER_01: So don't be shy about asking tough questions.
• 6:48
  SPEAKER_01: Um, the other thing, the the other way I think people can improve on their inputs is to be more structured.
• 6:54
  SPEAKER_01: So I think Seth and I have talked about this before, but a very simple starting point on a structured framework is is called the race framework.
• 7:03
  SPEAKER_01: And that that is where you actually give the model much more detail in terms of instruction.
• 7:08
  SPEAKER_01: So R for race is a role.
• 7:11
  SPEAKER_01: What is the role?
• 7:12
  SPEAKER_01: And of course, stick with the general aviation industry.
• 7:15
  SPEAKER_01: So I just say you are speaking as an expert in the general aviation industry.
• 7:20
  SPEAKER_01: So it gives you telling the model who they are.
• 7:23
  SPEAKER_01: Uh, A is for action.
• 7:24
  SPEAKER_01: I want you to write a white paper on the state of the industry.
• 7:29
  unknown: All right.
• 7:31
  SPEAKER_01: All right, here is I'm gonna give you tons of information about our company, about the industry, um, about what we're doing.
• 7:37
  SPEAKER_01: Here's a whole bunch of data.
• 7:39
  SPEAKER_01: You're just giving it a huge amount of background data.
• 7:41
  SPEAKER_01: And then that's a that's we we've we've got role, action, context, and then examples.
• 7:45
  SPEAKER_01: And I'll say, okay, here are five examples of really successful white papers we've done in the past.
• 7:51
  SPEAKER_01: So this could be this this could be a hundred pages of information.
• 7:54
  SPEAKER_01: And in the early days of these models, that was a bit of a stretch.
• 7:57
  SPEAKER_01: But something like Gemini, the context window, can take like over a million tokens now.
• 8:02
  SPEAKER_01: Um, so so you can put in a whole book in context.
• 8:06
  SPEAKER_01: So with this example, I'm structuring this.
• 8:08
  SPEAKER_01: I'm saying, hey, look, this is who you're speaking as, here's what I want you to do.
• 8:12
  SPEAKER_01: Here's a bunch of background information on the company and on the industry, and here's some previous examples.
• 8:18
  SPEAKER_01: So that's a very simple framework, but very, very powerful.
• 8:20
  SPEAKER_01: So I put that in, then I ask it, you know, based on this information, please write a white paper or case study or whatever it is.
• 8:28
  SPEAKER_01: That's that by being a bit more formal and structured in that way, you're you're you're almost guaranteed to get a much, much better output.
• 8:35
  SPEAKER_01: So usually disappointing output is not the problem with the model, it's the problem with just either ambiguity or insufficient information in the prompt.
• 8:44
  SPEAKER_02: Do you think that people also maybe misalign with some of the core core AI capabilities of a specific model, right?
• 8:53
  SPEAKER_02: Because we you know, talk about Claude, talk about, you know, um, you know, the one from uh from Google, talk about, you know, so there's there's varying abilities that these models have.
• 9:03
  SPEAKER_02: And I don't think people keep that in mind either, which by the way, is also a catch 22 because you get into this notion of if you want to good get you know aligned with good accuracy, take the same prompt and ask it across all the models to see what the responses are.
• 9:19
  SPEAKER_02: But at the same time, you may have undercomplicated or overcomplicated the prompt to a model that you really don't understand what the core capabilities are and what it could respond, because like Cloud3 is amazing, but is it Sonnet, haiku, or Opus?
• 9:32
  SPEAKER_02: Do you know what I mean?
• 9:32
  SPEAKER_02: Which are variances in synchronization andor just broad um you know decoration.
• 9:38
  SPEAKER_02: So I don't know what your thoughts are.
• 9:40
  SPEAKER_01: And it's tricky, and people, you know, like I said, Ann Handler uses Cloud3.
• 9:43
  SPEAKER_01: I know a lot of writers prefer Claude to say GPT-4.
• 9:47
  SPEAKER_01: I I still use I use Cloud3, I I use Gemini, I use all kinds of models, but I I'll probably still use Chat GPT-4 the most, um, just simply because I'm most familiar.
• 9:58
  SPEAKER_01: So it's it's almost like it's like um, you know, three different three or four different software tools for for working with images.
• 10:05
  SPEAKER_01: Could I learn all three or four?
• 10:07
  SPEAKER_01: I could, but my time is probably better served just learning a lot about one, unless there are significant capabilities that something else offers.
• 10:15
  SPEAKER_01: But to that point, sometimes what I'll do is I'll do something in in GBD4, and then I'll paste that into Gemini.
• 10:23
  SPEAKER_01: And I'll say, hey, what do you think?
• 10:25
  SPEAKER_01: So I'll have different models checking on each other's output.
• 10:29
  SPEAKER_04: Yeah, and uh tools like uh oh goodness, what's the name of it?
• 10:32
  SPEAKER_04: There's a tool that you can use to submit to multiple large language models.
• 10:36
  SPEAKER_04: I've I've talked about it before.
• 10:38
  SPEAKER_01: Oh, like perplexity.
• 10:39
  SPEAKER_04: Well, it'll give you the ability to do perplexity and to do uh Chat GPT and to do Claude and to and compare.
• 10:46
  SPEAKER_04: But I think to your point and Chris's point, you know, sometimes it can overcomplicate things and you have to make choices and now you have to integrate them and find the best one.
• 10:56
  SPEAKER_04: So I think it's it, I I like the way you provided the example of how Anne Hadley does it, because again, I I think I look at the same thing and I say, where can I improve it versus giving it tasks in the start?
• 11:08
  SPEAKER_04: But to your point around ambiguity, you know, when we when we were when we've been dealing with search problems over the years, you know, we always used to say search terms are short, ambiguous, and an approximation of the search's real information needs, right?
• 11:22
  SPEAKER_04: Because they would, you know, they would put in a very ambiguous uh query and expect a very specific result.
• 11:28
  SPEAKER_01: Well, and it's because it's because of the the the interface.
• 11:31
  SPEAKER_01: Search was built to actually accommodate the fact that we don't want to type in 100 words, we want to type in four words.
• 11:37
  SPEAKER_03: Right.
• 11:37
  SPEAKER_01: Um, so why that's why even even long tail keywords, you know, tend to be four, five, six, or seven words and not fifty.
• 11:43
  SPEAKER_01: Exactly.
• 11:44
  SPEAKER_01: So it it so but but so you you're right, people bring that kind of expectation or that behavior to these models and it and expect it to work.
• 11:52
  SPEAKER_01: But if this is something very new.
• 11:55
  SPEAKER_04: You know, it's also interesting is I've seen in different contexts, like uh in a one chat, one playbook for LLMs where you're giving it different tasks and you have a lot of pre-built context and every lot of pre-built prompts.
• 12:09
  SPEAKER_04: And in some cases, these uh uh these uh folks who are demonstrating this would say, get angry at it, or like berate it, you know, tell it that that was a terrible job and to try much harder and you're very disappointed, right?
• 12:23
  SPEAKER_04: And and then the model would say, I'm so sorry, you're right, I can do much better.
• 12:27
  SPEAKER_04: And then other times it's like encourage it, tell it that it's doing great.
• 12:31
  SPEAKER_04: Wow, that was awesome.
• 12:32
  SPEAKER_04: Thank you so much.
• 12:33
  SPEAKER_04: Can you do a little better job?
• 12:35
  SPEAKER_04: Like, thank you for your feedback.
• 12:36
  SPEAKER_04: I mean, it's so funny how some people believe that uh negative emotions uh impact results, some people believe that positive emotions impact the results, and you do get different results, but you know, it's a little bit uh black magic, a little bit of uh uh of right.
• 12:52
  SPEAKER_01: It is it is weird.
• 12:53
  SPEAKER_01: And in a sense, it's what makes this such an interesting area right now.
• 12:57
  SPEAKER_01: Even the people creating these models don't quite understand exactly what's going on sometimes.
• 13:02
  SPEAKER_01: Um there was a, I think it was last week or the week before, they had two AIs talking to each other and they got into a bit of a muddle.
• 13:10
  SPEAKER_01: One of the AIs was just like fought in a loop.
• 13:12
  SPEAKER_01: And the first AI said, Hey, are there any humans out there?
• 13:16
  SPEAKER_01: Because the AI I'm talking to needs to be shut down, which was like completely out of the blue.
• 13:22
  SPEAKER_01: And the researchers were like, Wait, what?
• 13:24
  SPEAKER_01: We just had these two AIs talking to each other.
• 13:26
  SPEAKER_01: One of them should not even think about saying, Hey, human, but my friend's got a problem here.
• 13:31
  SPEAKER_01: Can you close it down?
• 13:32
  SPEAKER_04: That's crazy.
• 13:33
  SPEAKER_04: So it does this fascinating.
• 13:34
  SPEAKER_01: Yeah, it's it's really interesting area.
• 13:37
  SPEAKER_04: Yeah.
• 13:37
  SPEAKER_04: Well, the emergent properties are very uh fascinating because theory of mind was supposed to be uniquely human, right?
• 13:44
  SPEAKER_04: When and children learn theory of mind, you know, around age three or something, where you can start to plan and understand what the other person is thinking, and that's when they start looking at deception.
• 13:55
  SPEAKER_04: No, I didn't uh you know eat the cookies, uh, you know, my baby brother ate the cookies, right?
• 14:00
  SPEAKER_04: And you're trying to deceive people that that emerges in children at a very young age, but it starts to mature over years.
• 14:07
  SPEAKER_04: Well, it spontaneously emerged in large language models around 2022, and by 2024, it's been able to perform at the level of an adult.
• 14:17
  SPEAKER_04: And nobody tried to predict that, nobody tried to produce that, nobody tried to program that.
• 14:22
  SPEAKER_04: That emerged spontaneously, which is crazy.
• 14:24
  SPEAKER_04: So when you start seeing some of these emergent behaviors and you start thinking about, well, what's really going on in these models?
• 14:32
  SPEAKER_04: I always look at it as you're trying to get closer in vector space across all these dimensions by giving it role, by giving it context.
• 14:40
  SPEAKER_04: But the emotional prompts, I I'm not sure I understand.
• 14:44
  SPEAKER_04: You know, the emotional prompts are what's very strange.
• 14:46
  SPEAKER_04: Do you want to talk more about that in in your work and how you see that impacting the quality and of outputs?
• 14:55
  SPEAKER_01: So, yeah, the emotional side, this is this is a like big area, and and as you know, it's an area that that really, really interests me.
• 15:01
  SPEAKER_01: But because of course, the these models they can't experience emotion, they can't fall in love, they can't eat ice cream, they can't walk on the beach and feel sand and water between their toes.
• 15:12
  SPEAKER_01: They don't have that experience of of the sensory, they don't have sensory experiences, they they they don't feel emotions firsthand.
• 15:19
  SPEAKER_01: But they read about it, you know, they've read the you know, they've read love stories, they've read Romeo and Juliet, they know what humans mean when we talk about love, but they've never felt it.
• 15:28
  SPEAKER_01: And and and there is there's a kind of limit there because of course they they can't feel and and experience emotions the way we too, we do.
• 15:37
  SPEAKER_01: But again, with the right mindset and the right prompting, they can get pretty close.
• 15:46
  SPEAKER_01: So I I I will encourage a model to hey, what one of the things these models do surprisingly well is build avatars uh say an ideal customer avatar, and even from an emotional perspective.
• 16:02
  SPEAKER_01: So I'll feed the model all kinds of information, like uh, you know, interview transcripts, product and service reviews, transcripts of customer service, just uh just all the input I get that's in the voice of customers, and I'll throw that all in and I'll say, hey, based on this information, create me a uh you know, an ideal customer avatar based on this.
• 16:24
  SPEAKER_01: And there's a lot of emotion in there.
• 16:26
  SPEAKER_01: And it does a really interesting job of that.
• 16:28
  SPEAKER_01: And and I've used this as a tool for many projects that I've done, is I start with the avatar.
• 16:34
  SPEAKER_01: And so I make that part of, in fact, my context.
• 16:37
  SPEAKER_01: I say, okay, today we are writing as Emma, who is a 45-year-old pilot in the general aviation industry.
• 16:45
  SPEAKER_01: Uh so I so I keep that and I might update it as I get more information.
• 16:49
  SPEAKER_01: But I can I can when it does that, or I can do, I can either say do the avatar, or I can another thing that I'll do is I'll say, hey, write me a create a sentiment analysis based on this input.
• 17:00
  SPEAKER_01: And the sentiment analysis is really interesting because that is very much focused on emotion.
• 17:05
  SPEAKER_01: So again, it's not human, it doesn't feel emotion firsthand, it doesn't have any kind of sensory experiential experience of this, but it does a pretty good job of like sentiment analysis.
• 17:16
  SPEAKER_01: And again, I'll put that in as part of my input.
• 17:18
  SPEAKER_01: And I and I've seen like before and after, if I'm writing copy or content, uh before or before and after including a sentiment analysis in there, there'll be a very, very different quality and feel in the output from that model.
• 17:33
  SPEAKER_01: So yeah, they they don't get emotion, but you can by feeding it, by helping it, you can get it into that place.
• 17:40
  SPEAKER_04: It's almost like getting it into that mind space, which is the vector space, right?
• 17:45
  SPEAKER_04: It's like, how do you get into that emotional mind space?
• 17:48
  SPEAKER_04: Well, it's like an actor getting into character, you're getting into that place in the an in-dimensional vector space, and there's 10 thousand and thousands of dimensions.
• 17:58
  SPEAKER_04: Chris, you had a uh I was just gonna I was just gonna ask you.
• 18:01
  SPEAKER_02: And so at what, because you brought this up right obviously, I wanted to double-click on a little bit, right?
• 18:06
  SPEAKER_02: In terms of when is it then appropriate for the human in the loop to come out, right?
• 18:12
  SPEAKER_02: And to be that.
• 18:13
  SPEAKER_02: I mean, I don't know if that's specific to copywriting and some of the stuff, the work that you do on a day-to-day, but just maybe in general, when you know, because it becomes a trust issue.
• 18:22
  SPEAKER_01: Right.
• 18:23
  SPEAKER_02: Well, but when does human in the loop really become essential in your mind?
• 18:27
  SPEAKER_01: So human in the loop, I think, is essential at every step.
• 18:29
  SPEAKER_01: And in fact, the term comes out of the training period for these models before they even became public.
• 18:35
  SPEAKER_01: You had a human in the loop sort of putting in guardrails.
• 18:38
  SPEAKER_01: No, do not tell people how to if they ask you.
• 18:41
  SPEAKER_01: Uh so that's the human in the loop in terms of training.
• 18:43
  SPEAKER_01: The human in the loop, in terms of our role on on the when we're using these models, is yeah, we have to check stuff.
• 18:50
  SPEAKER_01: Uh, there's this wonderful example of well, well, you we've seen, I think it was when Gemini first came out, and people asked it to show pictures of the founding fathers.
• 18:58
  SPEAKER_01: And Gemini tried to be a little bit too woke about it, I guess, and and and and had different colours and you know, peop pe people of different colours and ethnicity, where in fact they were all kind of white male Europeans, the founding fathers.
• 19:11
  SPEAKER_01: So people got upset about that.
• 19:13
  SPEAKER_01: So you have to check for stuff like that.
• 19:14
  SPEAKER_01: There's another one, somebody asked one of these models, I forgot which it was, is like, what's the record time for walking across the English Channel?
• 19:22
  SPEAKER_01: And for those who don't know, the English Channel is a body of sea between England and France.
• 19:28
  SPEAKER_01: You can't walk over it.
• 19:30
  SPEAKER_01: But the model gave a very, very convincing answer.
• 19:33
  SPEAKER_01: And it named the person and how long it took that person to walk across the channel.
• 19:37
  SPEAKER_01: So it just makes stuff up.
• 19:38
  SPEAKER_01: It's like hallucinating.
• 19:39
  SPEAKER_01: It's just make stuff up.
• 19:40
  SPEAKER_01: But you've got to be really careful there, because it makes stuff up very, very convincingly.
• 19:44
  SPEAKER_01: You've got you've got to look for bias.
• 19:47
  SPEAKER_01: Ivory has done this wonderful series of how to prompt these models not to be biased.
• 19:52
  SPEAKER_01: So if you go to these models and say, show me a picture of the most create a picture of the most beautiful woman in the world, it'll be Barbie.
• 19:59
  SPEAKER_01: It'll be blonde, blue-eyed Barbie.
• 20:02
  SPEAKER_01: And again, what it's it's not because the model is biased, it's because the data, the human data it's trained on is biased.
• 20:10
  SPEAKER_01: So, so Ivory has done this wonderful thing of actually teaching people how to prompt, where the idea of what beauty is is much, much more varied in terms of size and shape and color and ethnicity and gender and things like that.
• 20:23
  SPEAKER_01: So, yeah, in terms of the human in the loop, those are the things we watch for.
• 20:26
  SPEAKER_01: You have to check for accuracy.
• 20:28
  SPEAKER_01: I've done I've done work where it's on the medical side, where it's cited a whole bunch of references and sources, uh, which is great.
• 20:37
  SPEAKER_01: I love that, because it saves a huge amount of time and also inputs a lot of information I might not have found.
• 20:42
  SPEAKER_01: But you have to check those sources because sometimes they're not accurate or they're not relevant, or it's not exactly as said.
• 20:49
  SPEAKER_01: So, yes, and it's one of the things that worries me all the time is that as companies get into this, they are using these models almost like a conveyor belt in a factory.
• 20:59
  SPEAKER_01: You know, just the output just flies off the end of the conveyor belt.
• 21:02
  SPEAKER_01: There's no human in the loop checking it.
• 21:05
  SPEAKER_01: So that speaks to things like accuracy, bias, uh, things like that, but also to creativity.
• 21:12
  SPEAKER_01: Like that the the these models are having a huge impact in the advertising and marketing industry.
• 21:17
  SPEAKER_01: WPP, which is the largest agency group in the world, is integrating this across all of their 100 major agencies so that all their writers, designers are using this model now.
• 21:28
  SPEAKER_01: And and there's a marketing is being automated in in all kinds of ways already.
• 21:33
  SPEAKER_01: But what worries me is like I'm I'm an old school agency guy.
• 21:36
  SPEAKER_01: I started off in ad agencies in London in the 1980s, and we had a creative director, and nothing left the creative department without the creative director, the boss of all bosses, looking at it and saying, that's good enough.
• 21:48
  SPEAKER_01: That that is a brand voice voice is correct.
• 21:51
  SPEAKER_01: Yes, that is creative.
• 21:52
  SPEAKER_01: Yes, that works.
• 21:53
  SPEAKER_01: What worries me now is I see a whole bunch of stuff coming out at the end that doesn't have that creative director at the end.
• 22:00
  SPEAKER_01: And saying yes or no.
• 22:01
  SPEAKER_01: Because I used to get literally flapshed around the back of the head if I tried to put something out that wasn't good enough.
• 22:07
  SPEAKER_04: Right.
• 22:08
  SPEAKER_04: And there's a lot of the just this mediocrity of content that's being pushed out.
• 22:14
  SPEAKER_04: And I've had to reprimand some of my folks to say, hey, don't just give me a lot of AI power uh driven content or generated content unless it's relevant.
• 22:25
  SPEAKER_04: I was doing a facilitated session.
• 22:28
  SPEAKER_04: Uh and and I didn't want to say too much, but a lot of the stuff that was created in preparation was not necessarily relevant.
• 22:35
  SPEAKER_04: You know, it was just a lot of you know stuff that was not.
• 22:39
  SPEAKER_01: It can be very damaging.
• 22:40
  SPEAKER_01: I purchased a service from a company and they sell sent me this welcome onboarding package, like just to me, Nick, to make you feel welcome.
• 22:49
  SPEAKER_01: And I read it and I was three lines in that I recognized it had been written by AI and not edited by a human.
• 22:55
  SPEAKER_03: Right.
• 22:56
  SPEAKER_01: And I had that that moment of recognition was also a moment of deep disappointment.
• 23:01
  SPEAKER_01: It was also a moment where my perception of their value as a brand took a huge dive.
• 23:08
  SPEAKER_01: Yep.
• 23:09
  SPEAKER_01: They've broken trust with me.
• 23:11
  SPEAKER_01: Yep.
• 23:11
  SPEAKER_01: They they gave me the impression I was getting this personal welcome onboarding package, but it wasn't.
• 23:16
  SPEAKER_01: It was automated.
• 23:17
  SPEAKER_01: And my thought was not cool.
• 23:20
  SPEAKER_04: Where's where's the value?
• 23:21
  SPEAKER_04: Yeah.
• 23:22
  SPEAKER_04: And it does have like an instantaneous negative impact on the brand.
• 23:27
  SPEAKER_01: And I think and I think all of us who are interacting with this are getting better at recognizing.
• 23:32
  SPEAKER_01: Like, like when I look at like visual output from DALI or Mid Journey now, and I and I look at that, at the beginning, I used to think, oh my goodness, that's amazing.
• 23:40
  SPEAKER_01: I can't even tell that from the real thing.
• 23:42
  SPEAKER_01: But now we can, right?
• 23:43
  SPEAKER_01: You can instantly see when these things come out of those models.
• 23:46
  SPEAKER_01: And it's the same with writing.
• 23:47
  SPEAKER_01: I think at the beginning, we didn't we didn't have enough experience.
• 23:50
  SPEAKER_01: So we hadn't seen enough of it to be able to recognize output.
• 23:54
  SPEAKER_01: Right.
• 23:54
  SPEAKER_01: And that that's set to another point that you we've talked about before is the whole layering in emotional intelligence.
• 24:02
  SPEAKER_01: So this is a point of differentiation.
• 24:04
  SPEAKER_01: This is something that you want to do as a human.
• 24:07
  SPEAKER_01: It's at that end of the conveyor belt.
• 24:09
  SPEAKER_01: Before it goes off the end, there are things you can do.
• 24:11
  SPEAKER_01: You're going to be the human in the loop, checking for accuracy, bias, etc.
• 24:14
  SPEAKER_01: But you're also going to be the human human and saying, you know what?
• 24:17
  SPEAKER_01: And I mean, I've done, I have this iterative approach.
• 24:20
  SPEAKER_01: Human human, not the simulated human.
• 24:22
  SPEAKER_01: Yeah, me, the human human.
• 24:24
  SPEAKER_01: Still, still the real me.
• 24:26
  SPEAKER_01: So I'll have this iterative approach where I'll get, let's say I'm asking it to write an uh draft an email for me.
• 24:32
  SPEAKER_01: I'll say, do the email, here's the information.
• 24:34
  SPEAKER_01: I I give it all the race framework information.
• 24:37
  SPEAKER_01: And I say, ah, a bit too much feature, not enough benefits, but rewrite it with benefits up front.
• 24:42
  SPEAKER_01: And I'll say, okay, let's start with a story.
• 24:44
  SPEAKER_01: Start with a story or anecdote up front.
• 24:46
  SPEAKER_01: That catches people, that gets people engaged.
• 24:49
  SPEAKER_01: So it does.
• 24:50
  SPEAKER_01: It'll come up with a little anecdote or story.
• 24:52
  SPEAKER_01: And then what happens?
• 24:54
  SPEAKER_01: So now the kind of structure of the thing, it's done a lot of the work for me.
• 24:57
  SPEAKER_01: But then I will go in and I'll say, you know what?
• 25:00
  SPEAKER_01: That story reminds me of something that either happened to me or I've heard, I want to replace it with real life.
• 25:05
  SPEAKER_01: This is where these whole areas of emotional intelligence, which has been around for decades, you know, the study of emotional intelligence and also the use of emotional intelligence and marketing.
• 25:15
  SPEAKER_01: But it's very, very relevant now with AI because AI can pretend it can give you a good impression of emotional intelligence, but it can't have it.
• 25:26
  SPEAKER_01: And as humans, we are very, very attuned, very, very sensitive to connection.
• 25:33
  SPEAKER_01: I'm rambling a bit, but I just want to throw in another quick story.
• 25:36
  SPEAKER_01: I I was at a we're moving shortly, and I had to paint some walls here, clean up a bit.
• 25:42
  SPEAKER_01: So I went to a big box store, bought some paints, brushes, went to check out, and it was automated.
• 25:47
  SPEAKER_01: And it's so I you know I scanned, bagged, and paid myself and left.
• 25:51
  SPEAKER_01: And I again at that moment I felt this little twinge of loss and disappointment.
• 25:56
  SPEAKER_01: Because I actually wanted my moment with the cashier.
• 25:59
  SPEAKER_01: That the cashier is a stranger, would have maybe shared five words together.
• 26:04
  SPEAKER_01: But it matters.
• 26:05
  SPEAKER_01: It matters.
• 26:06
  SPEAKER_01: We're social animals, and that human connection matters.
• 26:09
  SPEAKER_01: And it struck me as being very, very powerful that that something as minor as that, five words with a complete stranger I'd probably never see before, mattered that much to me, but it does.
• 26:21
  SPEAKER_01: That is missing from these AI interactions and tools.
• 26:25
  SPEAKER_01: So that's why I have it like it's a part of my process is I will look at the the elements of emotional intelligence.
• 26:32
  SPEAKER_01: I will look at ways of adding a real human element to everything that I create with these models so that people get that moment of recognition, of connection, of like, hey, it's a person.
• 26:44
  SPEAKER_04: Yeah, yeah.
• 26:45
  SPEAKER_04: So so on that topic, how do you see training and education evolving to keep pace with um AI and the usage of AI, and especially in the realm of injecting that a bit of a little bit of emotional intelligence and connection?
• 27:02
  SPEAKER_04: And again, when we're trying to remove too much human uh family interactions, we we lose a lot.
• 27:09
  SPEAKER_04: And and yet, you know, we're trying to automate so that uh humans can engage, right?
• 27:14
  SPEAKER_04: You know, machines automate, humans engage.
• 27:16
  SPEAKER_04: And talk a little bit about being more human, uh, your project and how that's um uh training, focusing on training for marketing groups.
• 27:26
  SPEAKER_04: Like what does that entail?
• 27:28
  SPEAKER_04: And what what does that involve?
• 27:30
  SPEAKER_04: And how do what do companies learn from that uh initiative?
• 27:34
  SPEAKER_01: So going back to the first part of your setup of like where companies are at in terms of education, uh, I I think it's very, very early days right now.
• 27:43
  SPEAKER_01: Uh I was hearing some research recently um about basically employees.
• 27:49
  SPEAKER_01: I think it was 67% of employees said they're either getting no education or insufficient education with regard to AI.
• 27:56
  SPEAKER_01: A lot of companies don't even have AI policies in place.
• 28:00
  SPEAKER_01: Employees are using AI, they're using it, they're coming in, they're using it on their own phones, but there's no formal policy yet in a lot of businesses and a lot of a lot of individuals are saying we'd love to use this.
• 28:09
  SPEAKER_01: Our company is not giving us the training and they're not giving us the green light, and they're not telling us which models, and they're not telling us which databases we can access or not, what information we're allowed to, you know, make available to these models or not.
• 28:22
  SPEAKER_01: And I get it, it's incredibly complex, but but in terms of training, I think we're way behind.
• 28:28
  SPEAKER_01: With regards to be more human, this is my my effort to be helpful in this space.
• 28:32
  SPEAKER_01: So I have a website called bemorehuman.ai, and and the group I'm qualified, I guess, best qualified to speak to are people in the marketing department.
• 28:40
  SPEAKER_01: Uh of like, okay, within a company, what are the processes?
• 28:44
  SPEAKER_01: What what what are the steps we can take to, on the one hand, lean into AI and do everything, you know, use it as capabilities to the max, but also how to again be be the cashier at checkout, how to be more human, how to introduce that human element.
• 29:03
  SPEAKER_01: So again, it's kind of step-by-step process-based of where I'm talking about maxing out on AI, understanding how to integrate emotional intelligence, and then how to integrate the two.
• 29:15
  SPEAKER_01: I say you're getting the best of both worlds.
• 29:16
  SPEAKER_01: You're getting the maxing out of AI and you're maxing out on being more human.
• 29:21
  SPEAKER_02: Don't you feel like, too, that that's, you know, because I I know one of the you know hot topics for you is the sameness trap, right?
• 29:28
  SPEAKER_02: Right.
• 29:28
  SPEAKER_02: That's going to be one of the key areas to stay unique, to keep unique thoughts, brand awareness, whatever that is, because that's what made it, made the brand in the beginning was the uniqueness of it.
• 29:43
  SPEAKER_02: And or whatever it is, right?
• 29:45
  SPEAKER_02: The human creates the unique fingerprint that you know touches all of this kind of stuff.
• 29:50
  SPEAKER_02: And I feel like at some point too, um, we have diminishing returns when we are all starting to use the same models for good writing, but now all they start and they start to look and feel the same.
• 30:02
  SPEAKER_01: Right.
• 30:03
  SPEAKER_01: So that that is definitely part of the kind of be more be more human training.
• 30:07
  SPEAKER_01: Is that addressing the sameness trap?
• 30:09
  SPEAKER_01: Because you're right, people use the same models, they've downloaded the same top 20 prompts from LinkedIn or Twitter.
• 30:15
  SPEAKER_01: Uh, there are people offering frame, you know, frameworks for you know for email and for white papers and for sales pages.
• 30:22
  SPEAKER_01: If we all use the same tools, we're all going to get very similar output, which, from a marketing and sales point of view, is the last thing we want.
• 30:29
  SPEAKER_01: Like there are companies who invested billions of dollars in creating a unique voice and brand.
• 30:36
  SPEAKER_01: The difficulty here, and it's really interesting watching it play out, and it will continue to be interesting, is that the math is so irresistible.
• 30:45
  SPEAKER_01: If if I can use AI to create marketing at a hundredth of the hundred times the speed, but one hundredth of the price, it is irresistible.
• 30:54
  SPEAKER_01: I can automate everything, I can fire all the humans, and I can I shouldn't say that with a smile on my face because it's not funny at all.
• 31:01
  SPEAKER_01: But it's like it's like the math is like, oh my goodness.
• 31:06
  SPEAKER_01: But the more you do that, the more you sound the same as your competition.
• 31:10
  SPEAKER_01: And hey, if there's anything important to marketing, it's differentiation.
• 31:13
  SPEAKER_01: You've got to differentiate yourself.
• 31:14
  SPEAKER_01: Well, you you can't if you're doing everything in the same way.
• 31:17
  SPEAKER_01: So again, that's why I step back and say, yes, use the models, but then introduce that emotional element, use emotional intelligence, build build an emotional avatar for your company or cooperation, and say, say, what is that?
• 31:30
  SPEAKER_01: What is the emotional avatar for our business?
• 31:33
  SPEAKER_01: Um, and then include that in your process.
• 31:36
  SPEAKER_01: So, yeah, you're right.
• 31:37
  SPEAKER_01: The same as trap is is is a huge problem.
• 31:39
  SPEAKER_01: And it and it it'll it'll be a problem because, like I say, it is so irresistible to automate as much as you can.
• 31:46
  SPEAKER_04: Yeah, I I always say that if you do what everybody else does, do with some efficiency.
• 31:51
  SPEAKER_04: Standardization gives you efficiency, right?
• 31:53
  SPEAKER_04: You might need that for data interchange or something, but differentiation gives you competitive advantage, right?
• 31:59
  SPEAKER_04: So we need to have that competitive advantage of differentiation.
• 32:02
  SPEAKER_01: It's it's that's how it's how Nike did it, it's how Apple did it, it's how did it.
• 32:06
  SPEAKER_01: You've got to set yourself apart.
• 32:08
  SPEAKER_02: It comes back to the creative and emotional side of things of why humans, you know, are critical and still in all of this.
• 32:15
  SPEAKER_03: Right.
• 32:16
  SPEAKER_02: I mean, we still work within the boundaries and the parameters, and we're trying to blow past it if we can, but that's the beauty of it is the creative So I I I was saying earlier that I come from an agency world and the creative director was the boss.
• 32:29
  SPEAKER_01: And then as I with a bit more experience, I became a creative director myself.
• 32:33
  SPEAKER_01: And yeah, there there is this deep like creative part that the the really great marketing is deeply creative.
• 32:41
  SPEAKER_01: And so with with that behind that history behind me, I use these models a lot and I'm amazed and incredibly impressed by what they can do.
• 32:49
  SPEAKER_01: But you know what?
• 32:50
  SPEAKER_01: If all the kind of pages that I've asked it to write, whether it's content or copy or email, I have never seen it write a really great headline.
• 33:00
  SPEAKER_01: You know, the kind of headline that that a top performing copywriter or credit director might come up with.
• 33:06
  SPEAKER_01: I've never seen anything to compete with with the top tier of human creativity in marketing.
• 33:11
  SPEAKER_01: So yeah, I absolutely believe that the human needs to be part of that process.
• 33:15
  SPEAKER_01: And you you have to you have to allow the human to be creative at their speed.
• 33:20
  SPEAKER_01: So so again, it's really tricky because these models are blindingly fast.
• 33:26
  SPEAKER_01: But like in the back in the 1980s, we'd be creating a print advertisement.
• 33:31
  SPEAKER_01: And we might spend, as a team, the periods, we might spend 10 days on one ad.
• 33:36
  SPEAKER_01: And it's like that's the creative process of how you so again, I'm not sure how that can now fit in to the process of working with with AI and automation because nobody wants to allow the creative team two weeks to come up with something really amazing.
• 33:51
  SPEAKER_01: It's too slow, too expensive.
• 33:53
  SPEAKER_04: It's changed a lot of the expectations uh that are out there.
• 33:58
  SPEAKER_04: So so where do you see this kind of where do you see this kind of evolving over the next um several years?
• 34:06
  SPEAKER_04: Like, what do you think organizations are going to what what's that in store for them, especially when you think about the sameness trap and model collapse?
• 34:15
  SPEAKER_04: And where will the disillusionment come in and where will the ahas be?
• 34:19
  SPEAKER_04: Do you think?
• 34:21
  SPEAKER_01: Where will it dis so there's going to be disillusion disillusioned?
• 34:24
  SPEAKER_01: So I like I said, I I think people, I think companies in particular, and and and the larger the company, the slower they are proceeding, and again for understandable reasons, and they're looking at say Jet GPT 4.0, and they're they're trying to prepare for that.
• 34:40
  SPEAKER_01: The trouble is by the time they're ready to use that, we'll have GPT 5 or 6, which will be something totally different all over again, with with different implications for how do companies integrate this and use it.
• 34:53
  SPEAKER_01: And and so, in terms of how do I see the future, I honestly have no idea.
• 34:57
  SPEAKER_01: Um, I know it's gonna be messy.
• 34:59
  SPEAKER_01: I know it's it's it's gonna be really messy.
• 35:02
  SPEAKER_01: There's gonna be more politics, there's good legislation, there's legislation that, like in you know, in Europe, there's legis quite severe, but a lot of constraints on AI in Europe already because of legislation.
• 35:12
  SPEAKER_01: Uh so there's gonna be politics, technology, there's social implications, there's work implications.
• 35:17
  SPEAKER_01: And the trouble is I have no idea what the future holds because I have no idea when GPT 5 or some other equivalent will come out and what his capabilities will be.
• 35:27
  SPEAKER_01: Will it just be one step ahead of where we are, or will it be a huge step ahead of where we are?
• 35:31
  SPEAKER_01: Because if so, it's kind of all bets are off and we have to rethink a lot of stuff.
• 35:36
  SPEAKER_02: That's awesome.
• 35:36
  SPEAKER_02: So let me ask you this real real quick too.
• 35:38
  SPEAKER_02: If is there something, which which I generally do with a lot of our guests, is there something that that we didn't ask or that I didn't ask you that I should have, that is, you know, maybe kind of pulls this together?
• 35:51
  SPEAKER_02: Or something that I just missed, right?
• 35:54
  SPEAKER_02: In terms of of what this looks like with AI and you, AI and marketing, AI and well, generative really, but I think we've covered, hey, we could talk about this for hours and hours.
• 36:04
  SPEAKER_01: But but but I but I think generally, if I if I always have one last thing to say, whether it's someone working on their own or someone working within a company, is for goodness sake, lean into this.
• 36:13
  SPEAKER_01: Don't be part of the crowd who say, oh, it's just hard, it's not going to impact me.
• 36:17
  SPEAKER_01: It doesn't matter.
• 36:18
  SPEAKER_01: I think almost whatever the business you're in, you you need to lean into this because it will impact you.
• 36:24
  SPEAKER_01: I always, like I said, we're we're moving house and we went to the lawyer a few days ago.
• 36:28
  SPEAKER_01: And and I live in Montreal where the official language is French, so the contracts are in French.
• 36:33
  SPEAKER_01: And she sent me this contract, and I don't read French well enough for a legal document.
• 36:37
  SPEAKER_01: So I so I asked GPT ChatGPT to translate it, which she did.
• 36:41
  SPEAKER_01: And then I said to ChatGPT, I said, Oh, by the way, before I sign this thing, are there any points where you think I should you know ask questions or or you know important points I should consider carefully?
• 36:51
  SPEAKER_01: And it gave me a little list of three or four things.
• 36:54
  SPEAKER_01: So then we went in and spoke to the lawyer and and um I said what I'd done, and she was blown away.
• 37:00
  SPEAKER_01: She didn't realize she one, she didn't realize that ChatGPT could translate from French into English.
• 37:05
  SPEAKER_01: Two, when I showed her the questions that GPT suggested I ask her, she was blown away because she said, Well, if I was representing you as a client, that is exactly what I'd have told you to ask.
• 37:16
  SPEAKER_01: And she she was shocked.
• 37:19
  SPEAKER_01: And I think part of that is she started in her mind thinking, well, hang on, what is this gonna do for do for my business?
• 37:26
  SPEAKER_01: I don't know, you know, whether you're a writer, whether you're a marketer, whether you're a lawyer, an accountant, what whatever.
• 37:31
  SPEAKER_01: Final piece of advice to me was be lean into this because it's gonna impact you.
• 37:35
  SPEAKER_01: The more literate you are, as it were, in these tools, the better.
• 37:39
  SPEAKER_04: Yeah.
• 37:40
  SPEAKER_04: Well, Nick, thank you so much for sharing your insights and for joining us today on the Early AI podcast.
• 37:47
  SPEAKER_04: And uh thanks everyone for tuning in.
• 37:50
  SPEAKER_04: Appreciate your time, Chris, and for Carolyn behind the scenes.
• 37:53
  SPEAKER_04: And again, thank you, Nick, so much for uh your participation.
• 37:57
  SPEAKER_04: You're very welcome.
• 37:58
  SPEAKER_04: Hey, as you know, I love talking about this stuff.
• 38:00
  SPEAKER_04: Yeah, it's great stuff, and and we'll have to talk about some areas where we can uh further collaborate.
• 38:05
  SPEAKER_04: Um again, thanks to everyone for joining us.
• 38:08
  SPEAKER_04: We'll have some information about how to reach you in the show notes, and uh, we'll see you all next time.
• 38:15
  SPEAKER_04: Thanks again.
• 38:16
  SPEAKER_00: Visit early.com to find links to the full podcast on all audio platforms to listen on the go.
• 38:22
  SPEAKER_00: Thank you for watching.
• 38:24
  SPEAKER_00: Contact us today at infoearly.com.