← The Artificial Intelligence Show
The Artificial Intelligence Show
#154: AI Answers: The Future of AI Agents at Work, Building an AI Roadmap, Choosing the Right Tools, & Responsible AI Use
01:08:16
Loading summary
Transcript97 lines
- [00:00]
Paul Raitzer
I do think that three to five years from now it's going to be very commonplace, that it's just part of your job description to build and manage agents and agent systems. Welcome to AI Answers, a special Q and A series from the Artificial Intelligence Show. I'm Paul Raitzer, founder and CEO of SmartRx and marketing AI institute. Every time we host our live virtual events and online classes, we get dozens of great questions from business leaders and practitioners who are navigating this fast moving world of AI. But we never have enough time to get to all of them. So we created the AI Answers series to address more of these questions and share real time insights into the topics and challenges professionals like you are facing. Whether you're just starting your AI journey or already putting it to work in your organization, these are the practical insights, use cases and strategies you need to grow Smarter. Let's explore AI together. Welcome to episode 154 of the Artificial Intelligence Show. I'm your host, Paul Raitzer. Today I am joined by my co host Kathy McPhillips, our chief growth officer. This is the second edition of our new AI Answers series. So if you haven't heard this before, it is not replacing our weekly. Every Tuesday we drop the weekly episode that Mike and I do. AI Answers is a new series we just introduced weeks ago, Kathy Early June. Yeah, so basic premise here. Kathy And I do two free classes every month. So one is Intro to AI. We started that one in fall of 2021, is that right? Yeah. So we're coming up on episode or session 50 of that intro to AI class. I think we're on 49 is our next one correct? Yes. So we've had almost 30, 35,000 people roughly register for that class over the last four years or so. And then Scaling AI 5 Essential Steps to Scaling AI is another free class that we do each month. That one we are coming up on number nine or ten. I think we've been doing that one for almost a year now. And that one we've had probably a little over 10,000 people register for. So each of these classes gets, you know, somewhere between 800 and 1500 people, depending on, you know, what time of the month we're doing it and how many weeks in between. It's usually like four weeks in between. And we will get dozens of questions, sometimes 70 to 100 or more. And we can only get to like, I don't know, 7 to 10. On a good day, Kathy and I might tackle on a one of these sessions. So the AI Answers series is all about trying to answer more of those questions. So the idea is to try and provide as as much kind of input as we can. But also we just find it interesting to look at how the questions evolve. So the kinds of questions we were getting a year ago are completely different than the kinds of questions we're getting now. And so in some ways it's almost like real time insights into kind of where the market is and what people are thinking about related to AI. So hopefully this series is really helpful to people. I We have great feedback for the first episode. So we are planning to do this. There'll probably be 2ish a month. There might be a third because we'll also do these for our virtual events. And then we may mix in a couple other special AI answer sessions. But we've got the intro to AI and then we'll do the next one will be probably next week on Scaling AI because we have our Scaling AI class on Thursday, the day this is dropping. Yeah. So yeah, that's the background on AI Answers again, not replacing the weekly. The weekly still comes to you every Tuesday with me and Mike and then AI Answers is two to three times a month with me and Kathy. So today's episode is brought to us by Mekon at Macon 2025. This is the marketing AI conference that we started in 2019. This is the 6th annual marketing AI conference. This is the big thing. I mean there's lots of big things we do every year, but this is sort of like the first. This was kind of the origin of, you know, as we really started building out Marketing Institute. The Marketing AI Conference was the flagship event. Kathy works tirelessly along with many of the other people on our team to put this event on every year. It is happening August 14th to the 16th. This is in our hometown of Cleveland, Ohio, at the convention center, right across from the Rock and Roll hall of Fame and Lake Erie is a beautiful spot we are looking for. I don't know, last year we had what, 1100, I think came Kathy. We had 700 the year before, 300 the year before that, roughly. So we are trending, continuing to trend up. We're hoping for 1500 I think is the goal. I usually try and like, usually look up that. Kathy always like cringes whenever I start throwing out numbers. And I try real hard to be like conservative at these things. But 1500 is, is kind of what we're shooting for this year.
- [04:55]
Kathy McPhillips
That's my goal too. So you're good?
- [04:56]
Paul Raitzer
Good. We're aligned.
- [04:57]
Kathy McPhillips
Yes, we are. Aligned.
- [04:58]
Paul Raitzer
So we'd love to see everyone in Cleveland if you can be there. It's going to be an amazing three days. So you can go to Macon AI that is M A I C O N AI again, that is October 14th to the 16th. 16th in Cleveland, Ohio. The agenda is live. It's not full yet. The is not finalized. We're still working on some of the main staged and keynote talks, but I don't know what cat's about 80% or so of the agenda is probably up there. Okay. And the speaker lineup. So you can go get a sense of. You know what, what you can look forward to it at the event and.
- [05:29]
Kathy McPhillips
If you got yesterday's. Well, today's email, I guess yesterday's when. When this goes live. Yes. 10 things why 10 reasons why you should be in Cleveland in October.
- [05:38]
Paul Raitzer
October. I didn't, I didn't open that yet. I was actually in New York, so we're I guess context for people. We are recording this on Wednesday, June 18th at about 5pm Eastern Time because I was actually in New York this week at a Movable Inc. Event. So Movable Inc. Is an AI powered personalization platform for digital marketers. And I've done a series of talks with them this year. They've been a great partner of ours. And so I was actually doing a keynote for them at their Think Summit Tuesday morning. And then I, I was running a workshop for a group of some incredible marketing leaders Tuesday afternoon. I just landed back in Cleveland from that 45 minutes ago. And now Kathy and I are recording this. So. Yes. Drops on Thursday the 19th.
- [06:21]
Kathy McPhillips
I actually we were supposed to record yesterday. We had a little hiccup.
- [06:24]
Paul Raitzer
I wasn't going to get into that story.
- [06:25]
Kathy McPhillips
Well, I was just going to tell you that I was coming off a red eye. So I was a little bit happy that it got pushed to today.
- [06:31]
Paul Raitzer
We try. We probably don't want to get the story, but. No, we don't. We tried to thread the needle and record this. We had a perfect plan to hook it up at a production studio. And sometimes plans just don't work as intended. And it's okay. Like it ended up being working out fine. And I met some amazing people because it didn't work out the way we intended.
- [06:55]
Kathy McPhillips
And here we are.
- [06:56]
Paul Raitzer
And here we are. And you're not coming off a red eye. So. All right. So the plan is we're gonna. We got about 20 questions. These are all gonna be kind of rapid fire. As long as I don't talk too much. I. I said this word. I don't think some people believe me when I say this. I literally don't know what the questions are gonna be. I have not looked at this doc until three minutes ago. I opened this document. So this is completely unscripted. It's how most of the stuff we do works. Kathy coordinates everything, curates all the questions, organizes them, and then we get on and we just go. Because that's how it happens during the live class. And so I kind of prefer that feeling of just like, this is what it is. If I don't know an answer, I'll. I'll move. We'll move on. But we try and just kind of be as authentic as possible with these things.
- [07:39]
Kathy McPhillips
Yeah. So just to tell you again, so we did the class last week. Claire took all of the Q and A and the transcript. She ran it through some GPT she did, worked her magic. She gave me the list of the 20 questions that she thought best aligned with, you know, what people were asking how to make this flow with Paul, I went through, did a little bit of tweaking and then so Claire and I bounced back and forth a little bit. But again, behind the scenes, Claire did this heavy lift and it was her awesome idea to get these started. So this is fun. I'm excited because like you said, Paul, sometimes I would throw those questions in our Slack community, but there were still 20 or 30 that weren't getting asked and answered. So. Okay, so this week we have five different themes. Vision and Philosophy, Emerging Tech and Agent Ecosystems, Business Strategy, Adoption and Career Impact, Trust Ethics and Responsible Use, and then Future Outlook. So I am just going to jump right in. It's exciting.
- [08:32]
Paul Raitzer
We're not going to keep this easy at 5 o' clock on a Wednesday afternoon after traveling all week. All right, let's go.
- [08:37]
Kathy McPhillips
And some of these are actually repeats of what you did answer, because I. Because they were just good ones to ask. And I thought the public should know about some of these things. So let's start with the big picture. How do you define a human first approach to AI, especially as machines begin outperforming us in most areas?
- [08:53]
Paul Raitzer
Many areas, yeah. So anybody who's following us for a while. I published something called the Responsible AI Manifesto in early 2023. And it was basically 12 principles of how to do AI responsibly within an organization. And the main thing was that it had to be human centered. Which means every decision you make, every, you know, technology you're going to integrate how you think about the future of the organization. You have to think about the impact it's going to have on people. So if all we're thinking about is efficiency and, you know, cutting costs, that's not human centered per se. So I think of what is the, you know, what is the good of the. Not just your employees, but what's the impact on customers? So yeah, you can throw a chatbot up and it might save you a bunch of money and you need three less, you know, CSMs. But is it a great experience for your customers? Is, you know, are you, are you really thinking about the impact on the people? And so that can be your technology partners, it can be, you know, your service partners, it can be your customers, your employees. So that's what we mean when we talk about human centered is like don't just throw AI at things just to do things faster. You know, think about the impact and the downstream stuff too. Just, you know, how it affects people in lots of different ways. So yeah, I mean, the obvious thing is that you connect it to jobs and we don't just want to get rid of the people and the jobs, but it's actually way more than that. It thinks about all your different stakeholders.
- [10:18]
Kathy McPhillips
Yeah, we were just actually about an hour ago, some of the team was talking about something we're working on for Academy and we were talking about different technologies and what opportunities were, and all of us were like, okay, let's start with what's the best human? What's the best experience for our customers, for the humans? Everything else we probably could figure out. But let's make sure that we are putting that, that human in the center of all of that, which is like, that's been, that should be the case for everything in your whole entire life.
- [10:41]
Paul Raitzer
So, yeah, and when we launched Macon back in 2019, the tagline I created for that event was more intellig, intelligent, more human. And following that, we actually tried to like live that tagline. And when we create strategy documents, I'll, you know, often challenge our team, like, you know, think about those two things. What is the more intelligent part of this? Like, how are we going to infuse AI to do things smarter? But what's the more human side of this? What does that open up for us? So if we use AI to drive personalization through our email outreach and things like that, does it free us up to actually go have a coffee with someone who might be able to bring 10 people to the event? So it's like, what is the thing that AI can't do that we actually enjoy doing? We enjoy that FaceTime. We enjoy meeting with people and talking to them and having me free to be able to go and spend an afternoon at event at, you know, running a workshop. Like, that's the more human stuff to us. So, yeah, it can be carried out in a lot of different ways, but I think that's a good lens. What's the more intelligent? What's the more human.
- [11:34]
Kathy McPhillips
Okay, number two, what uniquely human qualities do you believe we must preserve in an AI driven world? Kind of feeding off what you just said.
- [11:42]
Paul Raitzer
Yeah, you know, it's interesting. I, I put a, like in our sandbox for the episode, the weekly episode next week with Mike. I've been having a lot of thoughts about this one lately, and I'm not sure they're fully baked yet. But I will say, you know, up front, like, I, more and more I just really look at the value of critical thinking. The easier it is to have the AI do the thing. I can see it already happening with myself. I can sometimes see it in our organization. I can see it in schools that I talk to. I can see it in, you know, enterprises that we consult with or have in workshops. It's like hitting the easy button. And sometimes when you hit the easy button, you don't have as much at stake in the output and you're not as, like, bought into the process of the learning that went into creating that output. And so, like, I guess the way I've been thinking about this, and again, I'm. This is totally off top of my head because I wasn't really ready to talk about this yet. But it's kind of like in high school, I remember you would have a reading assignment and it's like, God, I didn't read Tom Sawyer or whatever the book was. So I just go get the Cliff Notes. And you read the Cliff Notes and like, you think you're good to take the test and you get, get in there and realize, like, I actually don't know, like, the details of this book. And I kind of feel like that's what AI strategies and deep research projects have become for me. Like, I can just hit the easy button. I can create the 34 page document, but I didn't do anything to create the document. And like, all that energy that goes in and the research and the thinking that goes into creating it, like, yes, the doc may be great, maybe better than anybody else could have done in the company, but I didn't do the hard work. And like, I can't actually stand behind the document because I, I don't even really know the ins and outs of it. I just know it was good and I approved it. And so I think that this idea of, like, critical thinking, I think things like empathy and interpersonal communication and like, you know, all those things are going to matter. But it's the critical thinking part I'm really worried about. Like, I don't, I don't know how to preserve that when everything can just be created by hitting the button. And so I find myself thinking a lot about that. I think about, you know, imagination is uniquely human still. I, you know, I think. And so I think creativity and imagination and empathy and critical thinking, like, they're all going to matter. It's just like a moving target for me, like, how we preserve them and how we actually truly use AI, amplify them and not replace them.
- [14:07]
Kathy McPhillips
And we talked about this in the past before about, like, Mike uses AI to get ready for the podcast, but if he doesn't read those articles, if you don't read those articles and you just use AI to generate questions or to write the transcript to talk about the beginning of it, you can't have a good conversation about that because you don't understand, you don't really know fully what you're talking about.
- [14:26]
Paul Raitzer
Correct. Yeah. And that's why, like, for the podcast, I mean, we'll go through 40 to 50 sources that make the cut of the 150 to 200 things that I listen to or read every week. And yeah, like, I couldn't sit there and ask unscripted or give unscripted answers to the things Mike asks or presents if, like, I haven't actually consumed the information. So I can't just throw something in and hit summarize in Notebook LM and be like reading off of a study guide, basically. So, yeah, you can't fake expertise and thought leadership in my opinion. It becomes really obvious if you are. And the, the thing I've said, and I've said this to my own kids, is like, if you're going to do the work on a topic, I want you without notes in front of you, be able to stand up there and answer questions for 15 to 30 minutes about that topic. And if you can't do that, then you didn't do the right amount of work. And I'm not saying you have to be like, debate prep, like, ready to like debate somebody on topic, but if I can't take the notes away from you and have you explained to me the premise of what you did the research on? If you can't do that Then you relied too much on the AI, and in some instances that's fine, but not if you want to be a thought leader on something or if you actually want to be trusted or if you want to charge people money to, like, provide them advice and recommendations and insights. Like, you better put the work in and AI can't replace that. Like, I just don't see it. It can synthesize it or it can, like, simulate it, but it can't replace your ability to stand there and unscripted answer questions about something.
- [15:56]
Kathy McPhillips
All right, good answer. Okay, number three, we are hearing more about AGI. Where do you think we stand today? And how close are OpenAI, Anthropic, Google and Meta to making it real?
- [16:07]
Paul Raitzer
So if any of them could agree on what AGI is, I think they would all agree we're probably pretty close. They all, Even internally, like, OpenAI looks at it differently. I've talked about this recently on the podcast. Like, Sam is Altman is giving different definitions than what the OpenAI website gives. Like, it's just this moving target. But if we're talking about general intelligence that's roughly able to do what an average human can do, like the majority, what an average human can do, and we say, give me like a marketer and then say, okay, a marketer's job. There's, here's the 35 things that marketer does. I, I, I don't, I don't know that we're that far from being able to say, when you look at individual tasks, that the AI is often probably better than the average marketer at doing each of those things. Writing subject lines, drafting an email, writing a proposal. Proposal. Creating a blog post, Developing social shares, creating an image, creating a video, like, it's probably on par. Chat, GBT on its own is probably on par with an average marketer at the vast majority of those things. Now, that's not uniform across every industry, every profession, but if that's the definition, which is the one I generally look at, because I think of replacement value, well, if the AI is able to do what the average employee can do, then we're, we've kind of approached the thing we always thought was AGI before we started moving the goalposts. So I think that they all think we're really close. I, I think that whatever they define it as, it's probably sometime in this next, you know, two to five years. I think five is unlikely, would take that long, but I think probably two to three years is very realistic. I just don't know when they're going to think that they've achieved the benchmark that lets them claim it. But I would not be surprised at all if one of the labs in the next 12 months claims they've, they've done it.
- [17:53]
Kathy McPhillips
Okay, number four, if AI becomes smarter, faster and more accessible to all, how do individuals or companies stand out? Or is it just about being early?
- [18:05]
Paul Raitzer
So this kind of ties back to the one on individuals that I've been thinking a lot about. So there's in, in AI research, I don't know if it's carried out in other professions, but in AI research there's something called taste. So taste in AI research means you can go a lot of paths with how you try and make these models smarter, the algorithms you build, the systems you put in place. And taste is like your choice in which thing to go on, based partially on instinct, partially on experience. I would imagine this probably plays out in the arts as well. There's just the taste you have graphic design. You just know something when you see it. Can I have this instinct like I'm going to go after this that I think becomes even more valuable when everyone can kind of hit that easy button and create anything that the people who have the ability to look at the output look at a deep research and say, this is actually really important work. We need to spend 10, 15 hours vetting this thing. We talk about the AI verification gap was like something that we talked about on a recent episode. It's this idea that you have the ability to look at something and know that it, it matters, but it's not there yet. And so you have. And that can be applied to strategy, it can be applied to creative. And the hard part is I don't know how you get that without years of experience. And so I've been thinking a lot lately about which jobs are actually going to be most impacted. We've talked a lot about like entry level jobs and we might be, there might be a question related to this later on, but we talk a lot about entry level jobs. We've, we've looked at middle management, we looked at senior level and there's sometimes an argument that the senior level maybe goes first because they cost the most and it's easiest to cut out. There's certainly an argument that it's just entry level because it's task driven and we just don't need as many people doing the tasks. There could be an argument it's middle management because they maybe haven't developed the taste yet. Like they don't know really what Great looks like yet and I'm not sure where I fall yet. Like this is again one of those things I wasn't even ready to talk about yet. But I think the way you stand out is by finding the balance between using AI and I love it for strategy and creative thinking and things like that and outlining ideas, I love it for that, but I also get overloaded by it. Like there's so much strategy you can create so quickly that it's when to use the AI, how to use the AI, how to use the output of the AI and when to just be human and like allow yourself the permission to spend five hours on something that yes, the AI could do it in three minutes. But like you gotta put in the work to know the end. Like so for my presentations, like when I do keynotes or when I create courses, AI is assistive like ideation and maybe like vet things I've developed. But I have to create all those ideas myself. Like I have to write the stuff because I could never present it otherwise. And so I think that's going to be a differentiator at an individual level. And then the same probably applies when you zoom out at a company level. It's like all of us have access to the tech but sometimes you just can't take the shortcuts and there's no blueprint yet for how to know when that when you do and don't take the shortcuts basically. And so I think the people who spend a lot of time experimenting, you start to just sort of develop an instinct for when no, an AI output isn't enough here. Like I actually want you the employee, so me as a leader, I don't want you to do this one in chat GPT first. I actually want you to spend a week on this thing because you are going to own this and and you need to know it inside and out and you'll be able to stand behind it. I don't like again these are kind of like emerging thoughts from conversations I've been having in some cases in the last like 10 days and personal experiences in the last 10 days. But I think using AI, like knowing how to use it and when is it could be a huge differentiator for people if all else is equal and we assume everybody's using it. But right now the differentiator is a whole bunch of people who have no idea what to do with it. And so for a while that's the opportunity is like just to raise the head and do this because not everybody's doing It.
- [22:12]
Kathy McPhillips
And I'm guilty of that. You know, a few weeks ago, we have so much going on right now, and it's like, okay, I gotta start tackling some big things. And I started with one of my GPTs to answer some questions for me or to give me an outline. And then I was trying to, like, retrofit what I needed it to do, and I was like, wait, I'm not doing this the right way. And I actually stopped, scrapped it, and just started over.
- [22:34]
Paul Raitzer
Yeah, Yeah. I found the other thing is, like, I'll have these random thoughts to develop a strategy for something, and I'll have the. Either conversation, voicemail, I'm driving to pick up food, or I'll ask it, oh, I'm laying in bed at night, I'll think to, like, run a deep research project. And then like, two weeks later, I'm like, God, I feel like I did this before. Like, when did I. And then, like, you completely forget that you actually did the project already because, again, you had no stake in it. You literally just gave a prompt and it did the thing. And then you kind of forget that you even went through that process. That's why, like, Kathy knows I journal everything in business. Like, anytime I run a project, it's like I have journals for each component of the business, each business unit. Because, like, sometimes you just forget you've already done some of the work. And I find myself doing that all the time with AI.
- [23:17]
Kathy McPhillips
Yeah. Okay, section two, emerging technologies, number five. Do you see a future where AI agents can collaborate, like human teams? And how important will it be to know how to build and manage those agents?
- [23:30]
Paul Raitzer
Yeah. So agents collaborating with each other is already starting to happen. That is very much going to be a part of the future of every department, every business, every industry is agents working together. Hard part there is, like, how we manage those is who knows? I mean, you're. In some cases, you'll have leaders like Jensen Wong from Nvidia saying, we're gonna have millions of agents in every business unit. Like, how could we possibly, as humans, like, manage what they're doing? We can't even keep track of it all. So, yes and yes. I. I guess, like, they, they're. They're gonna be there, they're gonna be working with each other. Humans may have involvement in the early going, as these agents are, you know, they're kind of raw still. Like, they make mistakes. They're not fully autonomous in most cases, so there's a lot more management and oversight and connecting it to the right data sources and the right tools. But over time it, it's kind of probably going to just function more like you know, you're used to with chat GPT where you just give it a prompt and then if you've connected it to Google Drive and your CRM and like it has access to all the things you have access to, then it's just going to go do things. And you might not even know if it's calling on a different agent to do a thing. So as long as you've set up the permissions where this agent is allowed to go talk to these other agents, it'll. It would function again like this can be abstract for people, but it truly would function like if I went to Kathy and said, hey Kathy, we need to do this project next week, let's meet next Friday and review it. And then Kathy goes and brings in five people on the team and they each do a piece of the thing and then it comes back. And then Kathy and I meet and she goes, hey, here we go. And Kathy and I sit there and talk like I don't know who she worked with or what part they played in it. Kathy was just like the hub, basically. She was the lead agent and it. And she went and found the components to do a thing. And so that's how it's going to work. Except it would be like access to dozens or hundreds or thousands or millions of these agents like that. That's what AI labs envision the future being. And our agents will talk to other people's agents and things just get done. So yeah, I think that a lot of jobs in knowledge work is going to be managing these agent networks. So I do think that, you know, three to five years from now it's going to be very commonplace, that it's just part of your job description to build and manage agents and agent systems. Like, I mean, we have it in our job descriptions now that. Because we're starting to think that way. But you know, it's like you're building these distinct, like one agent for this, one agent for that. We're talking over time of like almost like a marketing ops. You almost think of like an agent ops thing where like your job is literally just to be the operations behind all these agent networks that work with the marketing team or the sales team or the customer success team. Right.
- [26:23]
Kathy McPhillips
So how can someone. This is question 5A, because this is not question 6 yet. But so how, if someone is thinking about this, where do they get started with learning about AI agents?
- [26:33]
Paul Raitzer
Yeah, I mean, part of it is just Using tools like Deep research From Google and OpenAI and starting to get a sense of how these agents will work and how they'll look. Because that's an early form of it where it just kind of goes and does the project for you. All the big tech companies are selling agents already positioning it as agents. Again, it's early and they're not like fully autonomous for the most part. And humans are still pretty heavily involved in building and running these things. But I would imagine like Salesforce, Google, Microsoft, HubSpot, like any, anybody major tech companies that are building around this idea of agents, they're going to have to provide education around it like we're creating. So Kathy mentioned Academy earlier. So we're, we are, we've had an AI academy for five years now, but it's only had piloting and scaling AI and then live components and then some other benefits to members. We're reimagining and rebuilding that like as we speak. Like I'm going to be recording all the videos here in the next like you know, three to four weeks for the, for the new courses and one of the ones is like agents 101. So like we're going to do our part to soar and try and help people understand the fundamentals. But then as part of our Gen AI app series that'll be part of AI Academy, we have a agents component to that where we're actually going to start doing weekly drops with Gen AI apps of productivity and vision and images and audio and agents to try and just make this stuff more approachable to everybody because it's just abstract until you start seeing it more and more. So we'll do our part but we're going to be more focused on kind of like the macro level, understanding agents and then showing examples. But I would probably like push heavy on places like Salesforce and Google and Microsoft and say what education are they offering that can be complimentary to the kind of stuff we're going to try and provide to people and probably think.
- [28:23]
Kathy McPhillips
More about what it's able to do versus what they're calling it.
- [28:27]
Paul Raitzer
Yeah, yeah. Because agents is basically just like automations with some intelligence baked in. It's just the new term that people.
- [28:34]
Kathy McPhillips
Have and they're using the term differently.
- [28:36]
Paul Raitzer
Yes, in a lot of instances, kind of like AGI, like everybody's already got their own definition of what an agent is.
- [28:41]
Kathy McPhillips
Right. Okay, number six, for those working with sensitive data, what does it make, when does it make sense to use a local LLM over a cloud based one?
- [28:51]
Paul Raitzer
So this Is one I will, I'm not going to punt it completely and like, not answer it at all, but I will say this is one where your IT department comes in. This is why the CIO is cto. Cio, they're often involved at the higher level, what's going on, especially if you're in a bigger enterprise. This is more technical stuff at a very high level. The concept here is, do you trust chat GPT, Google, Gemini, Anthropic, Claude to have your data that, like, I want to, I want to do an analysis where we take our marketing data or our profit and loss data or customer data, and I want to, I want to have ChatGPT run an analysis on it, find insights in it. So the core of this question is, are they trustworthy to provide that data so we can use these chatbots we're used to, to help us with this stuff? That is an individual company decision. It's an individual decision if it's just you. You have to look at the terms of use. You have to be comfortable with how secure your data is. It may be something you want to bring your attorney in to make sure you're fully understanding the terms of use and what the rights they have to your data and the different things you put in. In enterprises that have more sensitive data are more highly regulated. That is an instance where people may make that choice to build an LLM that can be on premise and that doesn't live in the cloud, and then you don't have as much concern. But again, it's hard to give one broad answer here knowing everybody's got different situations with their data. This does come up all the time, though. One of the questions I get the most is, is it safe to put my data into ChatGPT? Like, I want to use their data analysis, but, like, I'm not sure I'm comfortable giving it everything. And again, I think it's like a personal preference thing at this point as well as, you know, understanding the, the, the guardrails that your company provides about whether or not you should do that. Right.
- [30:52]
Kathy McPhillips
Okay, number seven, you answered this yesterday on the podcast, episode 153. But either you can do a Cliff's Notes, Cliff's Note version, or you can go expand a little bit. But what's the difference between a chat GPT, projects and custom GPTs and how do you decide which is better for a given task?
- [31:07]
Paul Raitzer
Yeah, so I did explain as best I could on episode 153. The gist of it is based on my current understanding because again, I'm still trying to make sure we're providing the best guidance here, but I looked into it. I use custom GPTs all the time. I do use projects. I think of projects as folders. Like if you have Google Drive or OneDrive, Microsoft, whatever, Dropbox Box, whatever, whatever your system is, you have folders. And in those folders you can put images and videos and, and chats and whatever, and they all live there. And you can kind of keep everything organized. So that's how I think of projects. Custom GPTs is I have distinct tasks or projects that I probably. We shouldn't use the word projects here. Let's say distinct tasks or workflows that I want to train a specific instance of ChatGPT to do, and then I might want to actually share that with my team or with the public. So we have Jobs GPT that helps people assess the impact of AI on their job. You can put your job title in, it'll break it down into tasks. And, and that's a publicly available free GPT. To my understanding, that is not something I can do in projects. Like, if I was doing projects, I can't share out a single GPT. So I think of GPTs as like, things we want to do that are distinct tasks. And sometimes we share them with our team, sometimes I keep them for my personal use, and sometimes I put them out in the public. Projects is a foldering system basically to keep everything organized.
- [32:36]
Kathy McPhillips
Well, that segues great into number eight. So almost like you planned this. Actually, I didn't Plan this, but ChatGPT must have known how you're going to answer that. Okay, number eight, if an agency or consultant, I guess even us, is managing dozens of GPTs, what are your best tips for organizing workflows, versioning, and staying sane at scale?
- [32:56]
Paul Raitzer
Yeah, this is a good one. I'm starting to feel this pain ourselves. We, we have been very aggressively been building out GPTs as an organization. Everybody has that ability in our company. And, uh, people have been way more proactive, I would say, in terms of creating GPTs or different processes and workflows. Um, we don't have like a structured naming convention for ours. They, you know, they're available to people within our team license, but we don't, to my knowledge, I mean, you know, Kathy, like, we don't have a Google sheet that tracks all of these things that just kind of in there. And as I'm saying this, I'm thinking like, maybe we need some better system than we currently have. For me, the ones I build and manage I journal again, like you'll sense a trend here. So I have a custom GPT Google Doc and all of the ones I build. I'll go in and say, problems. GPT made these five updates. Here's the system instructions, here's the clean version, here's the edited version from the prior version. So like I track GPTs the same way I would if I was building an actual app or product, which we have done some before. And so that's how I do mine. So I can always go back and see what happened. But like I don't, I don't know the team has access to that doc, even like I don't. I guess sometimes I'll say like, here's what I did not maybe show it to them. But that's not a uniform structure we have internally. So I would say I, I think of this as probably like pro, you know, project management style thing that maybe needs more structure. Um, maybe like prompt libraries is a good reference. If you've been trying to structure your prompts for sharing with your team, maybe it's following a similar flow. But I would imagine this probably fits into however your company manages projects.
- [34:45]
Kathy McPhillips
Maybe as these evolve these companies will figure out better systems for organization for, for their users.
- [34:51]
Paul Raitzer
Yeah, it'd be nice to be able to put the GPTs into different foldering systems for the company. Like if you're looking for customer success GPTs, here they are, here's number of usage, things like that. But yeah, unfortunately OpenAI has provided very minimal support to GPTs since they launched and they made a big deal out of it like it was going to be the next app store like Apple and then they just didn't do anything for them. And you know, every once in a while there's some little feature added. Like last week we got the ability to choose which model you would recommend. So now you can use any model within the GPTs, which actually probably did more to break them than anything because these things weren't written to be reasoning use reasoning models. And now all of a sudden a user can pick a reasoning model and it's going to like break the way the thing works. So yeah, unfortunately they just haven't put as much energy behind GPTs. But hopefully they, they do provide some ways to better organize it. Right now you're kind of on your own in Google Sheets or Excel or Asana or however you manage these things coming up with the system. I'll, I'll have to think about it more because that is, it's a, it's A good question. Something I honestly haven't really thought about or developed a system for our company to do.
- [35:56]
Kathy McPhillips
I mean, I've thought about it as, you know, team members are building things and I just need to remember to go back and look at what they've done, remember where they put it.
- [36:04]
Paul Raitzer
So I have had that where it's like, hey, Mike, didn't you build like a prompt generator or something? Like you're just kind of like. I feel like some point. I saw that somewhere. Yeah.
- [36:13]
Kathy McPhillips
Okay. Number nine, there's so much buzz about ChatGPT versus Gemini. How do you personally decide which tools to use and do you see a winner emerging?
- [36:20]
Paul Raitzer
I think the winner is just going to change every three to six months. I don't know. We're going to have a situation where like GPT4 from OpenAI was just the dominant model for a year and a half. Like it was far and away just the best model. I don't, I don't know that we're going to enter that phase again. Like, I think they're not like fully commoditized per se, but the models are so close in their abilities that it's hard to go wrong right now. Like, the difference with Gemini is 2.5 pro, which just yesterday I think went to general availability in Google Gemini. So if you have The Gemini app, 2.25 Pro is now generally available, I think in all Gemini accounts. And that model is both a traditional chat bot and a reasoning model combined like one unified model. ChatGPT is not. So they have a reasoning model which is O3 and O3 Pro, and then they have their traditional chat model. So I actually posted something about this on LinkedIn this week and we talked about it on episode 153. When we talk about the O3 Pro model, I use both. So I actually. So I think I, I might have said this on the 153, but right now in the building of AI Academy, I created a teaching assistant gem, like a Google Gem. And I created the same using the same instructions in a custom GPT. And, and so oftentimes I will actually put it into bold. I will say, like, here's how I'm going to describe a course. This is the course template I'm using for the description that'll appear in the learning management system. What do you think? Evaluate this in a critical way and I will give the same prompt, the same output to both systems and see what they do. So if it's a high value thing, I will just use both. And then sometimes you realize like, okay, Gemini is just better at this use case that I do all the time. So I'll use it and then once while check in with ChatGPT, see if we got any better. So if you can afford Both, I mean 20 bucks a month, like for the value you get from them, if you use them enough, there's, it's pretty good argument to just pay the 20 bucks a month for both and try them. But I also don't think you can go wrong with just picking one and using it all the time. Like just there's, there's some value there to just experiment and get really good at, at talking to one of them. So I don't know that there's a right answer here, honestly. If you can afford both and you have the capacity to be testing both, go for it. Worst case scenario, just pick one and spend a lot of time with it, experimenting with it and getting good at prompting.
- [38:54]
Kathy McPhillips
Right. Okay, number 10. What tools or platforms in the agent agent space like HubSpot, Salesforce or Chatbot integrations are actually ready for production today?
- [39:08]
Paul Raitzer
So we, I don't have personal experience with Salesforce. They introduced Agent Force fall of 2024 so it's still pretty fresh. It's like anything else. Like sometimes you get feedback that it's just a bunch of marketing and branding and there's really nothing to it. And sometimes you hear stories of no, it actually works. It's great. We've got, you know, these agents set up HubSpot builds on top of chat GPT or you know, GPT technology from OpenAI so they're starting to enable things. Like they just did a connection with deep research from Chat GPT so you can actually connect at the HubSpot and then that's agentic in a way. So it's able to go and look at your CRM data and provide reports. Our first experience with it is it just didn't work great. I've again I've heard awesome stories and I've heard things like our experience where it's like it just doesn't work. It like takes forever and it returns nothing of use. So I think just generally speaking, agents are just really early in terms of their reliability. I think there's the marketing from these companies haven't done them, the product team much favors in over promising like what these things do. I think there's a lot of early efforts made that to make them appear more autonomous than they are that, you know, you just thought you'd hit the button and it just Went and did the thing and it was great. So I don't know, I mean, everybody's kind of playing in this space. Even Nvidia, you know, is starting to move into this space, not only enabling other people to build it, but building their own things. I think it's probably going to be, you know, six to 12 months before a lot of the early stuff that's just not delivering on the promise right now truly starts to. But that, that's a very broad statement. I'm sure that there's lots of people, even if you're setting it up through like Zapier or make, where you're kind of building an agentic process and the human's pretty involved maybe in establishing what the workflow looks like, there's a lot of those that are working. So if we think of AI agents on this spectrum of autonomy, I would say that there's probably a lot of early stuff where humans are pretty heavily involved in writing some rules that are working great. If we're thinking about, I'm just going to go and get an email agent and it's going to take 80% of the work off of my team and they can go focus on this other stuff. I don't think that that's the reality for the vast majority of use cases that you would look at applying agents to today. You can go get a good sense of agent AI. So that's Dharmesh Shaw, one of the co founders of HubSpot. He has agent AI that's like a, almost like a social, not a social network, but like a marketplace for agents. And you can go see the kind of things that are being built and what you'll see is they're very distinct tasks like that Most of the agents are, are kind of still being built to do these very specific things. More hype than reality, I guess, is the short. Too long. Don't read. It's. It's probably more hype than anything at this point, but it's going to change real fast. And I wouldn't ignore it because of that.
- [42:07]
Kathy McPhillips
Yeah. Incidentally, I saw the back end of a workflow, of a make integration.
- [42:13]
Paul Raitzer
Yeah.
- [42:13]
Kathy McPhillips
I was just like, what in the world? And I'm so glad we had a human to help us do that because we knew what we wanted. But just seeing all that logic and the branching and everything, it was just like, wow. Yeah, we really need to understand that.
- [42:26]
Paul Raitzer
Yeah. I think that a lot of the agentic stuff today does start with understanding the actual workflow that needs to run and then finding the ways to integrate the agentic processes into those workflows. But it usually requires the human to first envision the workflow in some ways. Now again, there's exceptions to this like deep research from OpenAI and Google. You just say I would like this research project and it builds its research plan and it goes and does it and there's, that is agents at work. So yeah, it's a, it's a mixed bag. But again, I would say more hype than reality at the moment, but moving pretty quickly in the opposite direction. Yeah.
- [43:08]
Kathy McPhillips
Okay, our third section, business strategy, adoption and career impact. Question 11 for companies just getting started, how do you recommend they identify the right pain points and build their AI roadmap?
- [43:19]
Paul Raitzer
So we have a custom GPT called problems GPT. You can go to SmartRx, AI, we'll, we'll put this in the show notes and click on tools and the custom GPTs that we've created for this kind of stuff are right there. And Problems GPT is a free custom GPT. I actually was just showing this in the workshop I was running for Movable Link. What Problems GPT does is it helps you identify those pain points points like what are our problems? And then it helps you write problem statements and value statements and then it'll actually develop a strategic brief to help you solve in problems more intelligently. So identifying the right pain points is basically the same. It's always been what are your goals in the company? What is, what are the KPIs you're responsible for? Which ones aren't you meeting like pain points of pain points like so I don't know that that changes. What changes is the more you understand what AI is capable of, you look at how to solve those pain points and problems differently. And so that's what I built Problems GPT for was to help people identify and properly state their problems and then assign values to them and then try and prioritize which ones can be solved more intelligently with AI. And so when we talk about an AI roadmap, we think about all of the kind of smaller level projects you may be running. Like we often talk about like pilot projects that you're running. So okay, we're going to apply it to email, we're going to apply it to social, we're going to apply it to media buying or we're going to apply it to data analysis and whatever and you've got, you know, go find a tech and do these things and then the roadmap layers in and here's like the Five fundamental business problems we want to solve over the next 12 months. And then you probably have a third layer which is, and here's the innovation layer, here's the new stuff we're going to go do that we weren't doing before. And so the best AI roadmaps solve for efficiency and productivity immediately through these distinct projects. And then you're thinking about the higher value stuff through problem solving and innovation that actually drives the growth of the company and hopefully prevents you from having to lay people off because the efficiency is going to make it so you need fewer people doing that work. And then problem solving and innovation make it so you can redistribute the talent into these other areas that drive the growth and innovation.
- [45:35]
Kathy McPhillips
Right. Number 12, what AI tools do you believe delivered the most value to marketing leaders?
- [45:40]
Paul Raitzer
Right now this could vary. Like Kathy, you might say descript. I, I don't know, like you could answer this one as well. But I think just a chatbot like it using Gemini or Chat GPT well every day and building gems and custom G like that is for most organizations, most marketing teams in particular, that's enough. Now you might go get like a writer or Jasper that's specifically built for marketing as well. But at minimum you just go hard on one chatbot and you, you integrate it into the work. But would you answer that one differently, Kathy?
- [46:14]
Kathy McPhillips
I wouldn't. I mean, we have specific use cases for some specific tools, but 90% of my AI use is within chat GPT.
- [46:21]
Paul Raitzer
Yeah.
- [46:23]
Kathy McPhillips
Okay. Number 13, how is AI forcing agencies and consultants to rethink their models, especially with rising efficiency and lower costs?
- [46:30]
Paul Raitzer
This is a dynamic space. So if people again are kind of new to our ecosystem and what we do, we have an AI for agency summit. I owned an agency for 16 years. My first book was the Marketing Agency Blueprint. So I've sort of lived in this space for a really long time. It's a challenging time to be an agency, to be a consultant. I think you're under tremendous pressure because if you're using generative AI, which you should be, your clients are increasingly aware of that and that you're probably doing things more efficiently. So if you were using or still using some form of billable hours, that's a little tricky because you have to do a lot more work to make the same amount of money. If you're charging by the hour, if you're in a value based model where you're charging based on value creation. And I again, as I'm saying this, like flip the script if you're not an agency and you're on the brand side and maybe you pay agencies or consultants, freelancers. It's just a very up in the air space of how it's going to all play out. You could also get into the issue of if you're using genitive AI, are you passing copyrights over for the creative work you do for the outputs? The answer is no, you're not. Because as of right now, at least in the United States, the copyright law is if AI creates it, no one owns a copyright to it. So you're not passing a copyright to your client. Client may not know that. I have seen contracts from larger enterprises that outlaw their agencies from using generative AI unless they get specific permission. It is a total reinvention of the agency model and I'm not even trying to oversell this. Like over the next couple years the agency model is going to have to be completely reimagined. We're seeing some of the big agencies trying to do this. It's really hard to shift and stay stable financially while you're trying to reinvent this. It's probably a great time to start an agency or consultancy because you can do stuff that, I mean, honestly, I said it before, like at my peak, my agency I think was around 20 people. We, based on the way we do work now as an organization, we are more productive than that agency by far and probably on par with what a 50 to 80 person agency would have done back then. So I think that it's just so much easier to build and scale a professional service firm right now. It's a hard position to be in, to be an established one that's having to try and reinvent this. So AI native starting from the ground up is a way easier play than being an AI emergent where you've got all this traditional stuff. You may have a bunch of people, especially creatives, who don't want anything to do with AI or don't want to use it. And it's going to be a challenging change management process at a lot of agencies. I've seen some doing it well, but it's going to be hard.
- [49:26]
Kathy McPhillips
And there are so many people at agencies that want to figure this out. People at Macon are like, just tell me what to do, tell me what should I be thinking about.
- [49:35]
Paul Raitzer
Yeah, we have a huge, I mean our community, we have probably 110,000 plus subscribers at the Institute and there's a fair portion in the 20% range or something that, that are probably in that agency consulting umbrella. And so These, yeah, these are people we talk to all the time and, and we see the people doing great work and that are evolving and we do start to see a lot of people who just jump ship and like start their own thing and they can be, you know, one person can do the work of 10, basically. And so you, you see those kind of people having more kind of freedom to build their future. So yeah, great time to be building an AI native firm or consultancy. Tough time to be trying to steer the ship to, to build an AI emergent one from an existing traditional agency.
- [50:24]
Kathy McPhillips
Yeah. A little plug for our Slack community. We just hit 10, 10, 000 members this week.
- [50:28]
Paul Raitzer
Nice.
- [50:29]
Kathy McPhillips
And we have an agency channel within there that is very active with all these agencies trying to support each other, offer best practices, figuring this out together. So if you're an agency and looking for some support, come join us.
- [50:41]
Paul Raitzer
Yeah, the other thing I would add to this is like HubSpot. So that was how, you know, I came up as the first HubSpot partner back in 2007. And HubSpot's been doing an incredible job of helping to try and guide their partners. They have ecosystem partners, not just traditional marketing agencies, but you know, full blown solutions partners. And they're, they're doing great work trying to actually help level up those partners to help them make these kinds of shifts. And so, you know, if you are an agency, look for those kinds of partners who are invested in your future as well. It's cool to see what they've been doing with their partner ecosystem.
- [51:15]
Kathy McPhillips
Yeah, absolutely. Okay, number 14, what does great prompting actually look like and how should employers think about evaluating that skill and job candidates?
- [51:25]
Paul Raitzer
So great prompting and again, I'm top of mind. I'm building a prompting one on one course right now for Academy. The simplest way I explain this though is like just pretend like you're giving a project to an associate or an intern. Like how would you do that? So if you're asking ChatGPT, you know, if you're not treating as an advisor, if you're doing as like you want it to help you with an output. The way you would talk to an intern is listen like here's the project I want you to do, here's why you're doing it. This is the goal of the project. Here's five examples to look at and make sure you don't do like these couple things but like this is what we want out of it. So you just describe it. And so the easiest way to actually prompt is just talk to it. Like, you would talk to a person, and then from an advisor perspective, you flip it a little bit and you say, listen, I want you to function as my cfo. Like, I'm trying to understand the ins and outs of this, and I'm not an expert in finance. Like, help me understand this. Or I want you to function as an attorney and I want you to think critically from a legal perspective about the thing I'm trying to solve for that. So that's the difference is, like, just talk to it like what you want it to do, what the output needs to look like, and then if it's the opposite and you. It's a function as an advisor, then tell it you want it to function in that role. And here's what you're trying to solve for. And honestly, like, if all else fails, say, I'm not sure how, how to ask you this. Here's what I'm trying to do. Like, I, I say this in workshops all the time. People come up like, what should I do here I was like, what you just asked me. Ask AI. Like, you're, you just phrased it perfectly. You have a problem. You're not sure what to do. You don't know how to use AI to help you literally give the prompt that you just asked me. So sometimes just imagine you're talking to, like, a consultant or someone you know has the knowledge you need. How would you phrase it to them? So there are formulas you can follow and, like, do these five things. And like, that can work too. And we teach that. But if all else fails and you're just not sure, just talk to it like you would a human that you're seeking the knowledge or the output from.
- [53:22]
Kathy McPhillips
Yeah. And then once you get through a couple of those and you realize, okay, this is what I need to include in the beginning versus trying to, you know, do it 10 times. You just, you. You'll get better at it.
- [53:30]
Paul Raitzer
Yeah. And honestly, the AIs are being trained to get better and better at asking you qualifying questions, like making sure that they know exactly what you're trying to do. Say, hey, well, I can help you with that, but I would really need these five things. And what I'll do then is, like, I'll just answer one. I. What I'll say is, I'm going to give you answers one at a time, like, wait till I give you all the answers before you go and do the thing I want you to do, and then you just do it. And sometimes I'll actually keep a separate Google Doc and I'LL just like look at the five questions and I'll just write the answer fully and then I'll like throw it back in as a single answer. But yeah, I mean it's just, and it, and the biggest part is just experiment. Like you learn how to talk to them. It's, it's almost like as the other analogy is, if you have ever raised a kid and it's like when they're four or five and you're just trying to figure out how else can I say this to get through to you? Like I just, we need to figure out how to get you to do this thing. And sometimes it's like talking to a kid. Like you just gotta figure out how to say it. So it actually does the thing you want it to do or doesn't do the thing you don't want it to do, which I've definitely gone through. Or just keeps outputting something the wrong way and you're like, stop, what are you doing? And then you just have to try and rephrase it. It's like, okay, let me come at this a totally different way. So yeah, it's very much like a kid.
- [54:40]
Kathy McPhillips
Yeah, okay, 15. As AI reshapes roles, does age or experience become a liability? Or can being the most informed person in the room still win out?
- [54:50]
Paul Raitzer
So this one goes back to what I was saying earlier and I'm not sure yet, like I have to play this out a little bit more in my head, but there is a big part of me right now, I should think about this more before I say this. Okay, so I could be totally wrong here. I think middle management's screwed. I think the people that lose out the most in the interim are not the entry level because you can bring them in and they're cheaper and you can teach them and they bring a nativeness to this where like they've, they're just familiar with these things and you don't have to teach them new stuff. Like they just come out ready to work with these things. So like entry level is still super valuable and you can pay the entry level more than you normally would have because they're going to out produce their peers and they're going to, you know, produce it like 2, 3, 5x what they used to. You need the senior level because they actually have the experience to evaluate the models. They know what to ask, they know the right question questions to put in. They have some institutional knowledge and I think the middle management might be stuck in this position where they don't have that yet. They don't have all the critical thinking they need. They don't have all the ways to like, know if the outputs is good. But I don't, I don't know. Like, again, I'm literally thinking out loud here. But if I, if I even look at a microcosm of like our organization or like some of the companies I've recently talked to, the senior people need to be there, like you. You can't just get rid of them. If you, and if you don't have the entry level people, then, like, who are the future leaders? So I, I don't know. That's. Those are kind of like, I could be wrong and I could change my mind next week when I start talking about this more and I've had more time to think about it.
- [56:48]
Kathy McPhillips
This is live, folks. We are, we're on the spot.
- [56:52]
Paul Raitzer
Yeah.
- [56:52]
Kathy McPhillips
Okay. Number 16. What kind of changes should leaders expect in the workplace culture as AI adoption grows?
- [57:01]
Paul Raitzer
This is going to depend a lot on your organization. I could see there's going to be a lot of clashes. I think pretty soon in some industries and in society. I think there's going to start to be quite a bit of pushback against AI. And so there's a possibility that if you have cultures that don't want to change or that become so fearful of their jobs that there's actually pushback to AI adoption and resistance to it. If you're in a more innovative culture that welcomes change and is used to it, then it's probably going to go really smoothly. So I, I don't know. I think that the culture you have, the level of transparency and honesty from leadership, the willingness to invest in your talent and help them improve their careers. So if you have internal professional development programs, if you have a history of creating a workplace that's conducive to them, advancing their careers probably goes really well. If it's a very traditionalist organization that doesn't handle change well and has been through some of these digital transformations over the last 20 years and it was kind of painful. It probably isn't great, but I think it really comes down to leadership and their vision, their willingness to execute that vision, and then their honesty of having to go through that change. Because like, we saw. We'll talk about this on the next episode. But like Andy Jassy for the CEO of Amazon literally just put out a memo to his team yesterday. He's like, we're going to have fewer people. Like, we're just straight up, AI is going to drive efficiencies we will have a smaller workforce in the future. So that's part of it. It's like, okay, we've got the transparency part. We're at least admitting this is what's going to happen now, how you actually execute that and what that looks like to people. That's where the culture part comes in is like, what does this actually mean? Does it hurt our recruiting efforts if we're literally saying we're going to start getting rid of people? I don't know. And so I think culture becomes critical and I think the way you handle AI and whether you take a human first approach to it starts to really matter in your ability to recruit and retain people in that profession well, and.
- [59:15]
Kathy McPhillips
Even take out job replacement. Just think about people within the organization. Some love it, some don't love it. But like that honesty, that knowledge sharing. Look what I did, look what I learned. I want to show you something. Like just having that collaboration I think is really important.
- [59:29]
Paul Raitzer
Yeah, we see it with, in ours, we tried, you know, huge on the knowledge sharing side and we, we want it to be inspirational to people. But you also have to be, you have to know where that line is. Like, at some point are like, oh man, are we becoming too automated? Are we relying on the AI too much? And I honestly, like, I already kind of feel that sometimes I feel it myself. Like, sometimes I'm just like, yeah, I got to do the hard thing now. Like, I can't use AI for this thing. But I think strong cultures stay strong. Like, I think that, you know, again, I'll go back to like a company like HubSpot. I just knew their culture intimately for a long time when I was a partner. And it was always just a great place and had a great culture. And I think that if you trust your leaders and those leaders are transparent and open, then it could be good. But bad cultures, it's going to probably get amplified if you have a bad culture. And the other thing is, the problem you might run into is if overall the organization is not pushing, is not an AI forward organization, but you have AI Forward individuals within that organization that are trying to push for change that can go bad real fast.
- [60:42]
Kathy McPhillips
Yeah.
- [60:43]
Paul Raitzer
And those people are not going to stay there. They're going to go find a place that like, embraces their ability to be AI forward.
- [60:50]
Kathy McPhillips
We have four questions left and it's at the top of the hour, so let's rapid fire.
- [60:54]
Paul Raitzer
Okay.
- [60:54]
Kathy McPhillips
Trust, ethics and responsible use. Number 17. What is ChatGPT really storing in its memory? And how persistent is user data across.
- [61:02]
Paul Raitzer
Sessions, I would assume it's storing everything, unless you've told it not to. They the labs see memory as a fundamental element of achieving AGI and having a very sticky experience with ChatGPT. So you don't leave and go to Gemini. So if it knows you and everything about you, your preferences, your interests, your buying history, like they want to know everything, everything in your calendar, everything in your email, everything in your Google Drive, everything in your photos, the more they know, the more personalized the experience can become. So I would just assume that today it doesn't remember everything. It's not a perfect memory, but assume that's where they want it to go. And so how persistent user data is across sessions varies depending on the chatbot you're using. But again, I would just assume in the next couple years it's going to feel almost perfect, like it just remembers everything. And they have to. It's tricky to like, kind of figure out how to manage all those memories, but they're going to spend a lot of resources to solve memory. It is, like I said, fundamental to where these models are going.
- [62:11]
Kathy McPhillips
Right. Okay, number 18. How can businesses, especially in regulated industries, safely use LLMs while protecting personal and proprietary information?
- [62:20]
Paul Raitzer
Get this one all the time. So the first thing is safely using LLMs. If you're having trouble getting approval to do it, so you're having trouble getting chatgpt or Copilot or whatever it may be internally, steer into the concerns that the different stakeholders have about the use of those tools and find a bunch of use cases that are not impacted by that, so that don't require the personal information and things like that. The other is this is where you lean heavily on legal and IT to make sure you're doing everything safely.
- [62:56]
Kathy McPhillips
Okay, number 19, why do you think some companies still ban AI tools internally? And what will it take for those policies to shift?
- [63:03]
Paul Raitzer
Lots of risk and uncertainty. So it's logical to ban things that you don't understand or that you think have a higher risk. And so in some cases, banning is because the people making the decisions don't fully understand and don't realize that there's probably a bunch of use cases that don't cause risk and concern, so it's just easier to ban them. But you know, when we think about things like agents that you're going to give access to your computer to and access to company data, like there's all kinds of risks, including things you can't even fathom that are being considered, like data poisoning and prompt injection and like all these emerging research areas that it sees this stuff and it's like, whoa, whoa, hold on. Let's pump the brakes. Let's hold off on rolling things out. So, yeah, sometimes you just have to trust that the information security people, cybersecurity people, like, there's a reason why they're paid to manage the risk of a company, and you have to understand that and you have to be empathetic to that, that, like, everybody's trying to do their jobs here. And sometimes your job is to find the simpler use cases that can create value that don't cause these concerns or come up against them.
- [64:13]
Kathy McPhillips
Yep. Okay, number 20, that's shut this thing down. Okay, if AI tools are free or low cost, does that make us the product, or is there a more optimistic future where creators and users both win?
- [64:27]
Paul Raitzer
So that rule's generally pretty reliably true. So, yeah, if you're not paying for something, there's a pretty good chance your data is the product. That's the thing that they want access to. So, like, Facebook could be an example here. Everything you have ever put up there is basically being used to train their models. Now, you know, you could think of the same thing with like a Gmail or photo. It's like, yeah, the data is the product. And the data became more valuable because now it can train models that they think can generate billions of dollars in revenue and value every year. Tens of billions. Hundreds of billions. Trillions, potentially. So, yes, it's pretty safe. And I would say, like, just kind of a bigger picture to end with. I would just be really, really cautious of experimenting with a bunch of AI tools where you have to give it any data. Pictures of yourself as an example. If you don't know the company, you don't know who funds the company. You don't even know the founders are, what country it was built in, where your data is going. I just generally take a very cautious approach to the using of the tools and the connecting of any of those tools to any meaningful data source, because you just don't know. And it's often better. Now, I know that there's plenty of people who are pretty free with their data and just assume everybody's got it anyway. And I. I get that, too. I think it's going to be a generational thing. I think the next generation is going to be less and less, you know, cognizant of where their data is going. But generally speaking, I think it's good to just take a cautious approach to who you're giving your data to, what data it is that you're giving to them and you know, you got to find the companies you trust. And that's why in a company situation, I often say, like, start with the companies that are already through procurement, that are already approved in your tech stack. See what AI they have before you go. Try and like patch together a bunch of other tools that you might not trust or even be able to get through procurement.
- [66:31]
Kathy McPhillips
Absolutely. All right, that's 20 questions.
- [66:34]
Paul Raitzer
Okay, I went fast. About an hour. All right, well, thank, thank you everyone for the questions. Again. This is. These were from our intro AI class. Do you, off the top of your head, Kathy, know when the next intro to AI class is? We could, we probably have the ability to look July. We will put it in the show notes. July something the 9th or something like that. Maybe we just scheduled it. So the. That is coming up. I'll look it up right now. It is July 9th. Wow, look at that. Okay, Wednesday, July 9th at noon Eastern Time is the next Intro to AI class. So we do about 30, 35 minutes of presenting and then we do the ask me anything for 25 minutes and then same deal. Whatever doesn't get asked there, we'll kind of curate that and do another AI answer session. And then the other one I mentioned is scaling. That one is coming up the day this drops, so you might miss that one June 19th, and then we'll announce a July session for that as well. So again, every month intro and scaling happens and we appreciate the tens of thousands of people who have joined us in those classes and we plan on keeping them going. So thanks to everyone there. And Kathy, any final notes on this episode?
- [67:42]
Kathy McPhillips
See you at Macon.
- [67:43]
Paul Raitzer
Yeah, Macon. There we go. And Macon. M A I C O N AI. All right, thanks everyone. Thanks, Kathy and thanks, Claire, for helping put it all together. Thanks for listening to AI answers. To keep learning, visit SmarterX AI where you'll find on demand courses, upcoming classes, and practical resources to guide your AI journey. And if you've got a question for a future episode, we'd love to hear it. That's it for now. Continue exploring and keep asking great questions about AI.