Marketing AI Institute | Blog

[The Marketing AI Show Episode 50]: Prompt Engineering Best Practices from OpenAI, How GPT-4 Could Reshape Healthcare, and The Hidden Costs of AI Adoption

Written by Cathy McPhillips | Jun 6, 2023 12:03:59 PM

While AI breakthroughs slowed down this week, insights, best practices, and conversations continued. Paul Roetzer and Mike Kaput catch up on the artificial intelligence news impacting marketing and business leaders.

Listen or watch below—and see below for show notes and the transcript.

This episode is brought to you by MAICON, our 4th annual Marketing AI Conference. Taking place July 26-28, 2023 in Cleveland, OH.

Listen Now

Watch the Video

Timestamps

00:03:03 — GPT best practices from OpenAI

00:11:48 — Healthcare and generative AI

00:18:57 — How will costs of technology impact AI adoption?

00:28:32 — Center for AI Safety Statement on Existential AI Risk

00:33:17 — Falcon 40B 

00:36:11 — Apple AR/VR headset

Summary

OpenAI dropped chat prompt suggestions

A new guide from OpenAI offers six strategies for getting better results from GPTs: 1) Write clear instructions. 2) Provide reference text. 3) Split complex tasks into simpler subtasks. 4) Give GPTs time to "think". 5) Use external tools. 6) Test changes systematically. Is it that easy? What has OpenAI learned, and how can marketers follow these strategies while still differentiating themselves?

Could generative AI transform healthcare?

Could generative AI transform healthcare for the better? One expert thinks so. Dr. Robert M. Wachter, professor, and chair of the Department of Medicine at the University of California, San Francisco, outlines why in a new essay commissioned by Microsoft. In it, Dr. Wachter says he’s optimistic that generative AI systems like GPT-4 have the potential to reshape how healthcare works. This article caught Paul’s attention, and Paul and Mike break it down on the podcast, discussing not only marketing but also better patient outcomes and a reduction in healthcare costs.

High costs and AI adoption

According to a new report from The Information: “More than 600 of Microsoft’s largest customers, including Bank of America, Walmart, Ford, and Accenture, have been testing the AI features in its Microsoft Office 365 productivity apps, and at least 100 of the customers are paying a flat fee of $100,000 for up to 1,000 users for one year, according to a person with direct knowledge of the pilot program.” The proposed pricing models for AI features will impact business leaders' decision-making regarding AI adoption, especially small businesses.

This helpful episode of The Marketing AI Show can be found on your favorite podcast player and be sure to explore the links below.

Links referenced in the show

 

 

Main Topics

Rapid Fire

Read the Transcription

Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.

[00:00:00] Paul Roetzer: take a sample use case that you would use like writing a blog article or social media shares or developing a brief or something. And then like take your normal prompt and then start enriching that prompt using these recommendations and start to see for yourself.

[00:00:13] Paul Roetzer: How it evolves, and you'll probably start to see a lot better value outputs from the machine.

[00:00:18] Paul Roetzer: Welcome to the Marketing AI Show, the podcast that helps your business grow smarter by making artificial intelligence approachable and actionable. You'll hear from top authors, entrepreneurs, researchers, and executives as they share case studies, strategies, and technologies that have the power to transform your business and your career.

[00:00:38] Paul Roetzer: My name is Paul Roetzer. I'm the founder of Marketing AI Institute, and I'm your host.

[00:00:48] Paul Roetzer: Welcome to episode 50 of the Marketing AI Show. I'm your host, Paul Roetzer, along with my co-host Mike Kaput, how's it going, Mike? Good. How are you? Good. We are both in town this week. Not for long, but we are, yeah, right. We are both in Cleveland this week recording this thing live. It is Monday morning, June 5th at about 10:51 AM Eastern Time.

[00:01:12] Paul Roetzer: Now, it's important that I note this because as we'll mention later on, the Apple Developer Conference is today at 1:00 PM Eastern Time. So you'll be listening to this after the Apple event has happened, and we may then have brand new ar vr glasses from Apple, to sort of redefine that segment. But, we are recording this before that.

[00:01:37] Paul Roetzer: So if you're here to hear our take on Apple's, new AR glasses, it has not happened yet as we are recording. So, we'll talk a little more about, a little bit more about that later. But, This episode is brought to us by the Marketing AI Conference. If you're a regular listener, you've heard us talk about this event.

[00:01:55] Paul Roetzer: It is coming back to Cleveland, July 26th to the 28th. This is our fourth year. We're trending towards six to 700 attendees. It's looking like, I'll be there. Obviously Michael will be there. We've got about 30 to 40 speakers. Incredible lineup. So definitely check that out. If you're looking to be back in person for event and you want to be kind of at the center of what's going on in this space here, all about the latest technologies and, the best use cases and how to kind of pilot and scale this in your organization.

[00:02:25] Paul Roetzer: We would love to have you join us at the Convention Center in Cleveland, right across from the Rock and Roll Hall of Fame in beautiful Lake Erie. The address is www.maicon.ai. That's MAICON.ai. We hope to see you there. All right, Mike. Again, if you're new to the podcast, we pick three main topic. Each week we hit some rapid fires.

[00:02:49] Paul Roetzer: And today we had a last minute change to jump into the top spot because it's super practical and helpful, to us, and hopefully to you. So Mike, take it away. All

[00:03:01] Mike Kaput: right, Paul. So yeah, as you mentioned, we just saw, OpenAI drop a really interesting resource. It's called GPT Best Practices. And we first saw this because Logan Kilpatrick at OpenAI tweeted about it this morning.

[00:03:18] Mike Kaput: And this is an extensive resource from OpenAI that shares strategies and tactics for getting better results from GPTs. So, you know, the technology underlying things like ChatGPT and GPT-4. These are best practices. That are specifically for GPT-4 in the sense that all of them will only work with GPT-4, though I believe some of them would work in, other models like GPT

[00:03:46] Paul Roetzer: 3.5.

[00:03:47] Paul Roetzer: I've seen similar recommendations honestly, with like even like Bard and Claude too. So I do think that while what we're going to cover is specific to GPT-4 and from OpenAI, Based on our, you know, kind of look at this, it's a lot of tips that I think will be applied to any large language model you're working with, whatever your preferred model is, even if you're using an application like a Jasper writer.

[00:04:08] Paul Roetzer: These seem like really good tactical recommendations.

[00:04:12] Mike Kaput: And so they're basically six broad categories they cover. And I'm going to run through those real quick and we'll dive into the ones we find most fascinating. And under each one, this is why the guide is well worth a read. They not only include these larger strategies for getting better results, but for each one, they include tactics that you can also use to begin executing on their strategy.

[00:04:36] Mike Kaput: So the six strategies are as follows. First is write clear instructions. So they give a number of tactics to do this. Things like providing examples, asking the model to adopt a persona, et cetera. Number two is provide reference texts. So you're basically instructing the model to answer using some type of reference, an article, a citation, et cetera.

[00:05:01] Mike Kaput: Third is split complex tasks into simpler subtasks. Four is give GT's time to think. Five is use external tools to augment your prompting, and six is to test changes systematically. So Paul, there's a lot to run through here, but what, reading through this, I mean, how are you seeing marketers and business people getting value out of these

[00:05:27] Paul Roetzer: instructions?

[00:05:28] Paul Roetzer: Yeah, I just like that they're taking the step to just put this all out there because so far. For marketers, business, professionals to use these tools. We've talked about the importance of prompting in your ability to give guidance to the machine of what it is you're looking for, but there's v been very few, like authoritative takes on exactly how to do this.

[00:05:49] Paul Roetzer: There's lots of Twitter threads and lots of experimentation, but even for us, like I've just been trying to follow and kind of compile our own how-tos and what a good prompt looks like. So, I don't know, maybe just like I'll, I'll expand on a couple of uh or on the six you went through, because I think it's helpful for the context they have.

[00:06:05] Paul Roetzer: And again, we'll put the link in in the show notes so you can go check this out for yourself. But on the right, the right clear instructions, they say, you know, if their outputs are too long, ask for brief replies. So that's one thing people always seem to create like. If you don't get value on the first one, you can ask more.

[00:06:21] Paul Roetzer: You can give it more detail. And they're like, no, I'm, I'm looking for this. And so what they're saying here is like, really be clear with that. If their outputs are too simple, ask it to do an expert level. So if you get something and it's, it's too simple, say, can you, can you write this at an expert level?

[00:06:35] Paul Roetzer: Or vice versa. If they come back and it's too technical, say, can you write the, can you simplify this for me? If you dislike the format, demonstrate a format you'd like to see. So again, it gives you paragraphs saying, no, no, no. Can you do it like this and show it bullet points. So just like this whole idea, you know, we've talked about almost like briefing an intern, like all the depth of detail of exactly what you need.

[00:06:55] Paul Roetzer: The reference texts, we've talked about the importance of that before, but it says that they can confidently invent fake answers. What we've, you know, hallucinations, what have we've heard of the term, especially when asked about esoteric topics and for citations and URLs. In the same way that a sheet of notes can help a student do better, providing reference texts, GPTs can be helpful, getting fewer fabrications, the splitting complex tasks and the simpler tasks.

[00:07:20] Paul Roetzer: I've seen a lot of this actually, I believe a few OpenAI people have shared this on Twitter in, in recent weeks, but this whole idea of like, walk it through tasks, like walk through step by step. So rather than one big thing, Give it that, but then tell it to, to show its work or to do it in steps. And what it does is it enables it to make kind of less mistakes because it actually starts to like chunk these things off and do it in these logical, where there's logical break points and it reduces the air rates, in what you get back.

[00:07:50] Paul Roetzer: So that's a really cool one. The giving it time to think is not like a, it's a little abstract. They have thinking quotation marks, but it says if asked to multiply 17 by 28, you might not know it instantly, but can still work it out with time. Similar GPTs make more reasoning errors when trying to answer right away, rather than taking time to work out the answer.

[00:08:09] Paul Roetzer: So ask for chain of reasoning before an answer can help GPTs reason their way toward correct answers. So that's one that, again, is a little bit more abstractly. What is chain of reasoning? That's where the fact that they include these tactics in there, in this is really helpful to you. So again, take your time, go through this resource, click through and see exactly what they're recommending there.

[00:08:30] Paul Roetzer: Use external tools. So this is basically saying compensate by feeding them the outputs from other tools. For example, a text retrieval system, a code execution engine can help GPTs, I think, are they maybe referring to like plugins here? Where it's actually pulling in other resources, which Interesting.

[00:08:48] Paul Roetzer: This is total side note. I saw a tweet, I think I retweeted it last night. That. Sam Altman was quoted saying like, plug-ins don't have product market fit right now. So again, if you remember, Sam Altman was the head of Y Combinator product market fit was like his religion for startups. And so for him to say he's not seeing product market fit at the moment, I think what's happening is they pull these plug-ins into ChatGPT.

[00:09:11] Paul Roetzer: But they're not seeing massive adoption rates or value creation yet. And it almost seems like they're kind of wondering, is the plug-in model really going to work, or is it going to be, you know, is it going to be the ecosystem we thought it could be? So using external, then test changes systematically. So improving performance is easier if you can measure it.

[00:09:29] Paul Roetzer: In some cases, a modification to a prom will achieve better performance, but lead to worse overall performance on a more representative set of samples. Therefore, be sure to that a change is net positive performance. It may be necessary to define a comprehensive test suite, an eval as they talk about, these evals.

[00:09:46] Paul Roetzer: And then it kind of gets into all the tactical stuff. So, and there, there was a related resource that we'll put in the link that had the terms like around what are tokens, what are embeddings, what are prompts. So as, again, large language models are going to be a critical part of every marketer's job, every business.

[00:10:03] Paul Roetzer: So I think taking these steps to really start learning some of this stuff. And be more confident in your understanding of it is a really helpful thing and it's guides like these that can really help you start applying this stuff and start experimenting and don't, you know, again, don't just read it and move on.

[00:10:19] Paul Roetzer: Test it as you're reading it. Like go in and take, take a,

[00:10:22] Paul Roetzer: take a sample use case that you would use like writing a blog article or social media shares or developing a brief or something. And then like take your normal prompt and then start enriching that prompt using these recommendations and start to see for yourself.

[00:10:36] Paul Roetzer: How it evolves, and you'll probably start to see a lot better value outputs from the machine.

[00:10:41] Mike Kaput: Yeah, I've noticed, I think what you said at the beginning is really worth reiterating. If you have historically been unable to get good results from these tools for basic marketing and business tasks, you really do need to give this a revisit and experiment because oftentimes I've found the best outputs often come from more of what you might call a conversation with ChatGPT or what have you, versus just a single command working perfectly

[00:11:07] Paul Roetzer: the first time.

[00:11:09] Paul Roetzer: Yeah. One of the ones I think you've used this Mike, but I've seen this recommended too, is ask the ask g b t what, what do you need to know from me to do this? Yeah, yeah. You know, what questions should, should I be answering for you? So yes, like having a conversation and asking it what its needs, it really is worth the time if you're going to be using these tools as part of your daily workflow.

[00:11:31] Paul Roetzer: To become very proficient at prompting and the kind of the diff different technical ways to do it. And this is, you know, a good starting point. Hopefully, you know, can get you further along than you are if you haven't been doing these things yet. Cool. All right.

[00:11:46] Mike Kaput: Well next up we, Have an interesting resource that aims to answer a question.

[00:11:51] Mike Kaput: Could generative AI transform healthcare for the better? And at least according to one expert, the answer is yes. So we saw recently a new essay commissioned by Microsoft actually. And written by Dr. Robert Walker, who is the professor and chair of the Department of Medicine at the University of California San Francisco.

[00:12:14] Mike Kaput: And in the essay Walker says he's optimistic that generative AI systems like GPT-4. Have the potential to reshape how healthcare works now. He actually put GPT-4 through its paces in a variety of me medical or healthcare related scenarios, and he even says, quote, there is no question. That GPT-4 represents a breathtaking advance in medical ai.

[00:12:41] Mike Kaput: I fed it a series of very tough clinical scenarios, the kinds of twisty attorney cases that challenge our very best clinicians, and I found its overall clinical reasoning abilities akin to those of a very good medical resident, well beyond novice. But not quite expert. So Walker actually argues that wider deployment of AI in healthcare could lead to a

[00:13:03] number

[00:13:04] Mike Kaput: of different benefits, you know, from patients being able to manage some of their own health to reducing inefficiencies and expanding what each physician can do in a given day.

[00:13:14] Mike Kaput: And overall, this could lead to much better patient outcomes. And reduce healthcare costs. Now, what's really interesting is Walker also cautions that the barriers to achieving this cannot be minimized, and those include some significant barriers in the healthcare industry. One of them is he worries about the resistance to change from entrenched interests within the healthcare system.

[00:13:41] Mike Kaput: They will, he says, quote, push back mightily against substantive changes in the flow of dollars and work. He also cites worries around privacy and data concerns. You know, that could slow down AI innovation in healthcare as they try to navigate those issues. And also he worries that the training. Of healthcare professionals and administrators will need to seriously evolve as we learn to work with and oversee AI systems, and it has to do so in a way that minimizes automation, complacency, and overreliance on technology in this very critical kind of human function.

[00:14:19] Mike Kaput: Now, Paul, you had shared a post about this on LinkedIn. What did you find most interesting? What caught your attention about this article in particular?

[00:14:30] Paul Roetzer: Yeah, I mean, so we, I have a background in healthcare. When I ran my agency, we, we did a lot of work in the healthcare space. We have a lot of healthcare companies that are, you know, subscribers to the institute.

[00:14:42] Paul Roetzer: It's a topic we, we pay pretty close attention to. So, one, I was interested just from a healthcare perspective, but two, the thing that really jumped out it to me was this was a, a domain expert, a doctor who, you know, sees firsthand the benefits. But what I had said on LinkedIn, it was a, it was a wonderfully balanced article, based on both the opportunities and the obstacles of generative AI and healthcare.

[00:15:03] Paul Roetzer: But the thing that I really took away from it is this experience, this perspective I saw as being widely applicable to many industries that deal with sensitive data, high cost, both human and financial for AI errors. So if it, the AI goes wrong, the impact is significant. And concerns around liability of the AI outputs and the need for humans to own the outputs.

[00:15:27] Paul Roetzer: Because someone has to take ownership of what is created and recommended. So that was the main thing to me is I just thought it was such a well-written article that took a very balanced approach. Again, we've heard a lot of this existential threat to humanity, like it's going to be the downfall of humanity and all these concerns around the ethics and the safety.

[00:15:47] Paul Roetzer: And a lot of times what happens is you have articles that take an extreme on either side or, or leader thought leaders that take an extreme on either side. And what we try and do very hard with this show is to find the middle ground. Like what is the balance here? Understand the perspective of. Both sides, but say, okay, here's the reality of where we are.

[00:16:07] Paul Roetzer: And I felt like this article did that. Like he did a great job of saying, listen, we're not, we're not ignoring the fact that there are risks here. There are threats that can we, we have to accept there's potential negative outcomes, but the potential positive outcomes are so significant that we have to work to get this right.

[00:16:24] Paul Roetzer: But then as you called out, the one thing that jumped out to me in terms of like m many industries are going to deal with is, He said most of the stakeholders in the current healthcare system benefit from the status quo. That is so true. I mean, and we've talked about that, the law of uneven AI distribution previously, and that one of the ideas is that you're going to have to accept the technology to benefit from it.

[00:16:50] Paul Roetzer: They're going to be a lot of industries and a lot of companies and a lot of executives who don't want to accept the benefits because it changes things so dramatically. So we've talked about any professional service firm, for example, that's still charging bill blowers. That's going to be a really hard financial model moving forward.

[00:17:10] Paul Roetzer: If you're in knowledge work and you're charging by the minute or by the hour for what you do, it is not going to take you as long. So you're either going to let make less money. You have to find a way to move to a value-based model. And that's very hard for like lawyers, marketing agencies come to mind.

[00:17:27] Paul Roetzer: So I do think that there's just a lot of status quo. There's going to be a lot of resistance, especially in big organizations to change because a lot of people's careers and their success have been built on doing things a certain way. And they have no interest in doing it the other way. Another one comes to mind, like writers, we talk about writers all the time, we're writers by trade, right?

[00:17:50] Paul Roetzer: But a couple episodes ago we talked about like the Hollywood writer strike. And this is the basic premise. Like they don't even want to acknowledge, like they don't want to use ai. They just, they want to do what they do. And we're not saying it should in any way replace what they're doing. But we just have, there are so many people who have this resistance to even understand the tech, to, to figure out, okay, how can it actually make us better at what we do?

[00:18:11] Paul Roetzer: They just see it as a threat to replacing them and they don't want to hear about it. I've, I've experienced this at college universities with professors and administrators who don't want to hear about it. We hear, we've talked with writers, designers, artists like it, and that's our position is like I, again, I think this article just does a really good job of accepting that there are fears and uncertainties, and potential negative outcomes.

[00:18:34] Paul Roetzer: But if it can be done right, it can be dramatically transformative in a very positive way to an industry. And so this one is specifically for healthcare, but you can read this and apply the same kind of thinking to a lot of other industries. And so if you are in an industry or in a company where you're feeling this resistance,

[00:18:51] Paul Roetzer: this is probably a really good article to read, just to give you a little bit of perspective.

[00:18:57] Mike Kaput: So another factor that is not being talked about enough. Could dramatically affect AI adoption, and that's how much it costs customers to access the latest and greatest AI capabilities. There was just a new report from the publication, the information, and they said, More than 600 of Microsoft's largest customers, including Bank of America, Walmart, Ford, and Accenture have been testing the AI features in its Microsoft Office 365 productivity apps, and at least 100 of the customers are paying a flat fee of $100,000 for up to 1000 users.

[00:19:42] Mike Kaput: In one year, according to a person with direct knowledge of this pilot program. So this report goes on to say that Microsoft is basically trying to figure out how to price. The AI features, it is incorporating into its existing products. So in one possibility, Microsoft might charge an add-on fee to access the features in the other, it would add AI features automatically to Microsoft Office and increase the price of subscriptions per seat.

[00:20:11] Mike Kaput: So essentially what we're seeing here is that. As firms incorporate artificial intelligence, that is an expensive capability to run constantly as part of existing platforms and systems. So we could see a significant price rise in AI tools and actually other firms are paying pretty close attention to what Microsoft is doing.

[00:20:32] Mike Kaput: So there were two other companies, box, which is a cloud company, and Coda, a productivity app. That told the information they have considered raising their prices to cover the costs of running new AI capabilities. So Paul, I wanted to kick off by asking you, how do you anticipate that proposed pricing for AI features is going to impact the business leaders trying to use these tools?

[00:20:58] Paul Roetzer: I think there's going to be a lot of experimentation. I don't, I don't. I mean, a lot of these SaaS companies can draw on past pricing model experiences. Like I mean, I can speak firsthand for being HubSpot's first partner back in 2007. We went through dozens of iterations of their pricing model. So this is like a, a, a wonderful, highly successful SaaS company that was always iterating and probably still is iterating.

[00:21:25] Paul Roetzer: And some of them were fundamental shifts in terms of like charging by contacts, charging by usage, like complete changes in the paradigm of how it was done. And so this is very common in software companies overall. To explore and iterate on the model. The challenge here is going to be we've never faced a productivity gain, like could be, realized through these tools, and so then the question becomes, You know, if you, if you break down this math, a hundred thousand for a thousand users, it's only a hundred dollars per user per year.

[00:21:59] Paul Roetzer: Right? Like that, it almost doesn't, it seems almost stupid cheap. So I assume this is like a, a beta model, so it sounds expensive on the whole, it's like a hundred thousand dollars focus on that number. But when you break out, if that's not per month, that's just saying for one year, that's a hundred dollars per year.

[00:22:14] Paul Roetzer: Right? Like I would, I'd pay that in a second because like, I mean, you gain an hour of productivity in the year and you've paid for that person's time basically. So I think there's going to be a lot of experimentation and it's going to be highly competitive because if like, let, okay, let's just play this out.

[00:22:32] Paul Roetzer: So let's say Microsoft Office 365 comes out and it's available and it's good, like it actually works. And let's say I'm paying for a few other generative AI tools that I have. 19 bucks a month, $59 a month, $99 a month. Now maybe those applications have some enterprise capabilities or features that.

[00:22:52] Paul Roetzer: Microsoft often doesn't, or some like cool templates that we're used to using, or we have our preset prompts in there, like there's some sticky factor to these other tools. But if Microsoft and Google show up and all of a sudden for $9 a month per user, I can get all of these capabilities baked into Google or into Microsoft, I started asking myself, what do I need these other application companies for?

[00:23:16] Paul Roetzer: So one the big guys can have, can create pricing pressure on the market. By coming and saying, Hey, listen, we'll just break even for the first year or two and we'll just get rid of all these competitors. And now everybody's just going to use our tools. They could go upstream and they could offer enterprise capabilities that these other players can't.

[00:23:33] Paul Roetzer: So I mean, I think it's just going to be, it's going to take time to play out. I do, I'm unsure how players like Box, who has some awesome ai from what I've seen demoed so far, how some of these companies. Compete. Like it's, it is, it is easy to see a scenario where the big players get bigger and are the real winners here.

[00:24:00] Paul Roetzer: But yeah, I don't know. I mean, it, it was, it's the first time I've seen any real pricing data on this because I do think a lot of these SaaS companies are just pushing the AI capabilities out just to get people using them. Yeah. And it's like a, almost a freemium to date, but you know, it's not going to stay that way.

[00:24:14] Paul Roetzer: Because the cost for them to enable these features is significant at the moment. So, you know, when they're pulling the APIs from OpenAI or however they're building it, they're not getting that stuff for free. Yeah. So they're going to have to pass the costs on. And then the question becomes to, to what degree and can you compete with the bigger players?

[00:24:35] Paul Roetzer: And then the other variable is the cost is going to actually plummet for them to use these tools. Like the cost of the access from a OpenAI, they've shown they'll, they'll reduce the cost for the people to use the API over time. So I don't know, there's just, there's a lot of pricing factors here, but as you're building your AI roadmap for your company, you're trying to figure out which tools and which providers, the pricing factor is unknown at the moment, but is going to play a big part in it.

[00:25:02] Paul Roetzer: I don't expect a hundred dollars per user per year to be what this actually costs. I would imagine it's going to go way higher than that, but you should be able to build a business case for it by saying, okay, but if we have these tools, we're going to save 20% on administrative tasks. We spend a half a million dollars a year on administrative tasks across three people.

[00:25:27] Paul Roetzer: We can save X dollars. So yeah, we'll pay 300 bucks a month per person because we're saving 3000 a month per person. Like, it's going to, you're going to be able to build these kind of business cases to justify these costs. But again, I don't, I don't, I don't know too many organizations that are doing that yet.

[00:25:43] Paul Roetzer: They're that far along. So if I'm a small

[00:25:46] Mike Kaput: business, is there anything specific I need to be thinking about here? You know, I mean, in an ideal world, you're paying and testing all these tools, but small businesses often can have tight budgets and, you know, lack of time and

[00:25:58] Paul Roetzer: bandwidth to do that. Yeah, I think they're, the tools are going to be affordable to you.

[00:26:02] Paul Roetzer: So for one, I don't think you're going to get priced out of this. I, you know, I do believe that whether it's the application companies that are building on top of the APIs, or it's Microsoft and Google themselves, they have shown. Through their pricing strategies previously that they'll, they will make products for the small business.

[00:26:18] Paul Roetzer: Now, you may not have all the features that the enterprise players get. That would be the standard thing. You stack features for the higher grade ones. But the tools are going to be affordable. You do have to be careful with like tech creep though. Like I, again, I've, I've run small businesses. My agency was 18 employees.

[00:26:35] Paul Roetzer: Our institute's five employees, and it's real easy to throw 50 bucks a month at this tool and 99 bucks a month at that tool, and nine bucks a month here and 29 bucks there. And all of a sudden, as a small business, you have a $500, a thousand dollars a month tech bill across 10 different AI tools, eight of which you're barely even using.

[00:26:56] Paul Roetzer: So that's the thing you have to be careful of as a small business leader or marketer, is that you don't just start stacking a bunch of AI tools that you don't really use. You know, you really want to have, stay disciplined with your pilot programs, test tools for 90 days, fully use the features and capabilities, make sure you're trained on how to actually do 'em.

[00:27:13] Paul Roetzer: Like if you're using ai, AI tools, make sure you have proper, proper training on how to prompt them, like we talked about in the first topic. So make sure you're doing it right and then at the end of 90 days, make a decision whether to keep it or get rid of it. I mean, you, and you know, Mike, we, how many tools do we have sitting around that we just pay the license for every month that we don't even use.

[00:27:30] Paul Roetzer: Right?

[00:27:31] Mike Kaput: Yeah. Yeah. It's very, very easy for it to get out of control very quick. If you're a leader trying to

[00:27:39] effectively

[00:27:39] Mike Kaput: assess kind of if the benefits outweigh the costs, is it really just as simple as looking at those use cases, like you mentioned and saying, putting a

[00:27:47] Paul Roetzer: dollar value on 'em? I mean, I think it starts there, but I do believe you have to look at.

[00:27:53] Paul Roetzer: The people on your team who's using the tools, how does it affect their workflows and productivity? and then what's the value exchange there? I mean that's kind of how we always teach all this stuff is AI's just smarter technology. You have to look at it the way you've always purchased marketing technology.

[00:28:10] Paul Roetzer: You have to find what the value is to it. And it could be productivity gains, it could be increased revenue, it could be, it saves your cost because you can consolidate from three other tools like, but go in knowing what is the goal out for this technology and then measure against that. And you know, that's kind of the way you have to decide whether or not they're viable for you moving forward.

[00:28:30] Paul Roetzer: Gotcha.

[00:28:32] Mike Kaput: So let's dive into a couple rapid fire topics. And the first one is, a little heavy, but, top AI re leaders just released this kind of explosive statement about what they call, quote, risk of extinction that AI poses to humanity. This statement was released through the Center for AI Safety, and the statement is literally just a single sentence, and I'll talk about that in a second.

[00:28:59] Mike Kaput: The single sentence is mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks, such as pandemics and nuclear war. This statement was signed by top people in ai, like Jeff Hinton, formerly of Google, Demi heave. As well as OpenAI, Sam Altman, Stability AI, CEO, Emad Mostoque, and even Bill Gates.

[00:29:22] Mike Kaput: And the statement's designed to be short. So the Center for AI Safety basically said as we're kind of increasingly discussing, Important and urgent risks from ai. It can sometimes be difficult to voice concerns about some of advanced AI's most severe risks. So they intentionally created a succinct statement to essentially get past that obstacle and start a conversation.

[00:29:49] Mike Kaput: So, Paul, this was a pretty controversial statement. Some people are coming out obviously very, very for it and saying, we need to pay serious attention to existential risk here. Others thought it was wildly overblown and a little ridiculous. What did you think when you

[00:30:06] Paul Roetzer: saw this? I was confused, honestly, at first, I, when I read the statement, the 22 words, I kept scrolling down saying, okay, so where's the rest of this?

[00:30:16] Paul Roetzer: So I thought it was just a. It was kind of bizarre, and so I had to dig into it a little bit and go see what different people are saying. There's certainly a lot of entrepreneurs and AI researchers that sign this, that I have huge respect for and that we follow. Like de Saabas for example, like that's kind of my, he's sort of my North star in a lot of this, this area.

[00:30:35] Paul Roetzer: So if he's signing, it's like, okay, there's something to this. This is, first, I had to make sure it was legit that these people actually were signing this, but then they, one by one started tweeting that they did sign it, and here's why. The counterpoint to this that you're going to hear from the people who don't buy into this is pandemics in nuclear war.

[00:30:53] Paul Roetzer: Very clear existential risk. We understand how we go from A, it exists to B. We don't look that's obvious. No one in the AI world that claims it's an ex existential risk seems to be very good at explaining how or like what exactly it is about it. That's existential risk on par with pandemics and nuclear wars.

[00:31:16] Paul Roetzer: Or, or they'll say, okay, it could be, but it's not. And it might be years, or it might be never that it is, and are we just creating unnecessary conversation around this and distracting from the real focus. So that's what we've said on this show before is like, fine, great. I'm happy that people are researching this and thinking about it and talking about it.

[00:31:37] Paul Roetzer: But I'm more happy to know that people are focused on the near term issues that are, that are very real and tangible and we should be solving for like workplace and economic development and workforce disruption and, bias and algorithms and things like that. Like that's the stuff that's here now and we should be focused on it, education system disruption.

[00:31:58] Paul Roetzer: So, I don't know. I mean, it's still kind of weird to me. The whole premise of this, legit people signed it. I said on the LinkedIn post, like the most interesting part to me is probably that a bunch of these major AI research companies and labs seem to be unified in these concerns. Many of them and appear willing to collaborate on solutions.

[00:32:21] Paul Roetzer: And then I said it could be either they know government regulation is coming and they're trying to sort of band together to just solve this themselves. I don't, that doesn't really jive. Like I don't, I don't believe necessarily that's the reason. Right. And then the other I said is it could just be that they really do think this is a, a very near term threat to society and they gotta figure this out fast and they need support and there's probably other options of why they're doing it, but.

[00:32:51] Paul Roetzer: I don't know. I mean, like we've talked about on the shore, we, we surfaced this stuff just so you're aware of it. It's not hit the panic button time, in my opinion, but it is pay attention. There's lots of very intelligent, important people in the AI world who are a part of this. And so it's noteworthy for, you know, better, for worse at the moment.

[00:33:10] Paul Roetzer: It's worth at least paying attention to and seeing what comes of it.

[00:33:16] Mike Kaput: So, One other topic people may be hearing about this week is this name Falcon 40 B that is making the rounds. And what this is is an AI open source model, and there's a couple really interesting things about this. This is what we'll call foundational large language model, and it's got 40 billion parameters and it's trained on 1 trillion tokens.

[00:33:40] Mike Kaput: Now what this means, this is open source. For research and commercial use, so anyone can build on top of this model. Anyone can fine tune it to whatever purposes. That they would like it to do. Now, what's also interesting here is this model was released essentially by a government. So Abu Dhabi released Falcon 40 B as part of the Government's Advanced Technology Research Council, and as part of the Technology Innovation Institute, a research center within that council.

[00:34:12] Mike Kaput: So we literally have a government body releasing a major. Open source language model that anyone can use. So Paul, this is a pretty significant development it sounded

[00:34:23] Paul Roetzer: like. Yeah. Again, this is, you know, gets to more technical and I know, you know, some of our audience maybe isn't down with all the technical stuff and we're not going to get into like explaining the parameters and the.

[00:34:36] Paul Roetzer: The weights and the tokens and all that. Like it's, I think the key takeaway here is every week it seems there is news around innovation within the large language model space, which is the underlying architecture that's powering all the writing capabilities that we're all having as well as image generation, you know, video generation, like all this generative AI stuff.

[00:34:54] Paul Roetzer: But the language model is the core here, and open source versus closed is the conversation we've had. So we have our closed like OpenAI and anthro. Cohere. And then you have your open models like Llama, well sort of open when they release the weights, stability, lm, Falcon. And so these are things anybody can build on and they can, in the case of Falcon, get access to all the backgrounds of how it works and they can adjust it themselves.

[00:35:21] Paul Roetzer: And so I think it's just noteworthy that a lot of people in the AI space, This is what they were talking about last week, and it just indicates continued innovation of large language models, which also means continued complexity for marketers and business leaders who are trying to figure out what bet to make in the language model space.

[00:35:42] Paul Roetzer: What do you build on top of? If you're building a marketing sales service engine that's going to be powered by these language models, it's be, it's continuing to be, A noisy space, and there's very little guidance on how to actually pick the right language models for your company. And so it's important just to surface this for you so you're aware that things are continuing to move and there's innovations that we're still trying to comprehend every week.

[00:36:09] Mike Kaput: So, like you alluded to at the beginning, Paul, in a couple hours here, we're going to kick off Apple's Annual Worldwide Developer's Conference. It goes June 5th through the ninth, and this is huge because Apple is expected to be launching a rumored ar vr r. Headset. So, augmented reality and virtual reality, and this is big for a number of reasons.

[00:36:35] Mike Kaput: Obviously VR headsets exist already, but the industry has really struggled to kind of take off because the headsets and the use cases are kind of unwieldy and pretty limited. So the thinking is, and I think one VR developer called it in an interview with The Verge. The best thing that could ever happen to the industry is someone like Apple getting into the mix and creating hardware.

[00:36:57] Mike Kaput: Around vr, but also AR is exceptionally interesting as well, given that you could have some type of headset or glasses that overlay your physical world. And obviously AI is a critical component of creating these AR and VR experiences. So Paul, I know you're a huge Apple fan, so is this

[00:37:15] Paul Roetzer: exciting? I am a huge Apple fanboy.

[00:37:18] Paul Roetzer: I do pretty much buy anything they release. I'm not sure about this one. I'm not an ar vr guy. I just like three weeks ago, got an Oculus two, I think it is. It was gifted to me. So I, for the first time my life put on a headset like three weeks ago. I don't know. I've heard mixed reactions. I've heard from some people who have tested this and heard being, like, seen on Twitter, that it is, it is insane.

[00:37:45] Paul Roetzer: It's so good. The technology is so advanced, like the most beautiful screen they'd ever seen. I think it's going to be fascinating from a technological standpoint. I've heard that the price point is going to be close to $3,000. Again, this is all just hearsay and, and stuff you're seeing online.

[00:38:03] Paul Roetzer: Meaning it's, it's going to be a really high price point. So I don't think this is meant to be like the iPhone moment where we're all going to be walking around in ar vr glasses. But I guess we'll find out. We'll see what they have. Yeah, I know they've been working on this for a long time. It's been kind of hush hush.

[00:38:18] Paul Roetzer: They haven't talked much about it. So we'll find out in a couple hours. And again, by the time you're listening to this, go online and check it out for yourself. If you haven't seen it. It should be really interesting. I do think that while it is competition for meta. Meta needs apple probably to legitimize this space and to bring it mainstream.

[00:38:39] Paul Roetzer: And, we'll see what happens. I'm intrigued though. I mean, If it's not $3,000, I might go grab one just to Right, to experiment with it. That's a business expense, right. We gotta, we have to test this. You and I need to do our podcast in augmented virtual reality. We gotta go get a couple of headsets to try it.

[00:38:55] Paul Roetzer: That's idea.

[00:38:57] Mike Kaput: I like it. Yeah. So, Paul, as always, thanks for the insights and for taking the time to walk us through the developments in ai. I'll kinda let you wrap us up here. I know we have a couple of like quick announcements.

[00:39:08] Paul Roetzer: Yeah. So we have, so you're listening to this, like this comes out June 6th. The next two episodes of the podcast are going to divert from our standard weekly format because Mike and I are both traveling and it's become impossible to pull these off.

[00:39:22] Paul Roetzer: So we'll both be out of the country a little bit here and there. And doing some speaking tours. So we are going to episode 51, which will come out on June 13th. We are going to actually curate a list of the top questions we get in our Intro to AI class. So if you haven't taken the intro to AI class, we have an intro to AI for Marketers class that I teach every few weeks.

[00:39:43] Paul Roetzer: We've been doing this since November of 2021. It is free, it's done through Zoom. We have done 25 of them. We've done seven this year, I think, and we get, On average, let's see, we've been averaging probably around four to 500 attendees each time. So you can imagine we get a lot of questions. We've had over 11,000 people register for this series, so we have literally hundreds of questions and we're curating them with the help of ai and we're actually categorizing them.

[00:40:14] Paul Roetzer: And then picking probably 15, 20, 25, we'll see how many we get through. Well, we're basically going to tape the take the top questions that we get in as part of this intra AI class. And Cathy McPhillips, who moderates that for me, the intra AI class for me, our Chief Growth Officer. She's going to join me to moderate, q and a based on those FAQs from the Intrad class.

[00:40:34] Paul Roetzer: For episode 51. So super tactical, all focused on things that people are asking us all the time. So hopefully you'll get a ton of value and knowledge out of that one. And then episode 52, we decided it would be fun to do sort of a year in review of what are the biggest moments in AI up till this point.

[00:40:51] Paul Roetzer: So it'll be right toward the end of June and would air what? June 20th, I think. So that one will be focused on the top 10 or 15 things that have occurred this year. So it'll be a great way to catch up on all the news from the year. And if you haven't been listening to the podcast all along, a chance to sort of hear some of the big moments throughout the year.

[00:41:11] Paul Roetzer: And then Mike and I will be back for what, I think it's the June 27th or something like that? Yeah. With our regular weekly format, hopefully. Things don't go crazy for the next two weeks, and we have to have like a mega episode to catch up on all the news, but we will catch you up on anything major that happens in the next couple weeks while we're gone.

[00:41:31] Paul Roetzer: So stay tuned for that. So again, episode 51 is F A Q based on the Introed AI for Marketers class, and then episode 52 is going to be a mid-year in review of the top moments and news in AI so far. So as always, thanks for listening, and we love hearing from the listeners, so don't hesitate to reach out and connect with Mike and I on LinkedIn and, we'll look forward to being back with you again on June 27th.

[00:41:55] Paul Roetzer: And Cathy and I will be with you for the next couple weeks with the special editions. Thanks, Mike. Thanks Paul.

[00:42:01] Paul Roetzer:

[00:42:01] Paul Roetzer: Thanks for listening to the Marketing AI Show. If you like what you heard, you can subscribe on your favorite podcast app, and if you're ready to continue your learning, head over to www.marketingaiinstitute.com. Be sure to subscribe to our weekly newsletter, check out our free monthly webinars, and explore dozens of online courses and professional certifications.

[00:42:23] Paul Roetzer: Until next time, stay curious and explore AI.