Dashboard: Seeing Past the Workslop and Hallucinations

Welcome to another 21 Hats dashboard. I'm Lauren Feldman and I'm here with Jean Marks. Hello Jean. >> Hey Lauren. It's been a while. >> It has been a while. It's great to see you again. Uh appreciate you taking the time. >> Yeah, it's mostly my fault. I've been doing I've been traveling a lot. So, you know, >> as usual, traveling a lot, I might add. >> Yeah. >> Jean, you've been writing a lot lately, despite all your travels, uh about all kinds of things, but especially about why AI isn't quite ready for uh prime time. [snorts] Um, and anyone who's been paying any attention at all knows you got a point there. There are [laughter] always issues in the news. To be honest, I I wonder if sometimes the the negatives are overplayed, but but let's I I want to hear what you have to say. What you you wrote a piece for Forbes specifically uh about the image generation services. Um, but [snorts] te tell me what you're thinking. >> Yeah. So, first of all, this conversation, Lauren, is um yes, there there's two themes to this. It's not exactly ready for prime time, but but there are some things that are ready for prime time and are of business use, and I'd like to hopefully we can cover that in this conversation as well because I don't want to I don't want this to be like a like we're just on AI because that's not, [snorts] >> you know, that's not the case here. You know what I mean? like but but there there is some reason why there's been recent studies and recent articles in the New York Times and other places about work slop and you know AI is still hallucinating and whatever. So you know >> that work slop phrase is everywhere now. [clears throat] >> Yeah, it's everywhere and you know and and for good reason because you know people are trying to use this stuff and they wind up to spending more time fixing AI than you know than if they just did it themselves. And so I I I kind of my my story was similar on the image generation thing. I wrote a piece for Forbes and um and you know for if you're listening to this conversation you can go to Forbes and see it because I I put the images up there on the piece and I just did like a simple um you know simple prompt saying hey listen I I used uh um Google Gemini uh chatgpt's image creator and uh Grock okay [snorts] which is by the way Grock has been great but okay so I used those three AI assistants uh obviously there are a bunch of others out there and I gave it a prompt I said you know create an image of a Yorkshire Terrier hitting a home run uh in a [laughter] baseball game, you know, um because I thought that would be like kind of fun to do. And um we, you know, we own a Yorkshire Terrier, so you know, it's close to my heart. And uh so it did and it generated them within, you know, seconds, under a minute. And um each one and the Yorkshire Terriers were very cute in all of the three photos or the three images. Um very AI looking but cute. And they um but they were all wrong, you know. They they they weren't quite there. I mean, one showed the Yorkshire Terrier um pointed like like facing us like with his back to home plate, you know. Another was like it was like standing on home plate looking in the wrong direction. Um another didn't have a bat at all. The ball was just sitting in front of home plate, you know? [laughter] And then I went back and I started prompting again like no no no you know please you know have the Yorkshire Terrier batting on the right side right-handed side of the plate and you know pointed out to you know hitting the ball out towards the fence as a home run and it and it still generated it wrong. It was like the ball was going the right direction or it just didn't listen to me and still had the Yorkshire Terrier pointed in the right d wrong wrong direction or that it generated a an image and like there were other dogs on the field all of a sudden except they weren't even in the right positions. It's just going to look right. You know what I mean? And like and I want to just say first of all like because you and I, you know, we're the same age. Like we didn't have any of this stuff. So we're all still like amazed by this stuff. It is amazing that you can actually go onto these platforms and give it a prompt and within seconds it generates like this image. You're like, "Oh my god." I mean, it's really amazing, but it it's still not right, Lauren. You know, it's that's the story of AI in 2025. Like if you're a business and you're like, "Hey, I want to create this specific image that I want to use on my website or in my business materials or marketing, you know, brochure, whatever." You know, you want it to be exactly the way you want it to be. It's your business. This isn't a post on, you know, on on Facebook. And it's just, you know, you can't do it. And I was doing prompt after prompt. Now, listen, maybe I wasn't given the right prompts. What can I tell you? I'm not a prompt genius. you know, maybe uh maybe we I need to call in a developer from OpenAI to sit there and, you know, get underneath the whatever and, you know, and do it. You know what I mean? So, but it just wasn't it just wasn't right. You know what I mean? And and I want I just want people to know that that it's just you know, you can play with it. It's still a toy, but it you know, it just it just doesn't get there. The other examples that I give is um and I didn't give this in the Forbes article but you know Google has something called VO which is their video creation tool you know and Chad QPT or OpenAI has got something called Sora and again it's unbelievable I mean you give it a prompt you know create a video of two woolly mammoths you know walking through the snow uh with willow trees above them and a cloudless sky you know whatever and it'll take 10 minutes 15 minutes and then it creates this video for you and it's like everything else. It's like, "Oh my god, this is unbelievable stuff, you know, but it's still not exactly right." And you see these videos online. Um, and they're just people are playing with them, but I always think to myself, Lauren, like these tools have now been out for a couple years. I mean, there's new versions. They've been getting better and better, you know? Yep. But I know for a fact that there are countless young kids out there that are creators that that can take these tools, they don't have any money, so they can't go to Hollywood and make movies, but with these video, you know, generation tools, they can create their own stories and some awesome, you know, you know, films and, you know, it's unbelievable. And I believe that day is coming, but it's certainly not here yet. >> Let alone a a small business making use of it in its marketing or whatever. Let's use that as an example like you know like a small business hires one of these kids out of an art school or whatever and say hey here's Google VO create a really cool video and show people from my company going and doing work you know with our customers and whatever they come out and like you're you're not seeing it you're really not seeing any of it like why are we not seeing this yet I mean it's not I know it's relatively new but it's been around what and I think the evidence is clear it's just it's not it's not ready yet you know for for businesses to use. It looks goofy. I don't know if you saw the um there was this ad in the New York mayor race with you know uh uh that Andrew uh what Cuomo put out against Mammi and it was like this AI generated video and everybody's been talking about it because it was whatever. Um and I looked at the video so first of all it's like you can look it up anywhere and see it and it's it it looks very AI. You can always identify these things pretty quickly. But then secondly, I'm like, "All right, it looks pretty good, but I'm I'm betting they they put a lot of money behind this." Like they probably had developers working on the code behind the scenes for all I know. You know what I mean? Or um you know, because it looked that that good. So I guess if you if you spend enough, you can get it to some level of accuracy, but I don't know if the typical small business owner is is is going to do that. And that just defines where we are with AI. And I'll stop right there. Does that make sense? It does make sense. Now, you just got me curious. What What's the positive side that you were referring to that you were going to share? >> Yeah, I actually do have um some good positives. We're like AI and action that really, you know, does work and it's getting better and better. We all know the searching, you know, that you can do now and and that's improving and getting that much more accurate. But I will say like if you have co-pilot or Google sheets, their AI assistants um you know sorry if you have office or Google sheets uh the AI assistance co-pilot for Microsoft and Gemini for Google they're getting better and better and I do have clients that one of the better co-pilot applications is for PowerPoint and you know I and I've tried it out myself. you can, you know, you know, open up copilot inside of PowerPoint and you can connect documents to it and you can say create a PowerPoint presentation based on uh these documents and and it does a pretty good job of it. It does it like 80% right. You still have to go into PowerPoint and make some changes yourself or ask it to make some changes. Um, you're still going to have to tweak it. You're still going to have to spend some time. But I think if if we all as business people lower our expectations like it's not going to change the world. It's not going to eliminate people. It's not going to do that. But it will make you more productive. Like the PowerPoint presentations I had it do will do it um uh you know what it would take me four hours to do is now taking me like half the amount of time because it's doing a lot of the leg work for me. And I think that's valuable. You know that's real life working. >> I agree. And and that's kind of the point [clears throat] that I I was going to make. I I I think there's a distinction to be drawn between the types of jobs people give AI and I I think I haven't seen this articulated and it's I'm not doing a good job of it myself, but let me try. I I think it's one thing if you're relying on it to generate either research or images or anything like that. You're you know, you're asking it to start from scratch and create something. That's where you get the hallucination and the issues. But if you're feeding it your information, your data, and asking it to analyze that data or create something just within the confines of that data, I think it does a really good job. So if like if you're [snorts] analyzing, you know, a business strategy, it's less likely you're going to get a hallucination there. I mean, it's not like you're you're just going to do whatever it tells you to do, but >> but you're probably going to get if you put good information in, you're probably going to get something worthwhile back. Same thing with, you know, marketing materials. If you're asking it to create a blog post out of the information that you give it or a uh you know if you're responding to a request request for a proposal and it's creating a proposal for you based on how you run your business and what you do and it spits out uh a proposal it's probably a pretty good draft that you can work with and not likely to have these kinds of hallucinations in it. I don't know I don't know how you draw the distinction between those two types of of job but I I think there is one don't you >> there is and to me it's all about just a productivity and people are all up in arms about AI replacing you know workers and yeah at some levels you know you know like like bas basic you know computer programmers are going to be replaced by this stuff u some customer service jobs are going to replace I get that I get that but really for the you know for for at least you know you know for the cons foreseeable future the smartest employees will embracing this stuff and just being that much more productive. Like if I worked in a marketing u department at a company and we I was doing powerpoints for example as part of like what we do um I would be expert at using co-pilot so I could like produce 10 times the amount of powerpoints in the same amount of time and I'm that much more valuable to my employer. You're still needed because somebody's got to use this stuff to make it work the right way. You know, I'll give you another example like really good business use of AI. Google Meet um just recently has introduced its translation feature and I've tried it out and it's unbelievable. I mean, you turn it on and you can meet with I think it supports like 30 languages and you talk to somebody in a completely different, you know, in English, they speak Spanish, they speak back to you in Spanish, it either gets subtitled or a human voice will speak for you, you know, speak for them in English. And it's like I I find that to be like I just it's it's a little thing, but it's big, you know? I mean, like now you talk to people on a regular Google Meet call and you don't have to worry about language differences anymore. And I think that that is um I think that's like that's like extraordinary stuff. And then I'll give you one other like application that I've used it for. Like I have um I use co-pilot in one drive and um that's you know Microsoft's you know where you store stuff and um you know like we had an ad out for a developer and I got a bunch of you know resumes in the ad you know I placed it on like Craigslist and and I know homicidal maniacs I get that uh but it's a developer so they're all homicidal maniacs trust me. So anyway, I got like, you know, 20 resumes in and I put them in a folder in one drive and one drive has a co, you know, co-pilot and I go to co-pilot and I'm like review you these, you know, review these and then come back and uh create a table for me showing um you know name and address and experience and then and then you know uh sort it by experience and in two minutes it did that like it would have taken me to go through 20 resumes and build a spreadsheet. It would have taken me probably I don't know an hour or two to do. it did it for me like in two minutes and accurately. So, there's that. Okay. Then I, you know, and again, jump in if I'm rambling here, Lauren. Okay. >> Well, I was gonna say you also wrote a piece not too long ago uh about an AI platform called Slang that restaurants use to take reservations and >> um you know, [sighs] I mean, I'm sure it doesn't work perfectly. I'm I'm sure there are hiccups, but >> I pretty good. It's easy to imagine that it's making restaurants more efficient. >> Yeah. And I'm so glad you I didn't I wasn't even going to remember to bring that up. Yeah, you're right. I did write about that for Forbes and I interviewed a few people that use it, including the company's founder. They have I forget a few thousand implementations at restaurants around the country. And it is they're smart. You know, you call up the number, Slang picks up the call. It's a bot that's answering the call. However, you can configure it so that the bot says to the caller, I'm a bot. If you want to speak to a human, just press zero and you can speak to a human. So you don't have to feel like you're going through bot hell, you know, [clears throat] and but you know, and then you it but it says, but I can get this job done faster, you know, just faster if you want to work with me. And then you know, when would you like to make a reservation? And it understands your voice. We all know voice AI voice has gotten better and better recognizing us when we talk into our phones and whatever. And it makes restaurant reservations for for people. And then if people have more complex questions, dietary needs or parking or things like that, it then transfers you over to a human. And it's pretty good. And and they're they're, you know, these it it ain't cheap. It was like 450 bucks a month per, you know, for a restaurant to use. But the restaurant owners that I talked to were like, "We're not we're not getting rid of anybody. it's just that we have a host and he or she's very busy dealing with customers and then they're getting calls for people to make reservations and they're like the more that we can free up their time so they can be facetime with our customers there and and you know help them the the better off we are and they found it to be of huge value and so there's another example of like yeah this is like really happening right now in small examples but you know really tangible let me give you another one if I mean >> you know chat at GPT. Now, first of all, if you're running a business and you're not um you're not using, you know, an AI assistant, you're you're nuts. You you need to be doing it and you need to be coughing up to 20 bucks a month for whatever their pro or business vision version is or whatever. There's chat GPT, Microsoft Copilot, Google Gemini, Perplexity, Claude, and Grock are. And again, a year ago, I don't think I could I would have recommended half of them, but they're all good options. And there's other ones that are out, many others that are out there too, you know, but just just using chat GPT, I mean, once you pay for the for the business version of chat GPT, which it gives advanced reasoning and expanded messaging and you know, and and you know, projects and all that. I just want to make sure that people realize number one if you if you don't trust the first of all you can upload now any photos, documents, PDFs, files to it for analysis. So now it'll it'll handle all that stuff. That's number one. Number two is you can speak to it on the mobile app which is great. So you can be walking around driving around and talking back and forth about an issue or a problem or >> Have you [clears throat] done that, Jean? >> Yeah, like a mental patient. I'm driving around the city [laughter] with the chat GPT and I'm talking to it because I'm like I have a thing on my mind or whatever and it's answering me back. It's and it's good. It's like you know it's whatever. So you got that. >> Did you ask it why the Phillies lost to the Dodgers? >> I've had baseball conversation with Chad GBT. Lauren, I told you before we started don't bring [laughter] up the Phillies, right? Stay away from this. Okay. So anyway, I if you don't trust the answers in chat GBT there's the the the pro version is something called deep research. So it will you ask questions of it and it will take minutes to answer and the reason why it's taking longer is because it's it's going through a fast like a deeper process to really validate its responses to you. Now most of us have short attention spans so we want quick answers but I just want to say if you've got an issue with you know you you really want to get more accurate answers they have this deep research function which is more accurate. They have an agent mode. So, Lauren, I've created agents like I have a I have a client um if I told you this there you should interview this woman. She's great. She's a womanowned business in Wasilla, Alaska, a construction company, right? She's got like 40 people and I've been working with her for the past six years. >> Did she once run for vice president of the United States? >> She did not. She did not. Of course, the typical Laurens got to get political. No, we're gonna [laughter] keep that out. Keep it out. Priscilla is the town, wasn't >> I don't even remember. She hasn't she's never seen Russia so she [laughter] doesn't have any foreign policy experience. Anyway, she um so she great story construction company whatever. Anyway, in chat GPT um I created an agent and I said every Friday send me five stories that impact the construction industry in Wasilla, Alaska. And every Friday it sends me stories. Now, sometimes it can't find five and it'll tell me, but it'll send me a few stories and then I email her and I'll be like, "Hey, I heard there was this project going on here or heard whatever." And she's like, "Oh my god, how'd you know that? That's fantastic. Thanks for letting me know." Like she thinks I'm a genius because I'm like telling her what's going on. It's just chat GPT giving me updates. It's the same thing when I do, you know, I cover small business. So, I asked Chad GPT, Grock, and and Jen, they all have agents, and I'm like, "Send me an email every at the end of every week with 10 business stories that impact small businesses in in the country." And they do. And I get these these list of 10 stories and then I'm, you know, it it helps me figure out what to write about myself or talk about or whatever. These are agents, you know, they can create calendar items for you as well. That's little stuff. Productivity is good stuff. Do you know what I mean? I'm going to call up every womanowned construction business in Wasilla, Alaska, and let the owner know that she could be doing this herself. >> Don't tell her. Okay. She's like five hours behind. [laughter] She's not even know. She never heard of 21 ads. Although I I should tell her about your >> Yes, you should. >> I should. She She would be a great addition to your community. Um anyway, one final thing on ChatGpt. Uh you can People don't realize that because you don't dig down into it. You can connect it. You can connect chat GPT to Dropbox, Gmail, uh Google, your calendar, Google calendar, your Outlook calendar, uh HubSpot, Slack, uh Google Drive, SharePoint, Outlook, Teams. >> Okay, I'm going to expose my ignorance, but give me a few examples. Dropbox is the one that stopped me first. Why would you connected to Dropbox? What can you do? >> I'm so glad you asked that because I I connected it to my We use Dropbox. that's our, you know, what we share like in our company. So, I'm like, I'm budgeting right now for 2026 and I'm like, take a look at the last two years of presentations that I gave um in this folder and what was the average presentation fee that I charged and it went through all of the presentations and came back to me um with its average. And then I would I I ask it questions like where, you know, what presentations did I do in California in the past three years, you know. Then I have another folder of all of our quotes and proposals that we give out for our CRM clients. And some of them are specific. So we had a a prospective client that we were doing on um you know, like a potential HubSpot project. So I said, "Find all the quotes that we did HubSpot and and give me a summary of the type of procedures that we performed." And it looked into my data and came back to me and and told me what that was. You know what I mean? So it's not just the stuff that's on the internet, but now I'm querying my own files and spreadsheets and documents in Dropbox if I have a specific question about that. and I can see if I'm a if I'm a construction company and I've got a a bid for a building and and I say, "Okay, go through this folder in Dropbox." Um, find all the the similar bids that I've done. Recommend what materials do you did I use before. Maybe recommend other materials based on your chat GPT. Tell me if you think I could use better materials. But it's all basing it off of your own company's information. And you can be doing that by just connecting it to those services. How much technical skill does that require to connect chat GPT to Dropbox and >> you go you literally go into the the setup and configuration for chat GPT and then you'll see an option for enabled apps and connectors and then you'll see a list of those connect those apps and then you say okay connect to Dropbox. It will ask you for your username and password. We can talk about security in a minute if you want, but you you it asks you for your username and password and then you you get permission from Dropbox. You have to give it permission for chat GPT to access it and then it's accessed. That's it. You know >> what about to set up the agent? >> Not at all. You talk to it. So you go you go again if you're there at chatgbt plus and you hit the drop down on the main screen it will one of your options is agent mode and you choose that and you type in your your request send me an email every week showing me this or you know uh make sure I'm alerted when I have a calendar you know issue if you know and and the examples I'm giving you are so minor I mean like just ask chat GPT or Google and say give me some examples of how I can be using agent mode in chat GPT and it will give you a whole bunch of other ideas I'm not smart enough to give you right now that will that will help you and then the other thing in chat GPT that I use is they have projects so by the way I'm I know I'm I use chat GPT um you know and I use Grock but they all have this stuff >> they call it different names but it's same stuff >> yeah and but with projects if you have like if I'm working on a piece and I'm doing research for this piece and then you know I I could do various chats and then just save them under a project for that piece so that I have one it's just an organization thing you can go to that one area and see all the you know all your chats there in one area so you've got that organized in that way um but that's like again it's this whole conversation about productivity you know and that's where I see AI being used by small businesses that are smart just in in small in snippets small stuff >> well see you kind of made the point that uh I was heading toward myself which is there's so much conversation about the the flaws the work slop uh the hallucinations and I think there's this whole world of things out there that you know you got to be careful you got to be smart um you've got to experiment you got to keep an eye on it but it can do incredible stuff already >> well I'll give you the reason why there's so there's all this attention is being played do you know who Andre Cararpathy is >> no >> he was one of the founders of open AI. Okay. And uh so he's like this AI deep learning genius and he was literally just interviewed I forget the name of the podcast but you you can find it if you Google him. It's Cararpath is K A R P A T H Y. So he's interviewed on this tech podcast and he was talking about agents. Okay. So the the big tech companies dup us. They they say oh it's going to be agents. you know, AI is going to be, you know, you know, reconciling accounts and placing orders and, uh, you know, booking your travel and doing all this, you know, behind the scenes and you just give it a command and away it goes, you know, and this guy is like an AI, you know, one of the founders of of Open AI, you know, you know, he's like this whole agent thing, it's like it's going to be like a decade before all of this is actually really working well and reliable and useful. But what happens is is that big tech, they roll this stuff out like Microsoft Dynamics and Microsoft rolled out these agents promising all of these functions that AI will do in Dynamics where you won't need people to do it and it doesn't work very well yet. There were just too many challenges. The data is not good. The technology is not ready. I can't even create a, you know, freaking image of a dog hitting a baseball, [laughter] you know, with AI, let alone relying on it to process my payroll. You know what I mean? Yeah. >> So, even this guy is like, yeah, it's a it's a decades, you know, exercise before agents are really running the show. It's going to happen, but just not yet. And that's why they get pounded for work slop because they overpromise these tech companies as usual about what their great technology will do. But it shouldn't stop us from being like there are there there are parts of AI that small businesses can be using to just help with productivity to make our people more productive. And my final comment to you Lauren is because yeah and again this is what I do when I speak. U we all people ask for advice. I'm like, you know, cough up the money and get freaking trained on this stuff, you know, like we always blow this off. You know, we all know we're using 10% of what Office does, you know? I mean, there are a lot of AI functionalities with C-Pilot. You don't even know how to use it. So, go on LinkedIn, find a Microsoft consultant, you know, somebody that specialized or a chat GPT consultant. They're all out there. You know, pay them what, I don't know, 100 bucks, 200 bucks an hour, whatever. cough up for a few hours of work and have them really teach you about the stuff that you that you own already and that you're only using a small it's like you're driving a car, you know, in Texas in the summer and you're not using the air conditioning because you don't know where the button is. [snorts] So, you know, you [laughter] should be learning this stuff and your employees should be trained on this stuff so that they can be more productive and less fearful of AI and more confident in their jobs and more balanced as well because they'll be getting more stuff done during the day and less stressed hopefully. All I mean AI in small portions will help people do this stuff already. It does have use. So, that's my speech. >> What else are you working on, Jean? What else am I writing this week, Lauren? Um, I haven't decided actually what I'm going to write about for the uh for the inquire this week. It's sort of up in the air. Um, I'm working on a I've been interviewing people about co-ops um as an alternative to ESOPs where if you're a very small business, you can have a cooperatively owned business with your employees. >> I did a podcast about that recently. Great topic. >> You're kidding. >> No. >> Oh, I did not know that. I'll have to look at that. Um, I listen to that. Um because I'm like kind of kneede into it and I guess you talked about there's employee co-ops, right? There's purchasing co-ops, there's consumer co-ops. >> Yeah. All very different, but we talked about worker cooperatives as a, you know, a potential uh succession plan uh exit strategy for a business owner. >> What was your takeaway? >> It was really interesting. I think um I [clears throat] think it's I think it's going to do well. I think it scares people because uh a lot of people emphasize the idea of democracy that uh the worker co-ops result in one man one vote uh situations to run a business and there's a certain amount of confusion about that and uh that scares a lot of business owners especially if they plan on remaining with the company even after they sell it to the uh to the co-op. But the what my understanding and this could be flawed is that the format is very flexible. So if you want everybody [laughter] weighing in and voting on every decision, I don't know why anybody would want that, but you could create that. U but you can structure it the way you want. Um and basically you can have the people you want making the decisions you want them to make. And uh and you can do that in a way that's much less expensive, much less bureaucratic than than an ESOP. Doesn't have the tax advantages, but does have other advantages. >> Yeah, that's I'm finding from the tax side, it does not have the same advantages, but then again, ESOPs have um you know, a lot of disadvantages. You don't want to sugar coat it. There's a lot of costs for an ESOP um and compliance and all that. So, uh that's interesting. Okay, I'm glad you told me. So, yeah, I'm kind of like in the middle of talking to people and have a couple of interviews scheduled next week. So, whatever I learn, I will [music] write about. >> I will look forward to that. >> Thank you. >> Gene Marks is a CPA who writes weekly on small business for the Guardian, The Hill, The Philadelphia Inquirer, The Washington Times, the Chicago Daily Herald, Forbes, and Entrepreneur. You can also hear him on ABC Radio's Eye on the World [music] with John Bachelor. Gene hosts two small business podcasts with Paychecks Corporation, and the Hartford. Thank you, Gan. >> Thank you, Lauren. We'll talk to you soon. >> Have a great weekend, everybody.