Ep 48: Responsible Development of Custom AI Agents and Workflows to Empower Small Businesses with Avi Hacker

[00:00:09] Hi everyone, it's Bob Pulver. Welcome to another episode of Elevate Your AIQ. In this episode, I sit down with Avi Hacker, a dynamic entrepreneur whose journey from finance and law to AI consulting has positioned him as a go-to expert for small and medium-sized businesses embracing AI. Avi shares how custom GPTs and workflow automation can revolutionize operations, enabling companies to work smarter and faster while maintaining ethical oversight through human-in-the-loop systems.

[00:00:37] Avi Hacker, Elevate Your AIQ, Explore AI's potential in data-heavy industries like real estate and the importance of trust and governance in leveraging AI responsibly. Whether you're an AI beginner or looking to scale its use in your business, this conversation is packed with practical advice for anyone looking to integrate AI into their business and needs inspiration for taking the leap into AI-powered efficiency. Enjoy it and thanks for listening.

[00:01:02] Hi everyone, welcome to another episode of Elevate Your AIQ. This is Bob Pulver with me today. I have the pleasure of speaking with Avi Hacker. How are you doing Avi?

[00:01:10] Avi Hacker, Hi Bob, how are you? Thanks so much for having me on the show.

[00:01:13] Absolutely. I appreciate you being here. You and I met recently and I was fascinated by your story and so I thought, you know, by way of introduction, why don't you tell everyone a little bit about yourself and your background, how you got into AI.

[00:01:28] Yeah, thanks so much. So my story is kind of a bit unique in its way that how I got here. I don't have a tech background. I don't have a coding background. I actually didn't really like coding or tech in college.

[00:01:41] So I actually went to college and I got a accounting business management and finance degree and worked at PwC for two years doing risk assurance, doing IT general controls, security automation.

[00:01:53] So I kind of got some flavor of that, but really on like the accounting side and auditing.

[00:01:59] And then I discovered that I really liked interacting with the legal team. So somewhere during that stretch, I made my way over to law school.

[00:02:09] And while I was in law school, I was working for a solo practitioner doing real estate and corporate law, really getting my hands on and really learning the industry.

[00:02:17] And while I was doing all that, this is where the AI piece comes in is ChatGBT was released sometime during that piece.

[00:02:24] And I kind of really realized its technology and how it will revolutionize everything.

[00:02:32] And it kind of just really amazed me. And I was kind of trying to apply it to everything I was doing.

[00:02:39] And I was seeing the potential in the technology, even though it wasn't as good as it is right now.

[00:02:45] And as I was just getting more into it, I was just getting more obsessed with this could really change everything.

[00:02:51] So I kind of just jumped right in while everything was going on, learning everything about every large language model, how it works, how to apply it to really business use cases.

[00:03:01] And industry insights to really be applicable, not just to have fun and play games with it, because I want to leverage it for things I was doing.

[00:03:10] So then towards the end of my 3L of law school, I realized I have a culmination of all these skills from college, from PwC, and now the legal background, the real estate background, and now the AI knowledge.

[00:03:22] And I realized that there's a gap in the market, or there was always companies at the enterprise level always get these new technology and like these integrations.

[00:03:33] But the small to medium sized companies, they kind of just are fending for themselves, or they don't know how to integrate it properly.

[00:03:39] So I really wanted to create my own company and help out that market, the small to medium sized companies to help them leverage AI to really boost their efficiency, boost their efficiency, do whatever they're doing, just 10 times faster, better and cheaper, honestly.

[00:03:56] You really jumped around there.

[00:03:58] So I mean, you don't go to law school.

[00:04:02] That's not a decision you make lately.

[00:04:05] No.

[00:04:05] I imagine.

[00:04:06] So yeah, it must have been a pretty keen interest.

[00:04:09] But so yeah, it's a really, really interesting story.

[00:04:12] And I guess we'll get into this a little bit over the course of the discussion, because responsible AI, as we've talked about, is one of my focus areas.

[00:04:21] And it sounds like with some of your background at PwC and in law, I mean, undoubtedly, you think about compliance and the legalities of AI use and things like that.

[00:04:34] So it's certainly something that I think about when automating and not just tasks, but thinking about how you use agents to talk to other agents in the streamline, like workflows or whatever.

[00:04:48] I start to think about how are you tracking or having the traceability of all this data and how do you make sure that only the right people maintain visibility to certain data elements, certainly personally identifiable information or any sensitive data and things like that.

[00:05:07] So, I mean, any thoughts around that?

[00:05:09] Yeah.

[00:05:10] So I think when I advise my clients and when I look at third-party applications, I always ensure they at least have a SOC to report or some type of ISO certification to ensure that they're having some security around their data controls.

[00:05:24] Because a lot of AI companies are popping up right now and a lot of them are also shutting down.

[00:05:30] So you don't want to just go to any new shiny company that's promising the world where they're not really managing any of the controls, any of your data.

[00:05:37] Like you don't want to put your business information in those type of companies that are just a quick pop shop and then that's it.

[00:05:44] They have your data.

[00:05:45] So you really want to make sure that you know who you're trusting your data with and make sure they have those security controls in place to ensure that your data remains private.

[00:05:54] And regarding the AI agents is, I also advise my clients is, when you're setting up these type of workflows, as much as I automate everything, nothing happens or nothing goes into development or nothing goes to an external application to a client getting emailed before publishing before I review it.

[00:06:17] So I always make sure that the human still has to stay in the loop.

[00:06:21] As much as everything's going to be automated, it's kind of automating the work that doesn't like the research work, like scraping the web and doing an analysis of a market.

[00:06:34] But when you have to really do those critical decisions and the analysis that requires the human component, there always needs to be that human in the loop to make sure everything is running properly.

[00:06:43] There are no mistakes in place. And that's how you really do it ethically, because people are just trying to do it end to end automated.

[00:06:52] And if there's one mistake in the automation, the entire thing is going to be incorrect.

[00:06:58] And if it just gets shipped to development or product like it's then that's where the real issues happen.

[00:07:05] Absolutely. So certainly one of the things that I talk to people about is I try not to be a stickler around like terminology or whatever, but I want people to be really clear about what we what we mean when we say being responsible or we talk about AI governance and things like that.

[00:07:20] It's really doing exactly what you're doing, which is being responsible by design and thinking about these things from the beginning, from conception of, you know, what is this workflow that we're talking about?

[00:07:33] You know, adding automation to what are the steps that no matter what, even if you were doing basic, you know, rules based automation, you know, robotic process automation, any particular step of your entire work lifecycle that would typically get human eyes on it.

[00:07:55] Like, like, you've got to really understand that you haven't broken broken that because then then you run into all kinds of problems.

[00:08:02] But foundational to all of that is is trust. You're going to have people's trust if you expose anything.

[00:08:08] Yeah, like you could right now have your whole inbox automated, get read and draft replies through AI.

[00:08:15] And I think it's very useful if it goes into a draft so you could edit it or in a Google sheet.

[00:08:22] So it's not even a risk of being in your Outlook draft folder.

[00:08:26] But if they even ask me to be like, oh, can you just automate it where it gets sent out without even reviewing it?

[00:08:32] I won't build out that workflow because I think there needs to be a point where, no, like you can't just automate everything and like you still need to review everything that goes out.

[00:08:43] Yeah. So small businesses that you help any particular, you know, industries?

[00:08:51] I mean, are you going back to some of the folks you knew and legal profession or definitely reaching out to the real estate professionals, that industry as a whole?

[00:09:00] Because when I worked in that industry, I really saw a very big gap in people leveraging technology as a whole.

[00:09:09] And I think it's known that real estate is like adopting technology last usually.

[00:09:15] So there's so many applications that could be applicable in the real estate industry.

[00:09:21] And it's such a data rich industry where AI thrives on data.

[00:09:26] You're constantly reading contracts, title reports, doing due diligence.

[00:09:31] So there's so much possibility to really automate a lot of these tasks, especially where everything's written down and so document heavy.

[00:09:40] You're in New Jersey, right?

[00:09:42] Yeah.

[00:09:42] Yeah. I mean, I know in New York, I mean, it's been a while since I bought my house, but that was pretty, pretty painful.

[00:09:51] And there's a lot of different, you know, parties involved.

[00:09:54] There's a lot of documentation.

[00:09:56] It's also a lot of sensitive documentation, right?

[00:09:58] It's got all my banking and financial information.

[00:10:00] It's got employment information and then obviously legal information that's coming from a lot of different sources.

[00:10:07] So I understand on some level why they would be reluctant to apply technology for all of those sensitive reasons.

[00:10:17] But at the same time, I think to your point and the work that you're doing, we've reached the point where if you do it right and you're ethical and responsible about it, you know, there's a lot that you can accomplish to help you, you know, scale.

[00:10:31] I think it could also help you on the back end because as my time working as a paralegal, I saw tons of even while reviewing older contracts or on the web that there's always sometimes issues where people leave clauses in contracts that shouldn't be there or should be there.

[00:10:49] Or maybe something was mentioned in an email that was never applied.

[00:10:52] So I built out some type of workflow and custom GBT where you put in all the emails, you put in all the contracts, you put in all the due diligence, and then just do one final check to review everything to make sure that nothing was mentioned in an email that is in the contract or maybe something that was taken out.

[00:11:12] It just, you have so many opportunities to make sure your work is even better than it was.

[00:11:18] That doesn't affect you.

[00:11:19] So even if you don't want to use it to draft the contract, which I still think is so helpful in of itself to get the first draft done and then you could still edit it.

[00:11:28] But let's say you're taking that approach, why not have it review after you're done to make sure that you didn't miss anything?

[00:11:36] There's just so many ways to start applying it to your company where you don't have to go all in right away.

[00:11:42] Right.

[00:11:43] I guess that raises a question about when you take on a client and you start doing this for them, you go in and build an agent or you apply some automation in certain parts, but then you say, let's get you comfortable with that piece of it.

[00:12:02] And then maybe you come back when they've seen it run through successfully for a while and then you expand on it?

[00:12:09] Or how might a typical engagement work?

[00:12:10] Yeah.

[00:12:11] Yeah.

[00:12:11] What I like doing with a client is I kind of go on an intro call and see if we're a good fit.

[00:12:17] And if we decide we are, we kind of like map out the entire business and see what are like the different areas that they would need help with.

[00:12:25] And there's so many applications out there and so many third party tools that it's very overwhelming.

[00:12:31] And I've went through a lot of them.

[00:12:34] So I could really have good insights into which one will be helpful for their specific circumstance.

[00:12:41] So I'll help them start integrating the tools that are right for them just to start leveraging AI properly.

[00:12:48] And then I like meeting with clients at least once a week, at least for an hour, just to get them acquainted with the tools.

[00:12:57] Because I find that I had a few people that they just wanted the tools and then they didn't want to continue like with going for an hour a week.

[00:13:05] And I just find that they don't really do with it.

[00:13:09] They don't do anything with it at the end of the day because it's so overwhelming and it's so much information to learn that you really need someone besides you.

[00:13:18] Like showing you be like, okay, you had this task to do, but you also have this tool.

[00:13:23] Did you try to use it?

[00:13:25] Did you try to leverage it?

[00:13:26] Did you try to think of outside the box and how to make it better and faster?

[00:13:30] So you kind of need someone by your side for at least a little bit to get that lift to really start thinking differently.

[00:13:36] So then once we're comfortable with like the basic tools, then we'll start automating more.

[00:13:42] Because once you start automating things, even before they have a basic understanding of how to use a custom chatbot or how to interact with it, it just then it starts getting very messy.

[00:13:53] And that's where things start to go wrong.

[00:13:55] Right, right.

[00:13:56] No, that makes sense.

[00:13:57] Are they more of the mind like set it and forget it?

[00:14:01] Or do they actually learn how to do some tweaking themselves?

[00:14:07] I mean, are you sort of upskilling them in the process?

[00:14:09] Yes.

[00:14:10] So there's certain areas where I meet a client and I'll always build them.

[00:14:16] The first thing I'll do is I'll build them a personal custom GBT that's tailored to them as a personal assistant.

[00:14:22] Because I always say I think everyone should have one.

[00:14:25] It's kind of like everything on my website, everything on my LinkedIn profile, my resume, everything that I'm doing.

[00:14:31] And it's like I give it instructions that it's my personal assistant.

[00:14:35] So now when I need to draft emails or fill out forms or do all these questionnaires that take a lot of time, it's kind of I could take a screenshot of it, put it in my custom assistant, be like, OK, fill it out for me.

[00:14:47] Because it's my personal assistant.

[00:14:48] And you don't realize how much time this takes up on an average day of just this type of work.

[00:14:54] And it's kind of like my second brain that knows everything I'm doing so I can leverage it for everything.

[00:14:59] And that like right there is such a big lift.

[00:15:02] And then I'll start integrating certain CRM tools.

[00:15:07] A lot of people, believe it or not, don't have CRM dashboards.

[00:15:11] So their information is all over the place.

[00:15:14] So I'll integrate certain applications that's applicable to them and then have them start learning how to use it better.

[00:15:21] Because unless they want me to stay in their company indefinitely, they do need to know how to use it.

[00:15:27] So I give them step by step tools and like for them to go off and watch the trainings and then to see how they're using it next week and then to keep on going and learning.

[00:15:40] Right.

[00:15:41] You've sort of empowered them with this custom, you know, GPT that they now have available to use.

[00:15:48] You're doing a custom chat GPT?

[00:15:50] Yeah, so it depends on the application because I find that the different large language models have different pros and cons to them.

[00:15:58] And also chat GPTs, you could accompany them with actions, which could then be leveraged to make .com automation.

[00:16:05] So it's really gets into the need of the client.

[00:16:10] And for more writing tasks, writing based tasks, I like using Claude because I find it just much more, I guess, in the way it interacts for like for blog posts or LinkedIn posts.

[00:16:25] I like using Claude.

[00:16:26] So there are certain differentiations between them that based on the client I like using.

[00:16:31] But for the customization, I mean, do you, can you do that in Claude or you have to build like an artifact or how does that?

[00:16:37] Yeah, you just build a project.

[00:16:40] It's similar to a custom GPT where you just open up a new one, write the instructions, give it a knowledge base and then give it a name.

[00:16:49] And it's kind of there you have your own custom GPT in Claude.

[00:16:52] Okay, got it.

[00:16:53] And you do that.

[00:16:55] It's persistent across devices, across like your, as long as you're in your own account, like you can switch back and forth between mobile and desktop.

[00:17:03] Yeah, everything updates seamlessly like and it's very efficient.

[00:17:08] And actually, I like doing them with chat GPT.

[00:17:13] So on your phone, you're able to speak to your custom bot.

[00:17:18] So sometimes if I have documents on my computer that I want to interact with my custom GPT, I will upload those documents to the chat on my computer and then click the speaking button on my phone.

[00:17:30] So then I could have an interaction talking to my custom GPT.

[00:17:34] Now it's uploaded with my documents that I wanted to speak about from my phone.

[00:17:39] So that's where I would likely use chat GPT in that case because I wanted to have an interactive conversation as opposed to just typing back and forth.

[00:17:48] Wow, that's really cool.

[00:17:49] So with the voice interaction, could you then, if that was connected on the back end to your CRM, could you just ask it to, you know, sort of plan your...

[00:18:02] Before we move on, I need to let you know about my friend Mark Pfeffer and his show, People Tech.

[00:18:08] If you're looking for the latest on product development, marketing, funding, big deals happening in talent acquisition, HR, HCM, that's the show you need to listen to.

[00:18:21] Go to the Work Defined Network, search up People Tech, Mark Pfeffer.

[00:18:25] You can find them anywhere.

[00:18:28] You're weak for all your upcoming like sales meetings or pursuits or whatever?

[00:18:32] Yeah, you're able to integrate it into your Calendee or your CRM or ask it for information through a custom API action where it will go into your CRM, look up all the information.

[00:18:44] It gets a bit complex with the coding and the authorization keys, but essentially you're able to just go in there, ask it whatever you want, and it will answer you based on your availability.

[00:18:55] I actually set up a similar aspect for my company.

[00:19:01] I used an application, VAPI, where I bought a phone number, I wrote out the instructions for a custom helper, and I gave it, uploaded it with a knowledge base of how to leverage AI for a book that I wrote that's on my website.

[00:19:19] And then you could call the number and just start asking questions about AI and your company, and it'll give you tailored advice.

[00:19:28] So I could also hook that up, which I haven't done yet, to my Calendee, which it will then say, okay, if you want more information, I could set you up an appointment with Avi Hacker, and then it will book the appointment for you.

[00:19:40] So we're getting to a point where it's getting pretty crazy with the technology, and I'm definitely having a lot of fun with it because the sky's the limit in what I'm seeing.

[00:19:53] And every day I'm just exploring more and more and more, and I just love building out these really cool features that could be so helpful because right now I'm on my own.

[00:20:03] And anytime I think of hiring someone or paying someone to do it, I could essentially just build my own custom GBT or build out the workflow for myself.

[00:20:14] So I don't need to do it.

[00:20:17] So I really do 1 million things on a daily basis that I would not be able to do if AI wasn't around.

[00:20:25] Yeah, and I can certainly appreciate that scenario.

[00:20:29] You and I have talked about that.

[00:20:31] Being a solopreneur is not easy.

[00:20:33] You've got to wear many hats, and like you said, there are so many hours, only so many hours in the day.

[00:20:39] And I just can't help but think I'm missing out on opportunities because I haven't kept track of all of this.

[00:20:48] Or like you said before, maybe there's something embedded in an email that maybe I overlooked because somebody asked me three different questions in the same email.

[00:20:58] And I got two of them, but I forgot about the third or things like that.

[00:21:03] And it's like, it would be awesome to have another pair of eyes, some other entity that could basically make sure things don't fall through the cracks.

[00:21:11] Yeah, and one thing that I just, I did like a whole rabbit hole, like a wormhole, whatever it's called yesterday.

[00:21:19] So I built out a pretty good, what I thought was a pretty good prospecting make.com automation where if someone would book a calendar, calendar into my CRM, then I would do some prospecting through Folk that would get their description and information and kind of just see a brief description about them before on the call.

[00:21:41] And then I built out this workflow through Relevance AI where they had a certain template, but I made certain adjustments where you kind of just type in the person's name and the company.

[00:21:54] And then it does 12 automatic steps where it researches the web about them.

[00:22:00] It gets their company information from their website.

[00:22:03] It scrapes their LinkedIn, their company profile.

[00:22:06] Then it goes into articles that may have been written about them.

[00:22:10] It'll summarize like anything that happened to them online and then provide me a full detailed report under like a few minutes.

[00:22:19] And then it also has all of my company information and saying, okay, based on this client's company, their hobbies, their interests.

[00:22:28] And then this is the best way to approach the conversation.

[00:22:31] So doing this over and over and over again, it's unmatched in the information that you could obtain as one person.

[00:22:39] So it's your virtual assistant.

[00:22:41] It's your secretary.

[00:22:43] It's your market research assistant.

[00:22:46] It's all of these things all basically in one.

[00:22:49] I'm trying to literally automate.

[00:22:52] And I publicly say it like I post about and I show all my workflows that every single part of my company.

[00:22:58] So it's basically me sitting here orchestrating, I guess, 20 different quote unquote entities working for me, like me delegating my tasks while I oversee, review, edit, publish and post.

[00:23:15] Because anything that gets written, it's all AI generated.

[00:23:20] And people say the whole thing of it not being authentic.

[00:23:24] I think it's a misconception around it because the ideas behind my posts, the way the emotion is conveyed, that's all me.

[00:23:34] Like a lot of people create AI generated posts and it's terrible.

[00:23:40] So.

[00:23:42] I still have to think of the idea to be like, I want to post about this today.

[00:23:46] I want you to add these details.

[00:23:48] So if I wasn't adding what it would say, it wouldn't be a good post.

[00:23:53] Yeah, this has come up more and more frequently in my conversations and a lot of what I've been reading.

[00:23:59] Everything from kids in high school writing college essays all the way to job interviews and assessments to, you know, obviously what we see prevalent online now.

[00:24:12] And it's true.

[00:24:13] You can tell that there's like emotion and empathy and creativity like missing if it's just pure AI in the way that it's telling.

[00:24:24] Its grammar is great.

[00:24:27] Yeah.

[00:24:27] And it gets, you know, factually it may or may not, you know, be there, but there's just something about human storytelling that is clearly missing.

[00:24:39] And I think that's why, I mean, I just was on a webinar recently where college admissions people were talking about this.

[00:24:45] Like they say, you know, they can spot, you know, an essay written by AI.

[00:24:50] And I'm always apprehensive because you can have a lot of false positives and I don't want them to discount anyone's application if English isn't their first language or they're neurodiverse or, you know, whatever challenges they have where they may have used AI as a tutor or proofreader or something like that.

[00:25:08] So I'm sensitive to some of these issues.

[00:25:11] But on some level, I understand what they're saying when they say they can spot it because it just sounds very formulaic and just kind of dull.

[00:25:22] Yeah.

[00:25:22] Like I could spot it a million miles away.

[00:25:25] And when I read, when I read newspapers or articles, like I'm so good at spotting it just because like I read it every single day, all day.

[00:25:34] So and everything that goes out, like to what I mentioned before, I would say it's I'm getting it to the point where it's 85 percent there.

[00:25:42] So I still need to edit it.

[00:25:45] I still need to make those final touches.

[00:25:47] It's still never going out without my review and without me reading it over and editing it.

[00:25:55] So it's like I made a post and I actually did pretty well.

[00:25:58] It's like what's the difference between like a speaker having their speech written for them or a book writer having a book like they wrote the book.

[00:26:06] But then an editor comes and just fixes up everything and the grammar.

[00:26:10] So like what's the real difference?

[00:26:12] It's because I'm not hiring someone, let's say, one hundred thousand dollars a year to be my personal editor for my post.

[00:26:20] OK, so I have a I do it.

[00:26:22] So like what is the real difference?

[00:26:24] It's really helping me convey my message better.

[00:26:27] Yeah, I mean.

[00:26:29] Again, this so I spent a lot of time in the talent acquisition space and this comes up all the time about what are the appropriate uses of AI in the job and not just college application process, but more specifically in the job application process.

[00:26:42] And some talent acquisition teams have rejected some applications because they said AI wrote this this resume is like, are you applying for a resume writer job?

[00:26:53] I mean, doesn't everyone outsource that task or a lot of people outsource that task?

[00:26:58] That's why there's a whole cottage industry around resume writers and people who write cover letters for you or whatever.

[00:27:03] It's you have all that experience.

[00:27:06] You have all the skills that are being described.

[00:27:08] Describe if unless my job, unless writing is part of the job, like as a journalist or writer, like then that should not be considered a writing sample.

[00:27:20] This is me trying to put my best self forward.

[00:27:24] And yes, I had help in that it is not cheating.

[00:27:29] Now, there are other issues just getting into the responsibility aspects of this.

[00:27:35] Part of it, I think, is about appropriate use.

[00:27:38] And if you were going to outsource it to someone else anyway, what's the difference if you outsource it to a resume expert or an AI that's been trained to do that and is capable of doing it?

[00:27:49] The other is, especially when it comes to blogging and cranking out lots of content over and over, maybe on a daily basis.

[00:27:57] Hi there.

[00:27:57] I'm Peter Zollman.

[00:27:59] I'm a co-host of the Inside Job Boards and Recruitment Marketplaces podcast.

[00:28:03] And I'm Steven Rothberg.

[00:28:05] And I guess that makes me the other co-host.

[00:28:06] Every other week, we're joined by guests from the world's leading job sites.

[00:28:11] Together, we analyze news about general niche and aggregator job board and Recruitment Marketplaces sites.

[00:28:17] Make sure you sign up and subscribe today.

[00:28:21] Like be transparent, right?

[00:28:23] You should be transparent about the use.

[00:28:25] If you're cranking out a newsletter that's growing and it's, you know, 95% or more written by AI, like, I don't know, people should know that.

[00:28:36] Yeah.

[00:28:37] Yeah.

[00:28:37] And I play with the idea.

[00:28:39] So I run an AI-generated podcast that it's an AI host that gives news on AI and it's fully AI-generated.

[00:28:47] So, like, I play with the concept.

[00:28:50] The host acknowledges that it's an AI and it makes jokes around reporting on AI.

[00:28:56] And I just reached 1,000 listeners two days ago.

[00:29:00] And, like, in the past three days, 300 more listeners tuned in.

[00:29:05] And it's really, like, taking off.

[00:29:07] Because what I find that people are using AI to play off as being human.

[00:29:15] And I think that's where the discrepancy lies.

[00:29:17] And it's you could use AI, but they don't try to be like, okay, this is me.

[00:29:23] This is, like, my human self.

[00:29:25] So, like, I'm creating AI-generated news.

[00:29:28] And, yeah, it's created by AI.

[00:29:30] It's not me.

[00:29:32] And, yeah.

[00:29:33] And I think one of the things that, like, I don't like so much right now is the human clones that are getting so good of the person.

[00:29:42] Like, I would never use a clone of myself to replicate in a meeting representing me.

[00:29:48] Will I use a clone of an AI saying that he's Avi's assistant?

[00:29:53] Yeah, because that's not me.

[00:29:54] Like, acknowledging it's an AI assistant.

[00:29:57] But where the clones are becoming so good at me being an AI, that's where I think it goes a little too far.

[00:30:04] I mean, I can understand why people would want to do that.

[00:30:08] Especially if you're not created public speaking or something like that.

[00:30:11] You really could just have a digital twin.

[00:30:15] You feed it a script, and it basically looks and sounds, you know, like you.

[00:30:19] But I think, you know, someone like you would obviously spot that pretty far away.

[00:30:24] Because I listened to my first AI podcast recently that, I don't know if you're a car guy.

[00:30:31] I was looking at a car, I'm bringing a trailer.

[00:30:33] And he basically took everything, the comments, the pictures, the back and forth in there and the bids and everything.

[00:30:41] And he put it into something.

[00:30:43] And then, like 10 minutes later, he had like a full podcast episode talking about the pros and cons of this car and what should we watch out for.

[00:30:51] And so it did research on the car and the reliability of the car, that particular year, make, model.

[00:30:58] And then, of course, took into account that it actually also went out and grabbed, used the link to the car facts report and pulled that back too.

[00:31:06] Yeah, that's very cool.

[00:31:07] It wasn't perfect.

[00:31:08] And one of the things that caught my attention was there was one particular maintenance concern that people were talking about in the comments.

[00:31:18] And the podcast never talked about it.

[00:31:20] So the seller basically clarified and should have quelled any concerns about that particular warning light that was on.

[00:31:30] Right.

[00:31:31] So I say this with that level of detail because I've seen in other places where people are getting very concerned about AI's ability to properly summarize larger corpuses of text.

[00:31:47] Yeah.

[00:31:48] Of text, right?

[00:31:49] And if you, if there's, there could be some important context that you overlook.

[00:31:53] So I imagine, I want to ask you that specifically because with some of your legal background, either in law in general or through real estate or whatever,

[00:32:01] if you're summarizing some things and you're overlooking something that could give you, that could leave you exposed to some level of risk, like you got to be careful.

[00:32:13] Yeah, that's why I always say it should be used as a first draft if you're summarizing anything and not for you not to look at it at all.

[00:32:23] And it also comes into play of how good your instructions are going to be to extract exactly what you're looking for.

[00:32:31] Because I made a bot that's going to import, extract 15 important points from a loan document.

[00:32:37] And if you're looking at these 15 important points, every single time you get a loan document, I drafted very comprehensive instructions to go through the entire 80 page agreement, whatever it is, to extract these same 15 points every single time,

[00:32:52] as opposed to you getting someone less, less AI savvy that might just be like, oh, can you please extract the important points?

[00:33:01] So the AI doesn't know what important points are.

[00:33:05] Maybe it does have some context just because it was trained on it, but maybe that's not what you're looking for.

[00:33:10] So you really have to get in detail and describe exactly what you want.

[00:33:15] Give it the appropriate context.

[00:33:17] And the more information you provide it, the more instructions, the better your output will be.

[00:33:23] Yeah, absolutely.

[00:33:23] And that's where also the different large language models come into play because some of them have larger context windows.

[00:33:30] Some of them have smaller, some of them have better memory.

[00:33:33] So you have to know based on the type of task you're working with.

[00:33:38] Like Gemini, I'm pretty sure has over 2 million context tokens at this point.

[00:33:43] And Claude has over, I forget, like losing track because they're all changing all the time.

[00:33:48] But if you're using certain LLMs that are maybe smaller, you try to give it a thousand page document,

[00:33:56] you should know that it's not going to read all of it because it just can't.

[00:34:02] So you have to know which tool to use and when.

[00:34:04] That would explain why some of my podcast episode summaries, auto-generated show notes from this platform are not complete.

[00:34:17] I had one that went longer than usual.

[00:34:21] And then this whole summary, it was like it only transcribed and summarized the first half hour.

[00:34:29] Yeah.

[00:34:30] I was like, what do you mean?

[00:34:32] It gave timestamps with topics, so those timestamps.

[00:34:36] And the conversation was like an hour.

[00:34:38] And the last topic was like a minute 28.

[00:34:41] I'm like, what do you mean?

[00:34:43] What happened to the rest?

[00:34:44] Yeah.

[00:34:45] So yeah, these are not perfect.

[00:34:47] I will say I have been playing with Gemini.

[00:34:49] You know, it may have a large context window, but I haven't been all that thrilled with some of its output.

[00:34:57] Personally, yeah.

[00:34:59] I, my main ones right now are Claude, ChatGP, Perplexity.

[00:35:04] Those are kind of like my base ones that I go to for everything.

[00:35:07] And then if I have certain other tasks that I might, like I just leveraged Gemini yesterday for something that I had a lot of information that I wanted to just put an RSS feed from like all these tags and stuff for certain podcast episodes to just put them in a table and organize it and structure the data.

[00:35:27] But it was so large that it didn't even fit in Gemini.

[00:35:30] So what I did was I just broke it up into three separate chunks and did it three times.

[00:35:35] So it just knowing the capabilities and not just putting it all in the tool and being like, okay, now it's wrong.

[00:35:41] And AI is not helpful.

[00:35:42] It just, no, it just passed its capabilities.

[00:35:44] Yeah.

[00:35:45] I just wish it would acknowledge its own limitations sometimes or where it sort of stopped.

[00:35:51] Well, to the scenario I was talking about before, like because of your context window, you only were able to summarize the first half hour.

[00:36:00] Tell me that.

[00:36:00] Because if it looks like you completed the job, when in fact you've kind of phoned it in, like I'm just going to move on.

[00:36:09] But I guess it goes back to your primary point, which is the human in the loop.

[00:36:14] You've got to pay attention to what it's doing and don't just take everything at face value.

[00:36:19] Yeah.

[00:36:19] Yeah.

[00:36:20] Like when I speak to someone, it's like, oh, the AI said it could do this.

[00:36:23] But then I replied on an email saying, but it can't.

[00:36:26] So it's like, but he said, why did it say it could?

[00:36:30] It just, it does.

[00:36:31] It's very confident in its answers and people need to be aware of that.

[00:36:34] And that's where the expertise, like the domain expertise really comes into play because you could ask it to, let's say, draft a contract.

[00:36:44] And it might look like a really good contract if you're not a lawyer.

[00:36:47] But if you are a lawyer, you might notice certain clauses are missing.

[00:36:51] So you really still have to have that expertise to know if the output is still good or not.

[00:36:57] I also noticed, I guess, to use a courtroom term, you can sort of lead the witness.

[00:37:03] Yeah.

[00:37:04] Right.

[00:37:04] Like I can say just yesterday, I was on Claude on my phone and I said, how do I save this chat?

[00:37:12] And it said, oh, just click this icon in the top.

[00:37:14] Right.

[00:37:15] And I was like, no, that is to start a new chat.

[00:37:18] If I click that right now, this conversation would be over and I would have nothing.

[00:37:22] I said, do you mean this one?

[00:37:24] I'm so sorry about that.

[00:37:25] You're absolutely right.

[00:37:26] It's that icon on the bottom.

[00:37:27] Well, it wasn't that icon either.

[00:37:29] So it's just like feeding off of what you give it.

[00:37:34] And I appreciate the apology, but you can't just keep telling me I'm right when, in fact,

[00:37:40] I'm not right.

[00:37:42] So, you know, I don't know when we'll get there.

[00:37:46] It might be scary when we do, but at the same time, it just shows that we're still early days.

[00:37:51] Yeah.

[00:37:52] That's why there always has to be someone that really knows how to leverage it and knows how to properly integrate it into a company because the people using it might think, okay, great.

[00:38:04] They have AI right now and they don't have to do anything anymore.

[00:38:08] It's like, give me AI so I don't have to do my job anymore.

[00:38:11] And it's like, sorry, but like, you're still going to have to do your job.

[00:38:14] Right.

[00:38:15] Your job may change, but you're still going to have to do it.

[00:38:17] Yeah, it's going to change, but you're still going to have to do it and it just, you can leverage it to do things faster and better.

[00:38:24] But that doesn't mean you're like, you could just sit back, relax.

[00:38:28] Like I still work right now because I'm also building my own company.

[00:38:32] So I'm also doing that much more.

[00:38:33] I still work like almost a 12 hour day, 14 hour day, every single day.

[00:38:38] But it's like, oh, aren't you using AI so you're not doing everything?

[00:38:41] It's no, I'm, I'm personally using AI so I could get that much more done.

[00:38:46] Right.

[00:38:46] So instead of 18 hour days, it's a 12 hour days.

[00:38:50] Yeah.

[00:38:51] And also I'm in the beginning of the stages of like a startup for my company.

[00:38:55] So it's, I could have like 15, like I said, like 15 to 20 different, essentially that would be people working for me.

[00:39:05] But it just right now it's AI.

[00:39:06] So I could be in control of everything right now until it gets to a certain point where maybe the technology will be where I could hire an AI manager for my other AIs.

[00:39:16] But that will be in the future because Claude just released the computer use where I could kind of control your computer.

[00:39:22] And I did a video of trying to use it for business use cases.

[00:39:28] And honestly, it was pretty bad.

[00:39:30] Like it was keep on breaking and it couldn't really do stuff.

[00:39:33] But I think it was very cool to see where the technology was going, even though it couldn't do anything that I wanted to do right now.

[00:39:42] Yeah, I saw, I was reading Ethan Mollick's post about that.

[00:39:46] And yeah, he ran into some technical difficulties as well.

[00:39:50] Yeah, he was constantly giving me errors and like wasn't really working timing out.

[00:39:53] But even to get that far, it's pretty impressive.

[00:39:56] Yeah, exactly.

[00:39:57] That was his sentiment.

[00:39:59] Avi, this has been great.

[00:40:00] What's the best way for people to reach you if they want to get your help?

[00:40:05] Yeah, you can reach out to me on LinkedIn.

[00:40:08] You'll see me there every day.

[00:40:09] I'm posting very exciting stuff that I'm doing.

[00:40:13] You can check out my website on the AIconsultingnetwork.com.

[00:40:16] I do my blog there.

[00:40:18] If you want to listen to my AI podcast, it's AI Gen Daily.

[00:40:23] And yeah, reach out to me, connect.

[00:40:25] And if you have any questions, I'll be more than happy to help out.

[00:40:29] Sounds great.

[00:40:29] And I'll include those links in the show notes for this episode when it gets published.

[00:40:34] So this is great.

[00:40:35] Avi, any final advice for our AI laggards and newbies?

[00:40:41] I would say just start playing around with it and get started.

[00:40:46] A lot of people are very scared of the technology and don't know its capabilities.

[00:40:50] So if you don't know how to do something and people reach out to me and be like, okay, how

[00:40:57] could I do it?

[00:40:58] Why not just try to ask AI first and see what it gives you?

[00:41:02] And just try it.

[00:41:04] If you need to analyze Google Sheets or Excel documents and you're not sure if it's capable,

[00:41:10] like the only way I find out if it's capable is if I try it.

[00:41:14] So try to do all your tasks with it.

[00:41:18] See where it takes you.

[00:41:18] See its capabilities.

[00:41:19] And I'm telling you, you will be amazed on how much it could actually do.

[00:41:24] Yeah, absolutely.

[00:41:26] Avi, thanks so much for your time today.

[00:41:28] And I think there's a lot of important insights for my audience.

[00:41:33] So thank you again.

[00:41:34] Yeah, thank you.

[00:41:35] Thank you for having me.

[00:41:36] Have a good day.

[00:41:37] Absolutely.

[00:41:38] All right.

[00:41:38] Thanks, everyone, for listening.

[00:41:39] We'll see you next time.