029 The Deepfake Accountant

Hey, we're do a little different. It's like a, like a fun Friday show. Something a little different. But I think ultimately we'll be very relevant to us and the ways that we worked. Worked, work. Amazing. Absolutely amazing interview I watched. It's a new thing that came out recently with the guy who was made famous from the Tom Cruise deep fixed during the pandemic. Diving deep on kind of the psychology of deep fakes, what identity means going forward and what our identity ends up looking like as subject matter experts in professional services. Super cool stuff. Let's talk about it. Jason Daly. All right, I'll put a link to this in the show notes. Amazing interview. Probably worth watching after this. Also, if you haven't yet watched the episode we did on Chatbots and how easy it is to build your own and how I kind of talk through how kind of proprietary context I think is kind of the currency of the future. As the mass tools get bigger and more of what we do gets productized, it's worth checking that episode out first. Episode 20 Put this AI to work today. So deepfakes, what do deep fakes have to do with what you do? So they do like an hour long interview with the guy. If you haven't seen it during the pandemic, this is a couple few years ago now. He did some deepfake videos where he was acting as Tom Cruise and then and the technology they used overlays Tom Cruise face on his. And for most people it was indistinguishable from Tom Cruise himself. And so obviously this kicked off this sort of media craze of holy smokes, what does this all mean? What is like, you know, how do we ensure the legitimacy of what we see online and all that. And that was a few years ago and obviously now that's all becoming massive. Between the ability to generate absurd amounts of content on the fly with AI, the ability to create really compelling, you know, like the, the pope and the puffy jacket that we got a month ago and like the AI generated images of Trump being arrested and stuff like that. That's all getting more problematic. Another big development there has been in music now you may have seen there's like some AI generated Kanye and Jay Z songs that to me, especially the Jay Z one that I heard for me is indistinguishable from an actual song. So if we have these artists whose voice are out there and there's just a huge amount of music that models can be trained on and now anyone can create music with these people's voices, either with their permission or without their permission. What does the future of music look like? And so, like, an interesting spin on this is the artist Grimes. While everybody's scared about this and the label, the recording labels are like, don't quite know what to do. Grimes came out, she's an artist, came out and said, I'll give you a 50% royalty on anything that you create with my voice. So, like, she went the opposite approach and said, go ahead, like, make stuff with my voice and you'll get a royalty on it. But specifically, this interview with the guy behind the Tom Cruise Deep Fakes had some really amazing anecdotes. And it's a good example of Deepfakes, I think, are a great example of an AI thing that is like, kind of perceived as evil and can be used in many evil ways, but can also be used in some amazing ways. And that's kind of like the theme with AI, right? Is kind of like the Internet. It will enable a bunch of amazing things and a bunch of awful things. So how do you get the best of both worlds? But one of the anecdotes from this interview that really stuck with me, this the guy behind the Deepfake videos. They now have a creative studio and they've gone on to do work on a bunch of movies. And there's a film coming out on Netflix soon where Tom Hanks is basically de aged to. So Hanks did all of the acting, but he will look like a much younger version of himself throughout the movie. And the quality of this stuff's amazing. But one of the really interesting anecdotes from this interview was creating the ability to talk with people from the past or people who have died. And so someone was talking about the idea of talking with, say, your great grandparent. Visually, we kind of have the ability to do that depending on the imagery that we have of these people. And you could kind of manufacture an identity there based on what you know about them, but it wouldn't be very high fidelity because of the limited imagery you have of them and. And the limited context you have of them, of the things that they have said and done and things as simple as what does their voice sound like and what sort of vernacular did they use when they talk and all that. Right. So the further back in time you go, the lower fidelity that version of that is going to be. But that is a We are living in a period of time where for human beings that were, that will never be the case again. So how many emails have you written in your Lifetime. If you think about that going all the way back. We are better documented human beings. Maybe better is not the right word. More documented human beings than any who have ever come before us. Like a really obvious example of this is the number of photos I have of me as a child versus the number of photos that my kids have of them as children. Like, they'll have this amazing, like much higher fidelity, you know, by day digital photo book that they can go back through and like see their whole development and growing up with their siblings and all that stuff, as opposed to what I have, which is like a, you know, physical album of just kind of those sort of like milestone moments. Right. But everything that we're doing is creating context. So context being like, how does this person talk? What are their thoughts on certain subjects, that sort of thing. And we talked about this in the past, that other episode, episode 20, about how being a more documented person potentially gives you advantage in the AI age, because you can build things on top of that context. Chatbots being an example. But looking ahead to the future, if my great great grandchildren wanted to create a version of me that they could talk to, they absolutely could, right? Because of this video, because of the transcripts from everything I've created from how well documented I am, from all of the emails that I've sent, they can not only go back and talk to me, they can go back and talk to a 60 year old version of me. They can go back and talk to a 25 year old version of me, me, because I am so documented. And we're not all documented to the same degree right now. But the, the idea that these sort of like on demand avatars are becoming more realistic and more accessible, I do think impacts the ways that we work and the way that we communicate going forward. Right now we already have really good chatbot technology and you can limit those chatbots to specific context. So I'll pull up a conversation I had with Ed Zollers on, on Twitter the other day. He was talking with Ashley, the CPA about like tax research with chat GPT and how ultimately, because it draws on the entirety of the Internet, there's some good information and there's some bad information in there. I hopped into the discussion and this is kind of my thoughts on an interesting direction for all of this right now. I said, ed, a really exciting version of this is pairing the capabilities of GPT with context created by thought leaders, context being the underlying source material. It's possible now with GPT4 to reliably get it to only Reference the context you provide it with. So if you feed it the transcript from your Corporate Transparency act talk, Ed's doing a two hour talk on that next week, I think, and you ask it how to make ice cream, it'll tell you it doesn't know this wasn't possible. With 3.5, it would make up an answer. So in the previous model of GPT, you couldn't reliably restrict it to a certain set of context. If it didn't see the answer in the context you gave it, it would fall back on its general knowledge. Which isn't what you want, right? You want to be able to put a chatbot on rails to only give responses based on the context you provide it. Continuing here, as you said, pairing authoritative text with articles from trusted sources like thought leaders could make for a fantastic assistant. So Ed's an example of somebody who's got a blog where he's done a ton of writing and a chat experience that is only connected to that trusted source. The person that I trust that I'm willing to let advise me in decision making. That's something I would love to have access to. And in many ways this is like, you can call it a deep fake. I mean, maybe it's not. Maybe you don't see them talking and you don't hear them and all of that. But let's say we took the transcripts from the talks Ed gave and plugged his blog in there and all of that. I am talking well and truly with a version of that person. And in the same way that it's hard for me to talk with a version of my great great grandfather, it's going to be much easier for my grandkids to talk with a version of me because of the context they'll have access to. So up inside that little noodle of yours, what's the context that you have that's most valuable to a certain type of person? Because the way that we communicate right now is largely like one to one, you know, low leverage. And I talk about this stuff a lot. Like the difference between exclusively doing one to one work versus doing one to many work. The show is a great example of a wildly efficient way of sharing my thoughts on something because one more person can come watch it and it doesn't require any incremental effort from me, right? Like it's available on video, it's available on podcasts. It's a very high leverage way for me to talk about the stuff that I think is important. And traditional communication like emailing with a client, very, very low leverage. But a video that's about a specific topic or a blog that's about a specific topic, not a lot of specificity there. And so on one end of the spectrum, you have content that is about a very specific thing that may or may not be relevant to what somebody needs to hear. On the other end, you have sitting down and having a conversation with that person and historically, nothing in between. Right. And so for most people, if they don't create any content, literally the only way to get an answer out of them is talking with them, sending them an email, having a phone call, something like that. The people who are producing content around certain subjects, great. At least then around those subjects, people can come and consume that stuff and get an answer as long as it's relevant to what they're doing. But that's it. And that's a, it's a big gap between, you don't ever need to talk to me because I've written 100 blog posts and the information they can get from that and sitting down with the person and AI atop your context kind of, in my mind, sort of bridges that gap. Right. And so if the piece de resistance is sitting down and having lunch with you, surely there's a version of your identity, your avatar, your expertise, short of sitting down and having lunch with you, that has value. Right? You see where this is going? So, good example. Matt Metrus, met him on Twitter, super authoritative digital currency, web3guy. He knows way more about this stuff than I do. He knows the ins and outs of, of what stuff there is clear rules for and what stuff there isn't clear rules for. So even if I had a perfect tax research bot, the fact is there's a lot of situations where the rules for how to do that thing just don't exist. And in life, I mean, there's a hundred examples of this. Most things don't have an absolute answer. If somebody asks me what cloud practice management system should I use? It depends, right? It's very firm, specific. So in, in cases where there is like an absolute black and white answer, like fine, AI will help us find those answers faster and like the general purpose tools will do that well. And this is where we need like our tax research partners to develop better tools for getting to answers faster. But what about all the situations where there's not a black and white answer or where you want that specific person's take? This is where Matt, as someone who works like almost exclusively with web3 types of clients, has had to make a bunch of those Hard, squishy decisions. And for me, if I'm a tax professional, boy, I would sure love to have a Matt on my team. Like, when it comes to making a hard, squishy decision on the treatment of a specific gray area, I don't have that expertise. And so if I can sit down and have a call with Matt and say, hey, here's my client situation and this and that and what do you like? What's the best thing to do? I will absolutely go with what he says. The right answer is. So in a perfect world, yeah, we all have, like, we can clone ourselves and we have those people on our team and we have everything that we could possibly need to get to those answers. But an example of like a thought leader becoming more accessible through AI is like, Matt. He does a whole bunch of talks around the country on digital assets and all that. Give me the transcripts from those talks from blogs, maybe even building, like, Matt building some additional context to fill in the gaps. And maybe that's just voice memos or sitting down and writing a thousand words on this thing that maybe he hasn't documented yet. That is a snapshot of sort of his thoughts on this nuanced thing right now. And all of that comes together to this body of context that if you wanted, you could license, you could give people access to that. And it's not the real deal. It isn't. But GPT4 is good enough to only give answers based on that context and cite the source. So even if you were afraid that from the context to what it gives you, if there was some sort of miscommunication in that transmission, it'll even point you to the original source so you can get verbatim what it was the person said. Now, usually when it comes to answering questions, it's actually pulling a combination of maybe several things that they said to get to that kind of fleshed out synthesized answer, but it can point you to those sources too. And so if there's five different things that, oh, Matt gave this talk in Ohio here that said this one time in this other talk at this other conference that said this this one time. But if I combine those three things, it looks like this is the answer. AI does that for you, and it will point you back to the source for those things. So if I can't have Matt as a team member in my firm who just sits in a chair all day long until those like three or four sticky client situations come through a year where I really need that expertise, if he's not going to Sit in that chair until I need him. This feels like the next best thing, right? So not necessarily for the general tax knowledge that is clear cut and black and white in the U.S. i think of that as like quick finder stuff. Quick finder is like a really good concise guide for just all the little fiddly things that you have to remember that like, you're not gonna commit to memory because there's just so many little details. That stuff's black and white, but for the stuff that's not black and white, which is advising is what we do is the value of what we do. Having that person on speed dial in some form or another, that's really valuable. And I know, I know we are all very scared of AI hallucinating and saying things wrong, but it's worth understanding the experience you have with ChatGPT, which sits on top of an incomprehensibly large volume of information that we have no control over and can't see into. That AI is very different from the AI where we do know what it's drawing from and it can't draw from anything else. And it gives you a citation right next to that statement to say, hey, here's where I pulled this thing from. And these sources were curated by this person that you trust. That seems pretty solid to me. I know there's a lot of trust with AI issues, but I have a hard time seeing how that goes sideways, how the AI gives you something that's fundamentally incorrect and it will only get better from here. So if you can see all the sources and all of that, that seems really cool to me. So then if we go back to all of the ways in which you could communicate with a subject matter expert, and I'm picking on Matt here, talking with Matt on Zoom, having lunch, that's the best, right? That's the highest fidelity. Reading a blog post, that's still better than nothing. Not having a blog post, that's a bummer. Then I'm like, there's no help there. A chatbot on top of context that Matt has curated, that's way better than a blog post right now. You too are a subject matter expert. And ultimately most small business owners, they want a relationship with somebody who can understand their nuance and their context and all of that. And so I don't see a future where there's is not really a place for like the human advisor. And not everybody is that way, especially like younger entrepreneurs. They just want to get stuff done with a product when that's all they need and it can be frustrating sometimes to have to go and ask a human for advice because the product wouldn't like take care of that for them. But ultimately most people still want that personal relationship. And there still has to be an element of accountability. Somebody's got to come in to save the day when something goes sideways. And so if you filed a tax return with AI and something was wrong, what do you do? And so that aspect of accountability I think is where like that personal touch will be necessary for the foreseeable future. But if you had two accounting firms, and I've used this analogy before with cas, if you have two similar accounting firms and you run a small business and you come to one and one only does tax, and you can't tell the difference between how that guy does your tax return versus how the lady across the street does your tax return. If you have a hard time finding bookkeeping staff and the other firm will do your bookkeeping also, they're probably going to go to the other firm because they can get their accounting and their tax solved all at one shop. They've got no idea if one shop does it better than the other. But just because the other offers it, they're probably going to go to that one. And if you think about this whole notion of this secondary level of access to expertise, the thing that is all you can eat, the thing that they can diy, the thing that they can do in the middle of the night, the thing that has zero latency because they get an answer immediately. If you compare firm A to firm B and firm A has this first party source of context that they have built around the types of clients that they serve and their specific expertise. And clients have access to that, not for free, but they have access to that and can get immediate answers, can have conversations with it. It won't replace the really technical questions or the questions that fall outside of that context or maybe just the big decisions. I don't see that replacing the one on one conversations with the advisor. But most of the questions you get from clients are not that. They're not all high stakes, really sweaty questions. They're can I do an IRA this year? I read this article and that article and I don't quite understand what's happening or something about HSA or like it's, it's not always going to be super technical stuff. But I do think the future of communication, and I get this sounds really sci fi, is not exclusively going to be sitting down with that person. I think there's a version of communication that's actually enabled by your context. And maybe, maybe a helpful framing for this is me and what I do. So I shared in that last video, like one of the interesting things to me about doing a daily show is the context that will come from it. Like, I am as well documented on my thoughts as virtually anybody and I do think going forward that's going to become more and more of an asset. And so, you know, people ask for a lot of things from me, from technology companies wanting me to help them with, you know, sort of, sort of like AI technological stuff, like how to implement certain things. What things are going to be the most viable to accountants. Then you got a lot of accountants on the other side who are like, do you do any consulting or coaching, that sort of thing. And I have to just like, as a rule I have to say no to virtually all of that stuff because that one on one work for me takes away from this, which is the one to many work, which is the highest leverage thing I can possibly do with my time. So just as kind of a life choice, this is where I invest my time. Because I think long term that's best for me and it's best for the largest number of people. But I now have this like wealth of context of all the things that I've talked about and tweeted about and made videos about and all that stuff. So taking my context, for example, how would you build a product around that? Well, people are gonna want different things from you. And so I don't think it's like a one size fits all sort of thing. So for example, I have a bunch of public content that anybody can go consume, but that doesn't mean that you should give it away for free. I heard Derek Sivers, he's an author, talking in an interview where he's published a number of books. But he's got like this, this blog that's really popular and he does all this writing online. But the books are just like tidied up, rolled up versions of this writing and content that he has done elsewhere. And it's stuff where if you wanted to go out and find it, you, you could, you could go out and find all of this same information in a bunch of different places online. But the value of the book is it's all in one place and somebody can read to it from beginning through it, from beginning to end and that's just easier. So you charge for the book, right? So I have this public body of work that I could enable access to through a chatbot. I mean, you Just think about, goodness sakes, even just the show, how much content there is. Like, how are you going to go back and find a statement that I made in a show about a specific thing? I can't even do it, goodness sakes. So the idea of like a chatbot that can see into all of that stuff and you have a conversation with it, it's limited to the context of the stuff that I've said and then it pulls back an answer according to that in a single interface, obviously that's wildly more efficient than having to dig through, oh my gosh, every little thing I've said in a newsletter and Twitter and all of that stuff. And so there's this public body of work that you could absolutely build something on top of, but there's also like another layer there where you could build it atop private stuff that you develop. So let's say that people want to know more about my thoughts on like what's the right cloud practice management system right now? Or you know, how to, how to build a team to better sell advisory, like specific stuff like that. I could absolutely. I have a public body of context, right? And that's all the stuff that anybody can go out and find on YouTube, on Twitter, newsletters, stuff like that, that's a public body of context. I can develop other bodies of context that are proprietary to me that only exist within the context behind that AI. So if it's around like certain very specific types of things that people want to ask me about, that's like another, I guess, like body of context that you could license to people. And so giving somebody access to something like this that you've developed, it could, it could even be like a granular thing where there's like different packages and different sorts of things that you can pull in. It'll be interesting longer term. I think the best versions of these things, when you are kind of like developing an assistant that's as helpful as possible for you, probably gives you the ability to pull in multiple sources of context. So like, if I wanted to make the ultimate tax assistant that anybody in my team could use, I could pull like a general tax thing from Quick Finder where there's like a general body of context there that the bot can access. Maybe I also want to layer in mattmetric bottom maybe I wanna layer in something from the AICPA that's helpful or something from the Oregon Department of Revenue or something like that. Longer term, I think you'll be able to build assistants that pull in multiple sources of context. But specifically, as we Think about this through the lens of how we develop our firms and how we monetize the expertise that we have going forward. I think the big takeaway here is one on one sitting down. Having a live conversation probably won't be the only option. It sounds sci fi to call it an avatar, but there is a version of you that is short of sitting down with you that can provide a lot of value to you. And like this video is like, it's an example of that. But if you think past that, but, but something short of sitting down with the person, if that person has a body of context that the AI can be trained on. I think we all have this expertise in our head that if well documented is valuable to people and your clients can access that 24,7 with no incremental amount of effort from you. Just think about if you rolled this thing out in your firm tomorrow and 50% of your clients would use it and it covered half the questions that they would send you. You just knocked down the questions you got to handle by 25%. That's valuable, right? I think what's stopping most people from doing that right now is a lack of context, is that we aren't that well documented and context exists in places. You've obviously sent a lot of emails, stuff like that. So things could be built for from that. But it's not necessarily like intentionally developed context that you could then like build a chatbot on top of. And this is yet another reason why I think wading into content creation, even though none of us identify as influencers and that sort of thing is yet another reason why I think wading into content creation is really valuable because it makes you a more documented person and it forces you to think, think through how you actually feel about something. We all have these conflicting thoughts in our head and it's not until you take the time to write them and articulate them that they will all kind of fit together. And as we're headed to a future where there's gonna be chatbots absolutely everywhere and AI solutions for this and that, what people would love to have is Steve Bot U Bot. If you rolled out a bot tomorrow that your clients could ask questions of and not all these chatbot tools just use GPT4 I think so I've shared ask my AI before. I think it's using 3.5, which is really, really hard to restrict to a context. I've had better luck with Chat Thing, Chat Thing IO. I think the middle tier of that will give you GPT4. I've put that on some context. And if I ask it how to make ice cream, it says it doesn't know. But if you put this kind of avatar of yourself out there, it has value. And there's some aspect of like our self worth where maybe that could be an uncomfortable thing to put out there and say that like, hey, here's this version of me that I actually think that you can get value from. Because that's there's like this weird sort of narcissistic, I don't know, thing that may be hard for us to kind of process, but the future looks like clients working with a deep fake of you. I genuinely believe that it's a version of you that is informed by the things that you said that is infinitely scalable. And one way or another, I think we're really close to a version of that. And that is more like what professional services and thought leadership look like longer term. Weird, right? Weird but exciting, but also weird. What do you think? We got pretty sci fi today. I linked that Tom Cruise deepfake interview in the show notes. Check that out. It is very like mind opening when it comes to the notion of what is identity going forward. If somebody can just create something that looks like anybody else and sounds exactly like anybody else, that's scary that they could do it. But also does that mean that I should be doing it? Does it mean that I should be more mindful of what context I could put on top of that? So that there's an essence of me that people could draw value from asynchronously. Like, you know, a Jason Bot that Anybody can talk with 24, 7 and ask questions of and all that. Right. What domain expertise do you have that you can give people access to in a higher fidelity way than a list of blog posts? I think that's what that comes down to. Cool stuff. Hey, have a lovely weekend. Thanks for coming and hanging and I'll see you Monday.