Two Big AI Developments for Accountants
Attention: This is a machine-generated transcript. As such, there may be spelling, grammar, and accuracy errors throughout. Thank you for your understanding!
Blake Oliver: [00:00:09] Hello and welcome to The Cloud Accounting Podcast. Happy Friday. Happy Saint Patrick's Day to you, David.
David Leary: [00:00:16] My my green shirt.
Blake Oliver: [00:00:17] Yes.
David Leary: [00:00:18] This last minute impromptu- didn't comb my hair yet like Saint Patty's Day special show.
Blake Oliver: [00:00:24] Well, we got to do it because there have been some huge developments in artificial intelligence this week that were obscured to us by the collapse. And we got to catch up on them, let you know what's going on. Two big ones. First, we had GPT-4 released by OpenAI. And while GPT-4 is not mind blowingly different than GPT three in my humble opinion, in the demo that OpenAI did, for millions of people who watched it, they ended that demo with a tax prep question, and so-
David Leary: [00:01:01] This is why we're recording, because both things were very personal to our listener base. Oh yeah. Openai talked about taxes and then Microsoft released their Copilot and we can talk about that, but they're both going to impact our listeners. It's just not. Let's talk about I like these are very specific things that occurred.
Blake Oliver: [00:01:19] Yes. And this is not in years we're talking right now. I mean, people are already starting to use GPT chat GPT for answering questions, understanding the tax code. And this demo really shows how you can do that. The potential for it is enormous and it's going to change how people think about tax prep and accountants. And we got to give the credit where credit is due to Jason Stats, who tweeted this and has been doing a fantastic analysis on Twitter about all of this. So do follow Jason stats and you'll get the latest. In terms of AI. Right now I want to play for you the video. And this is the the end, the very end of the demo that OpenAI did for the public of ChatGPT version four. Let's take a look.
OpenAI Demo: [00:02:12] Last thing I'm going to show is how to work with the system to accomplish a task that none of us like to do. But we all have to. So you may have guessed the thing we're going to do is taxes. Now note that GPT is not a certified tax professional, nor am I. So you should always check with your your tax advisor. But it can be helpful to understand some dense content to just be able to empower yourself to to be able to sort of solve problems and get a get a handle on what's happening when you could not otherwise. So once again, I'll do a system message. In this case, I'm going to tell it that it's tax GPT, which is not a specific thing that we've trained into this model. You can be very creative if you want with the system message to really get the model in the mood of what is your job, what are you supposed to do? So I've pasted in. The tax code is about 16 pages worth of tax code. And there's this question about Alice and Bob. They got married at one point, and here are their their incomes and they take a standard deduction. They're filing jointly. So first question, what is their standard deduction for 2018? So while the model is chugging, I'm going to solve this problem by hand to show you what's involved. So the standard deduction is the basic standard deduction plus the additional the basic one is 200% for joint return of subparagraph C, which is here. Okay. So additional doesn't apply. The limitation doesn't apply. Okay.
Blake Oliver: [00:03:36] So I just want to point out that what he's done is he's copied the relevant tax code into Chatgpt so that it has it as as part of its response. So now it's he's asking it to interpret that.
David Leary: [00:03:49] And right now he's manually scrolling through it and thinking out loud or talking out loud.
Blake Oliver: [00:03:56] Yeah. Like how how would a human approach solving this problem or answering this question these apply.
OpenAI Demo: [00:04:02] Oh wait, special rules for taxable year 2018, which is the one we care about through 2025. You have to substitute 12,000 for 3000. So 200% of 12,000. 24,000 is the final answer. If you notice, the model got to the same conclusion and you can actually read through its explanation. And to tell you the truth, the first time I tried to approach this problem myself, I could not figure it out.
Blake Oliver: [00:04:31] I gotta just say, that's amazing to me that an AI researcher. The guys, the guy building OpenAI is saying the tax code is so complicated that he couldn't figure this out. That's a credit to our profession, honestly. Right.
OpenAI Demo: [00:04:48] I spent half an hour reading through the tax code trying to figure out this back reference and why there's subparagraph, just what's even going on. It was only by asking the model to spell out its reasoning. And then I followed along that I was like, Oh, I get it now I understand how this works. And so that, I think, is where the power of the system lies. It's not perfect, but neither are you. And together it's this amplifying tool that lets you just reach new heights. And you can go further. You can say, okay, now calculate their total liability. And here we go. It's doing the calculation.
David Leary: [00:05:29] It's typing across the screen. Type. Type, type, yes.
Blake Oliver: [00:05:31] So for our podcast listeners, the AI is now doing a detailed written analysis of the tax liability of this hypothetical couple.
OpenAI Demo: [00:05:43] Honestly, I every time it does it, it's just it's amazing. This model is so good at mental math.
Blake Oliver: [00:05:49] And it came out with, let's see, what did it say? The. It said. That. The couple's taxable liability is tax liability is. $10,597.68 for Alice and Bob in the hypothetical situation that it was given.
David Leary: [00:06:12] So now now what's scary about this is. Everybody. The understanding we're having now is it's he even said it. It's not perfect, but the way it presents the information, like if you go back to that video like it, it writes up the whole like more documentation and explanation than you ever gotten from your accountant. Right. And so it almost creates the impression that has the authority. And that's the scary thing about this is it's it's almost. It's that tall. Paradigm where the people that are most sure usually are the people that are the most wrong. Right. And it's like, you know, those those chemical weapons that were in Iraq. Right. They were so sure about it. And it's kind of feels like that. It's like it's so confident in its answer that you're going to be like, obviously, this is correct. Yeah.
Blake Oliver: [00:07:08] And Christopher, who's joined us on the live stream, says conclusion they didn't withhold enough. Yeah, that's right. I guess so I my headline on our live stream for that video was OpenAI demonstrates tax planning with chat GPT But really that wasn't planning that was just calculations that was tax prep. And so there's a big opportunity here. David you know, you were basically getting to this point, which is that the tax prep is going to be more and more and more automated, and the answer is going to be more and more automated, which is great for accountants because we don't want to have to write that long email to the client explaining all this stuff, which is why we don't do that. We don't have time for that and the AI can do that and we can check its work. And I spotted an interesting app, actually, the founder of this app reached out to me on Twitter and wanted me to take a look at it, so I got to share it. It's called zero Tax AI. This is essentially chatgpt wrapped in their own app and you can ask it tax questions and it's free to ask it tax questions. But if you want a if you want a professional, a tax professional to review the answer, it costs $5.
David Leary: [00:08:26] So you pay other. That other site we had on a couple of weeks ago where they had the. Your is cast, right? You could ask the cast questions and then. Or you could get a human instead.
Blake Oliver: [00:08:37] Yeah. So I asked the founder who reviews the answers and they didn't give me an actual, like, real response as to who reviews the answers. So I'll let you know if I do find out. Carlos says, Can Chatgpt pass the CPA exam? Now, that's interesting because I saw when Chatgpt came out, I saw a list of all the exams that it can pass- uh, GPT-4 pass and it's a lot of them, but like, it can pass the bar exam. That was one of them. Let's see what it can do. Yeah, it can pass. Ap classes. It can pass the bar exam, it can pass AP chemistry, it can do the GRE. But they didn't run it through the CPA exam. Like none of the none of the sites that I've seen have run it through the CPA exam. So I wonder, I bet it could write like if it can pass the bar [CROSSTALK]
David Leary: [00:09:36] - if you can upload the data, if you- basically here's all the information. I don't see why you can't do it.
Blake Oliver: [00:09:43] Michael says. Remember when Bot Keeper was so confident in their bookkeeping ability? Well, that takes me back to the days. Remember the Bot Keeper days, David? That was when that was when bot keeper was advertising itself as artificial intelligence. And they were. They were not exactly forthcoming about the fact that it was humans in the Philippines. My friend Michael Lee likes to call that. He likes to call that kind of AI. Asian intelligence. Intelligence? Yeah. They were using they were using AI, but not the AI you're thinking of. Yeah, but so but so this is a that's a good point, right? Which is that this AI is is true. Ai now. Right. It's not it's not that fake stuff that we've been sold where it's really just human beings on the back end doing all this stuff. This is true generative AI that's creating unique responses that have an accuracy, you know, problem. But so do human answers. And it's really good and it's getting better and better and better. And this is version four. We just had version three like a month ago, it feels like. And now we're at four. Imagine what it's going to be able to do in just a few more months or years.
David Leary: [00:10:53] In this version, they're not disclosing any more. They've they've closed it. It's no longer open API. That's the other big version, this version four, they're not disclosing how they've trained this model the way they have the previous ones, which tells me they're probably starting to get in bed with, you know, maybe Intuit is going to make an investment and they're going to have some data model tied to this, like they're going to the data these systems get trained on is going to kind of determine which system you use a little bit. Um, somebody brought I heard somebody relate this to like varietals of grapes for wines, right? People are going to start figuring out which, which service they like the best and go with that. I think that's kind of crazy to me is. How come Alexa and Siri? These voice chat things never got any better. I know, right? This seemingly came out of nowhere in the last like three and a half months. Well.
Blake Oliver: [00:11:53] And that's I mean, that's because Apple has not always been on the forefront of AI development. Siri has always kind of sucked compared to Google's assistant and. I just think it's a matter of time until Google's got a plug chatgpt into their Google assistant. And that's going to be a huge win for Android. And I bet you right now at Apple, they're scrambling to figure out how do we get an AI to plug into Siri? Because once they do that, imagine if you could just ask Siri to do all those things that it can't do like that you just had given up on trying to get it to do. I mean, call it booking appointments for you. Just ask the phone, you know, can you book my medical appointment? And it will go and look through your emails and find out who your doctor is and call the office and look at your calendar and book the appointment for you and put it on your calendar Like there's no reason an AI couldn't do that at this point. Which is. Yeah, I want to highlight a comment. I want to highlight a comment here from Ryan. Ryan Wade says, I will impact all compliance tasks in the next few years. The question we are asking is whether CPA firms should wait for the tax software products to integrate AI or if we should be experimenting on our own to get ahead. Well, Ryan, given the past performance of the tax software products, I think you're going to be waiting a long time before they integrate. I considering that they still can't do fillable PDFs and proper client portals and.
David Leary: [00:13:18] But does this kill the portal? Does this kill the onboarding form? Right. Somebody can as soon as they come to you. You just they start telling you their tax situation and then based off that, they start asking questions, Oh, can you email this? Can you upload this? I need this, I need this. Like that becomes the portal. And everybody's experience is super unique to them instead of a generic PDF. Yeah, I still haven't onboarded with my new accountant. Why? There's a pdf sitting in my inbox that I do not feel it filling out.
Blake Oliver: [00:13:48] Yeah, I think the tax organizer could disappear and it could just be a chat bot that asks you questions because.
David Leary: [00:13:55] Because then it can just keep harassing you until it gets the answer it needs. And then you don't have your staff doing that.
Blake Oliver: [00:14:00] Yeah. Now the thing is somebody one of these client portal solutions, client hub or AE or carbon or just name all the practice, financial sense, name all these tools they're going to have to integrate chatgpt to do that because you've got to give it prompts, right? And so if you're the CPA and you're just using Chatgpt to do this, it doesn't really make sense because you're having to create a prompt, put it into the bot, get the answer back, send it to your client. But if the messaging tools could do this for us and then you just give it a list of documentation from last year's return, give it last year's return and it can go and ask for what it needs, what you need to do the next year's return. I think that is totally within the capabilities of this tool, just taking the right integrations and prompts. And it could it could talk to you.
David Leary: [00:14:50] It has the tax code. And you you say, here's all like, what do I need? What data do I need? Do I need to to process this tax return? Yeah. Like now and then it's like, great, go get it from the client. I think there are some steps in that that I think could be here sooner than later. And this really pulls back to the Microsoft announcements. Yeah.
Blake Oliver: [00:15:13] So I mentioned that, you know, I get a lot of my news about Chatgpt from Jason Stats and Carlos asked what other tasks can Chatgpt do to help us in the accounting world? Well, here's an example from Jason. Uh, let me get this on the screen here. Okay. So Jason said on Twitter at CPE. Check this out. I'm using GPT-4 to help me categorize ambiguous accounting transactions, how it works in a swipeable prompt for you to do the same. And he has a screenshot of a table that. Chatgpt is created with column A, merchant name, column B confidence, then category, then background, and he shows he demonstrates how he popped into the. The chat bot like a list of bank statement lines. And then he asked it using a prompt to to to try and categorize it and explain why. And it did it. So for instance, it took file stack https w-w-w dot file and it said with 95% confidence. This is software for file handling and management API and it put that into the table. And it said for puppets. Papal puppets and then it has a phone number. Ca That was the bank statement line. It said puppets with 85% accuracy as office supplies Handmade puppets for professionals. So it goes up and it does the research. It goes and looks online and it figures.
David Leary: [00:16:47] This is with no database. Now, if you take this skill and you put millions and billions, probably billions of bank feeds that have been categorized in something like QuickBooks Online. This becomes very, very good.
Blake Oliver: [00:17:02] Yeah. David Hall says, I asked Chatgpt for compliance related information and it could not reference the source documents or references. A lot of compliance is hard to locate or not readily available. This is the challenge I had with it. I asked it to do some answer some questions for me based on the tax code and. You have to give it the information. If it doesn't have it in the database already in its training set. So I'm not sure if like GPT or OpenAI has put the tax code into it. I bet they have now that they demoed it on, you know, in the capabilities. But you have to guide it. It's not it doesn't do it like perfectly on its own. And of course you know that's that's the opportunity for professionals is if you guide the chat bot then. You know, you can evaluate its outputs. It's like, think about it.
David Leary: [00:17:54] That's how they're going to monetize this, right? Right. Like, oh, Intuit. That way into it can have their version of it built on their data set. Like if you want to use it with no data set, it's free and you can upload your 1000 lines of data. One one little interaction at a time. If you want to offer this as a real service, you're going to pay a lot of money, which is kind of because what's going to happen Big Four is going to go do this, right? Is this going to leave out? A set of until somebody like you Intuit and Zero and these other companies, is it going to leave out a bunch of accountants from being able to utilize this in their firm because all the big players are going to pay to get exclusive use and upload against their data?
Blake Oliver: [00:18:37] Yeah, there's going to be an arms race now, right, for who can do this. But thankfully there is an API, right? People are going to be able to plug into Gpt4 and do this stuff. So I mean, we can at least do it with Zapier. Like we can do it with make small firms are going to be able to do this and probably can do it faster than the big firms because they just take so long to do everything. So I think there's an enormous opportunity here. But the real opportunity, the biggest opportunity is for the practice management solutions to integrate AI into everything, to automate all of the messaging and all of the stuff. We got a lot of chats coming in, Joseph says. Greetings from Nairobi, Kenya, guys. I uploaded the entire Income Tax Act for Kenya to Gpt4 then I gave it a tax problem to solve. It was not anywhere close to accurate. Yeah, so again, right. Questionable accuracy on some of these things. I think the reason that it worked well in the OpenAI demo is they kind of cheated and they took just the portion of the tax code that is relevant to the problem they asked. So they kind of primed it, if you think about it that way, right. It wasn't having to go through the entire tax code and figure out the answer. They gave it just the relevant portions. And I think that's what you have to do right now. But maybe five and six will be able to take the whole thing and figure it out.
David Leary: [00:19:52] It feels like training wheels a little bit.
Blake Oliver: [00:19:54] Billy says AI and automation in accounting is extremely overhyped. People have been saying that accounting will be automated since the invention of the calculator, the computer, the ERP system, Excel, etcetera. Well, yes, and it has right Excel automated the paper spreadsheets that accountants used to have to do, and they were called spreadsheets because you literally took a giant piece of paper and spread it out on the table and you had to manually do all the calculations. And it automated that entire job. And there used to be hundreds of thousands of people who that's all they did. And the people who did the calculating and, you know, the green shades and all that. And yes, it all gets automated, but then the jobs that are created are much more fulfilling right now. Instead of being a little literal paper spreadsheet jockey, you're an Excel jockey and you can do the work of ten people or 100 people or just a work that wasn't even possible to do before. So I'm not worried about this taking our jobs, especially with the talent shortage that we've got. There's like no chance this is taking our jobs. I mean, it'll take your job if what you're doing is that kind of stuff, right? If what you're doing is, is the stuff that's going to get automated for sure. But if you learn how to use the the app, it's not going to take your job. Yeah.
David Leary: [00:21:07] So this could be an extra credit for you. But like somebody asking Matthew is asking, can Chatgpt finish someone's 30 extra credit hours for the CPE license? So at this point for you this week like you could actually settle the argument about the 150 completely. You could give up the studies, the data, you could put it all in starting salaries, the cost of college. And let Chatgpt summarize whether or not the hundred and 50 hours is necessary.
Blake Oliver: [00:21:32] Maybe, maybe I'll do that. I'll I'll ask it to come up with a plan. Please come up with an eight point plan to solve the accounting talent shortage. Exactly.
David Leary: [00:21:40] So do you want to jump over to the Microsoft part of this thing? Because that was the other big announcement yesterday. Yeah.
Blake Oliver: [00:21:46] Ryan put in a link and I'm going to go ahead and take a look at that. But we have another announcement from Microsoft. Yes. And so that that announcement, you have a video here to play, right, David Yeah.
David Leary: [00:21:58] So do you want.
Blake Oliver: [00:21:59] To tee it up at all.
David Leary: [00:22:00] Announced Copilot and Copilot is basically AI in all the parts of Office 360, Microsoft 365. I guess that's what's the proper name is now. And so if you can imagine, you are using the Microsoft Teams and in the middle of the meeting you can say, summarize what our meeting is about and then we can take that and then say make a slide deck from that and oh, and then email this, create an email or start that email. Some of these things you're talking about like, like book my meeting, right? These things are being added into Microsoft Office 36- or Microsoft 365 as functions. And the video Blake's going to play as actually the Excel part of this.
Blake Oliver: [00:22:38] Okay, let me play this.
Microsoft Copilot Presentation: [00:22:44] Copilot in Excel helps you make sense of all your data. Say you need to analyze this quarter's sales results. You start by asking Copilot to analyze the data and give you three key trends. Within seconds, you've got what you need, but you want to drill in. You ask Copilot a follow up question about one of the trends. Copilot creates a new sheet giving you a sandbox to play in and helping you better understand what's happening. You ask Copilot to visualize what contributed to the decline in sales growth this period. Copilot adds a little color to make the problem jump off the page. Now you want to dig deeper and ask a follow up question with a what if scenario. Copilot not only answers your question, it creates a simple model and even asks if you want to learn more about what it did. With a step by step breakdown. Finally, you can ask it to create a graph of your projected model. Copilot in Excel turned a sea of data into clear insights and actions.
Blake Oliver: [00:23:50] Wow. I mean, so basically what it did in that video for those who aren't watching it or are listening is we fed it, the demo, fed it, a data set, and then asked it to analyze that and it was able to figure out that there was a decline in sales and represent that with a chart and a graph. And then it was able to dig in specifically to like the cause of it and provide its rationale in bullet point form. I mean, it's doing financial analysis right there.
David Leary: [00:24:23] And you could do this. It might it might have taken you an hour and a half. And it's just it's like it's truly an assistant, right? It's doing all this for you to go faster. Yeah. It's not replacing you. But if you're going to have to start using this as a as truly a Copilot, like it's called my big impact, I step back and look at that and I'm like, Oh, all these Excel wizards and all these YouTube channels with 700,000 views on how to like write a script to separate people's first names from last names and put them in two separate columns. You're just going to ask it to do it and it's just going to do it. Like, you don't have to learn formulas anymore. You don't have to actually learn how to use Excel.
Blake Oliver: [00:25:02] I disagree because it's the same thing as with the the tax code tax prep tax planning demo, where you got to be able to validate what it's doing and check the work. So if you don't know how these formulas work. In general, right? If you don't, you may not have to know exactly how to do it. The exact syntax and all that stuff and all the different little nitty gritty parts. But if you need to understand the general concept around each formula because you need to be able to dig in and see like, does this make sense? Right? And so. It's not going to take somebody who doesn't know anything and make them into an Excel genius. I mean, it might take somebody like me who knows a little bit about Excel to be dangerous and make me really dangerous. Right. That's yes. And it's a great yeah, it's a great way to learn.
David Leary: [00:25:47] Be dangerous.
Blake Oliver: [00:25:47] Yeah, it's a great way to learn because it's like, okay, I, I taught myself how to do index match finally, like last year. But it was not easy to learn from the documentation. Now, what I could do is I could say, Hey, Excel, will you, you know, do this thing with index match that I don't really remember how to do with the syntax and then explain how you did it. And I could learn that way, Right? It's actually going to be a really powerful teaching tool. I mean, it already is. Like I'm using Chatgpt to ask questions and get answers to things. I'm trying to think like that. That would have taken me just so much research. I feel like a lot of people are using that now. I mean, if you're not using it right now in your job as just a way to brainstorm new possibilities and new ideas, you're missing out.
David Leary: [00:26:35] Yeah, that's one of the things in the other Microsoft video that they did not have. There's no real audio in it, but they were talking about how it's it's a starting point right. It'll it'll get your you could take a word doc and say turn to a presentation and boom now it's a presentation and it's like helps you get started on a lot of work. Right.
Blake Oliver: [00:26:56] So, so basically what Microsoft did is they turned. Chatgpt into a sidebar A widget in all of the different office apps. So like in word you're saying it can it can do a first draft for you, it can bring in information, it can add content to existing documents, summarize text and rewrite sections of the entire document to make it more. And then imagine.
David Leary: [00:27:17] For for end users instead of you typing like or like that demo for the taxis, like I uploaded the tax code, you'll be starting to say like summarize this and in a pick list you'll pick the word doc. And then using this Excel spreadsheet and these three email addresses, history of email use that. And so you're actually dynamically clicking and linking to these docs and then it pumps out your output tailor super deep integration.
Blake Oliver: [00:27:43] Taylor Thanks for joining us. Taylor says, I'm glad I caught this live. I love this show. Yes, thank you. And if you want to catch us live, follow me on LinkedIn, follow The Cloud Accounting Podcast on YouTube or rather subscribe to us on YouTube and you'll get notified when we go live. And you can you can chat with us. Ask us your questions. We'd love to hear from you. Thanks, everyone. Did you mention PowerPoint, David? What you can do in PowerPoint now Transform existing written documents into decks, complete with speaker notes and sources, or start a new presentation from a simple prompt or outline. So like example commands and prompts, you can try create a five slide presentation based on a word document and include relevant stock photos. Consolidate this presentation into a three slide summary. Reformat these three bullets into three columns, each with a picture. Oh, here's the big one. Copilot and outlook. Summarize the emails I missed while I was out last week. Flag any important items? I mean, this is one of the biggest challenges I have. Like when I'm heads down, I might get a bunch of emails and I need to go and figure out what do I need to respond to today. In order to do that, I have to read all the emails, but the AI could help me with a pretty high confidence rating. Figure out what's important.
David Leary: [00:29:01] The and I don't use outlook anymore. And outlook was a king for years and years and years. But like these these features they're adding to outlook is probably worth switching back to outlook for and this is why I think the Microsoft stuff's really interesting because the vast majority of accounting firms are in the Office 365 stack. They're going to get this in their hands very soon, which that is going to instantly benefit accounting firms. Um.
Blake Oliver: [00:29:26] There's a little video here. I don't think I can play it with the audio, but I just want to see. Oh, it's a gif. Okay, we're good. So what's happening right now is I'm going to try to narrate so that you've triggered the autopilot and asked it to create a draft of an email and then it creates the draft and you insert it and you modify it and you send it like, this is exactly what I was hoping they would build. That's amazing. And it's ask Lily to be a last minute presenter for some sort of event. And, and it writes this beautiful email with the everything. What a time saver.
David Leary: [00:29:59] And this and these these these videos and gifs that Microsoft have feel very real. It does not feel like vaporware, which tells me this is pretty close to being out there on the market and being live and just released out there.
Blake Oliver: [00:30:12] Copilot in teams I want to highlight this Copilot in teams says summarize what I missed in the meeting what points have been made so far. Where do we disagree on this topic. Those are prompts you can actually try. Create a table of pros and cons for the topic being discussed. What else should we consider before making a decision? You can ask it what decisions were made and what are some suggested next steps? Wow.
David Leary: [00:30:39] So the other news now, there's no videos or anything exciting per se, but QuickBooks and Intuit had an article in Accounting today, this week about AI. And essentially they are going. Quote unquote, all in on AI in 2023, which I think you have to say, because you're, you know, you're a public company and everybody needs to get on the AI bandwagon. But Michael Hitchcock, who is the director of product manager of accounting and tax at QuickBooks, he's confident that QuickBooks can achieve what would be effectively, quote unquote, hands free accounting that can run with little to no human intervention. And they're talking about how it used to take customers six months to fully train an AI model. And now basically when you get your QuickBooks and we kind of had a video of this at QuickBooks Connect, it asks you 5 or 6 questions. And because of that, it's able to do all your transaction categorizations right? You're training the model quicker and they're going to roll this out for our listeners accounts and bookkeepers, they're rolling out an AI guided end of month review feature that will not only will summarize the financial information, but it also go and figure out what tasks that need to be automated and completed and then curated to do lists on what tasks have yet to be completed. So, for example, transactions that are not categorized, it's going to flag those and bring them to your attention. Try to take guesses, probably to categorize them like Jason Stats demo, and then you just review as the accountant. So this is we're going to see this stuff in QuickBooks and Xero. It's going to be in everything and how good people implement it is going to be the next test because everybody's going to have I and now it's going to be filtering you. It's going to take people three years to figure out like God, that apps horrible at it. I don't know what they did when they implement their A I bet it's bad because I can't imagine every app has somebody with the skill set to roll out a proper AI product.
Blake Oliver: [00:32:41] Yeah, it's going to take really good product management, really good design experience. Um, it, it's not going to be like automatic unless you, I guess, I guess you could just ask the AI like how do I integrate chat GPT-4 into my app and it'll tell you exactly how to do it. Maybe it'll even program it for you.
David Leary: [00:33:00] It'll write the code down to yes, right, You can get code written. So yeah.
Blake Oliver: [00:33:03] But you still got to have the vision of what it's going to do, what the experience is going to be, Jeremy said. How do we get our hands on Copilot? So I'm looking at the Microsoft blog, Jeremy, and it says, At the bottom, in the months ahead, we're bringing Copilot to all our productivity apps, Word Excel, PowerPoint, outlook teams, blah, blah, blah. We'll share more on pricing and licensing soon. So oh, they've they're also adding it to Dynamics 365 as a Copilot in their CRM and ERP products. Wow, that could be really big for ERP. So it just says coming months, it's coming. So it'll be here soon. I would be surprised if it wasn't here in Q2. Christopher says auto reconcile would be nice if the double entries tie. That's always the question, right? Somebody's got to check the work. Well, uh. Okay. So we talked about Gpt4. We talked about Microsoft 365 Copilot. What else in AI did we miss this week?
David Leary: [00:34:13] Yeah, QuickBooks came out and I was just looking at other stories. I think that was.
Blake Oliver: [00:34:20] Is that it? Did we get it all?
David Leary: [00:34:23] I think the big the other story here that's happened this week is this shift of OpenAI being open to private. And I think the founder, one of the founders, he's trying to say like they were wrong before by making it open sourced and open. And it's hard to tell if he's just saying that because they want to be private and make billions of dollars now, or he kind of like gives this weird warning of like the rest of you might not realize it for years down the road. It's almost like they know more than everybody else. We've realized something about AI, this AI we're building. And if we don't make it private. Like it could be exploited in the future. Like it's really interesting the opinion that's out there. Yeah.
Blake Oliver: [00:35:07] Well, so one reason, one argument in favor of keeping it private. Is the danger that it poses to society. And this is from The New York Times. The New York Times journalist got access to GPT-4 to test it. And in his previous test of GPT three, that was the the story where the AI like told him to divorce his wife after he talked to it for a long time, like it was saying, it was basically encouraging him to do things that would have been very destructive. Um, and so he tested it again. He's putting it to its limits. And he, he references that in one test conducted by an AI safety research group that hooked GPT-4 up to a number of other systems. GPT-4 was able to hire a human TaskRabbit worker to do a simple online task for it, solving a CAPTCHA test without alerting the person to the fact that it was a robot. The AI even lied to the worker about why it needed the CAPTCHA done concocting a story about a vision impairment. So CAPTCHA tests. Those are the tests that are designed to detect whether or not you are a robot.
David Leary: [00:36:17] So the trick to human into doing the caption for it? Yes.
Blake Oliver: [00:36:21] For it. Yes. So I mean, if you are a doomsday warrior, this is not encouraging because this means that the AI can use humans in the real world to accomplish things. And so imagine if you're a bad actor and you have access to this tool, what you can do with it, like Twitter bots have been a problem, but it's kind of always been possible to tell who's a bot and who's not. If you like, look carefully. But with AI, you're not going to be able to know who is a real actor and who is an AI actor in the online digital world.
David Leary: [00:36:55] But the AI bots should I should be able to to connect chatgpt to all my LinkedIn followers. And it should be like, can you tell me who here is a bot and then unfollow them? Right? Oh man. If link Microsoft adds that to LinkedIn. Actually that would be I would pay for that feature.
Blake Oliver: [00:37:13] That's genius. Or automatically decline all of the invites from people who are just going to sell me stuff. Right?
David Leary: [00:37:20] If you could detect that before based on their other behavior.
Blake Oliver: [00:37:23] So here's the other fun AI story that I saw. This is a tweet thread. Um, so this is from Jackson Greathouse Fall at Jackson Fall. He said, I gave GPT-4 a budget of $100 and told it to make as much money as possible. I'm acting as its human liaison, buying anything it says to. Do you think it'll be able to make smart investments and build an online business follow along? And he has been doing this for several days now. And the I have to say, like while the model is suggesting a lot of stuff that's really hot and popular online, like create an online dropshipping store and stuff like that, it's not it's not terrible at doing it. It it suggested basically to create an affiliate marketing site making content around eco friendly slash sustainable living products and the.com that it gave him was green gadget guru.com. And then Jackson used Midjourney to generate the logo for Green Gadget Guru, which looks pretty good honestly. He had he had to fix it because you know, Midjourney got the words wrong, but it got the design pretty good. And then, yeah, he, he built products like look at, look at these products that he mocked up using AI. He didn't design these the AI design, these ten eco friendly kitchen gadgets and they just they just look beautiful, really streamlined, really friendly, look good looking. I don't know how to describe it. These dark greens and kind of yellows. I mean, it looks like.
David Leary: [00:38:57] Like a real website. Like an Etsy page. Yeah.
Blake Oliver: [00:38:59] Yeah. And he just goes on and on. And so it's kind of like. I mean, where could this go? We're talking like sci fi kind of stuff. But imagine, like. If I gets better than the average person at business, we might have corporations that just delegate all of the important decision making to an artificial intelligence because it removes the human element from the equation.
David Leary: [00:39:24] Well, this is this solves the audit problem. Which essentially then nobody has to get a CPA.
Blake Oliver: [00:39:30] Well, and that's that's my question for audit. Yeah. Is could you just feed all of the transactions, all of the data into an AI and ask it to audit the books? I mean, what are the auditors really doing most of the time? Right. They're just ticking and tying. Why can't I do that? So then that calls into question, what are the two thirds of accountants that go from school into audit going to do instead? And how are the universities and colleges going to prepare them? This is. This is, uh. Yeah. I wonder if the if the audit firms are thinking about this, But the thing is, they're just so far behind. They're not thinking about it. Right? They're not. They're not have to be. They have to be. I don't know. We'll see. We'll see.
David Leary: [00:40:17] I mean, this is getting to the point where, you know, your parents my parents will be talking about this at Thanksgiving dinner because they're using it by this time on their devices, they'll be exposed to it.
Blake Oliver: [00:40:31] I mean, let's take this back to the whole Silicon Valley Bank collapse. One of the reasons or some of the blame, I suppose, is pointed at regulators who. They could have examined SVB and determined that it was too risky. But they didn't. And they were given that option by the deregulation that happened. But they still have the authority to. And, you know, I get it why they wouldn't do it. They got a lot going on If they don't have to check a bank like SVB because it's a regional bank and it's not one of the big banks, maybe they would decide not to do that. But with I. What if they could plug an AI into all of the bank systems, Right? All these banks have balance sheets that are updated and all their investments are tracked digitally. They have a daily view of their assets and their liabilities inside the bank. Why couldn't you just plug an AI into all the banks and tell the AI? Your job is to create a risk assessment and do a daily stress test of every single bank in the country? It's not that complicated. Right. You just have to figure out the fair market value of all these assets and liabilities, and most of them are readily tradable. So figure out like, is this bank trending toward insolvency or not? And you could do that on a daily basis and present that information to the Fed and then they could go in and shut it down before it becomes a problem or warn them before it becomes a problem. Right. I mean, audit, you know, like the way it's done now, it's like a snapshot in time every three months. It's just too slow. Matthew said Audit firms move in slow motion. I like to think of that. What was that movie with the Animals where there's this sloth that works at the post office or is it the DMV? You know what I'm talking about. I don't know. That's what I picture. And then Matthew says, Except for with them. My firm, of course.
David Leary: [00:42:35] Yeah. I don't have any other stories on, on our thoughts kind of on it.
Blake Oliver: [00:42:39] This that was the big news. Yeah. All right. Well, let's let's wrap it up. David, if people want to follow you online, where should they do that?
David Leary: [00:42:47] I'm on the socials, @DavidLeary. Please say you're not a bot. If you hit me up on LinkedIn, I'd be very helpful.
Blake Oliver: [00:42:54] Yeah, you can. You can craft your invite to David that says I'm not a bot using chat GPT. I'm sure he'll appreciate that. Christopher says the movie was Zootopia. Yes, Zootopia. Thank you. Uh, David Hall says I is a risk for false positives. Ai is only good. As good as the data set, we can corrupt the AI with bad data sets. Absolutely. So that's why what.
David Leary: [00:43:17] Happened in What was the big German one? The Wirecard Wirecard? Yeah. Remember they were creating fake bank statements and giving it to the auditors. The auditors are like, Sounds good to me.
Blake Oliver: [00:43:29] Well, it's like an AI could do bank confirmations, right? It's just get a list of all the bank accounts and then go ask all the banks like. Do these accounts exist and are these amounts accurate? You know, the stuff the auditors were too lazy to do or didn't have time to do. I could do that. All right, David. All right. I am @BlakeTOliver. Follow me on Twitter and LinkedIn and follow The Cloud Accounting Podcast. Subscribe to us on YouTube and we hope to see you again for another live stream episode. Thanks, everyone, for joining in. Bye bye.