329 Is The New OpenAI Model Any Good at Accounting?

Gang, it's a big day. Anytime you get a new AI model from the folks at OpenAI, it causes everyone to kind of stop and reassess and be like, okay, what's possible with AI today that wasn't yesterday? And it's really unlike any other technology we've ever had where overnight, like, I mean, entire markets can potentially shift with a new AI model coming out that's capable of really cool, powerful things. And these models are just going to keep getting more impactful. There's been some really, really, really interesting things to come out from this latest release with the new model solving like PhD level, like math and physics sort of stuff. Folks saying for certain applications, the new model is as useful as basically a graduate level student. So some, some super compelling examples of specific things it does amazingly well. But does that translate to accounting? Ultimately, buddy, that's how I put food on my table. So that's all I really care about. So today we're kicking the tires on the new model to see can it re. Can it review a set of financial statements? What if I chuck in a set of books? What does it have to say? Because I think sooner than later, we're actually going to get to a crossover point where you can throw kind of abstract things at a language model and it'll do that stuff as well as, I don't know, an intern, staff level, senior level. Let's how about this? That's the goal of today. How good of a financial statement reviewer is the new OpenAI model? By the end of the episode, we're going to figure it out. Come on in. Let's do it. So some housekeeping around the new model. The new model is called 01. It's like a little letter O and then the number one. What an awful naming convention. If you are a ChatGPT user, you will see two models you can use from the dropdown. Two new models. As of right now, you'll see O1 Preview and O1 Mini. O1 Mini is like the lighter weight, faster. Not quite as smart, but still pretty smart. Version 01 preview is the big boy, but it's still just a preview and they're set to release the non preview version about a month from now. And it will be better. So even the one that we have right now is not as good as the non preview version we will ultimately get. And the one really interesting thing to know about these new models is behind the scenes. Let's channel a secret behind the scenes. It's actually not a new model. It is the old model being used a different way. And so if you've played around with AI assistants very much or waited into prompt engineering, there's all sorts of different ways to massage better results out of what the AI assistant will give you. And a really straightforward example of this was maybe a year ago when somebody ran a study of the CHAT GPT taking the CPA exam. And it was when GPT4 came out, which was a big step improvement last spring, a really good new model. And when they just like copy pasted the multiple choice questions from the CPA exam and had it take the test, it scored, I think it was something like a 65 or 70 or something that wasn't passing. But when they gave it a few examples like here's a question, here's an answer, here's a question, here's an answer, here's a question, here's an answer, here's another question, now give me your answer. When they gave it a few examples like that and the questions had nothing to do with the question that it was answering, the score improved to like an 84 on average or something like that across all the sections. And then it went on to pass pretty much every accounting, other accounting certification, the EA, the CIA, GPT4, like cruised through all of those, but it performed much better when you prompted it a specific way. So there's always been this sorcery to how to get the most out of a model. And actually most of the products we use that are leveraging AI assistance on the back end, they have it generate something. The reality is behind the scenes, what's actually happening there most of the time is actually a sequence of prompts. Because by doing kind of this internal dialogue with a language model where it generates something and then you ask another model with a different perspective to review it and be like, does this seem right? And if not, make these changes like that sort of having that conversation with itself yields a much higher quality output. And that is the whole secret sauce of this new model that OpenAI put out is it is a model that basically reasons with itself. And so they say it's the first model that is capable of like a deeper level of reasoning. And behind the scenes, most people think there actually isn't any new model there. It is a new approach to using existing models that makes them much more smart. And in early testing, it greatly, greatly reduces hallucinations and performs much better at larger step by step like tasks that you would have to break down into sort of multiple sequences. And one interesting sort of sidebar on this is that there's been a, there's a lot of talk about like, have, have AI models kind of reached a plateau? Like will they really be able to continue improving at the rate they are right now? Because man, the improvement we've seen in the last 18 months is unbelievable. And if you project that out a couple few years, it's wild how good these could be. And so the big question is, yeah, are we going to kind of plateau though? Like, are you really going to just be able to keep improving? This is a form of improvement that can actually be applied to any model that has nothing to do with like the number of parameters trained into the model or how big of a model it is. It's entire. It's an entirely new way they have found to improve the quality of the output. And so this is a big step forward just using like the models we already have. But it will be an even bigger step forward when we have GPT 4, 5, GPT 5, Orion, whatever they end up calling it. And so it's interesting that not only the underlying models are being developed and getting better, but also the ways that we use them. We're still learning how to like best make use of that stuff and get the most out of the models after they've already been trained. Okay, enough geeking on that stuff. Let's dive into this little test that I want to run it through because I think this is really interesting. Historically what I have said is the right way to use AI assistants is to give it explicit step by step instructions and give it a job and say this is what I want you to do for me and then cut it loose and have it do that job. That's great, that's helpful. It's, it's like having an intern. Don't just cut them loose with no instructions on how to do the thing and then expect them to come back with what you actually wanted. And that is still the best way to use AI to explicitly instruct it on what you want. But the bigger test, long term. And what I think is more interesting and a more fun conversation is what does it just know? Like I meet a 10 year industry professional and I give them a project, they're just going to know how to do it right. This is an example of where we talk about, you know, not using systems, creating an over reliance on expertise. You can create a system and the more systematized your process is, the less expertise is required to go into it. And that's the case with AI too. The more you break it down and the more turnkey you make it, the more likely I will be able to do it correctly for you. But the other end of the spectrum is being fully reliant on expertise and just being able to check it in and see what it does. And that's what we're doing today. So I'm going to throw a balance sheet at it, since that's kind of the cornerstone of, like, how do you review the. Not the. We're not looking at the financial health of the business. We're looking at the technical accuracy of the accounting. And that's actually another wrinkle in this test is can we get that AI model to not be a CFO and give me insights about the health of the company? I'm an accountant. I don't care about that. I shouldn't say that, but you know what I mean. We're just looking at the technical correctness here to ensure that we put out a technically correct set of books. And the company that we're looking at, I mean, you ought to know who it is, is Craig. Craig's Design and Landscaping Services. The sample company in qbo. We had to create, like, oh, man, you know what would be awesome to do? We could really piss into it off with this. I don't know that I want to do that. We could create a whole Persona around Craig and his business, and Craig could be, like, kind of slimy, a little bit skeevy. Maybe Craig needs to be the new Steve. I don't know if that's a bear I want to poke. I just think it'd be really funny. All right, so if you log into your QBO file, this is actually super useful. And they. They, like, hide it in a really weird place. But if you're logged into qbo, in the top, right in the dropdown from qboa, like the accountants view, there is a sample company file. I don't know why they don't just put it in a list of, like, client files. That makes it way easier to find. But there's a sample company file there. Craig's Design and Landscaping Services. This is actually a godsend for testing AI stuff, too. If you're considering, like, security wise, what do I want to put in this? What do I not want to put into this? First you got to decide, like, what's even possible? What will it even do for me? And if it does something compelling, then we decide what. From a security standpoint, how do we want to do this? That's what makes a dummy file like this super, super helpful. So what I'VE done is I've run the August 31st Craig's Design and Landscaping Services comparative balance sheet. It's August 31st and July 31st. And let me first skim through this with my seasoned accounting. I mean, I have, I have most of us listening to this. We've reviewed our share of balance sheets, right. And the demo file in qbo, it's kind of janky. And I think it's probably because they have to do this thing like where they refresh it and roll it forward every single month. It's kind of weird, but initial things that I see when I look at this, things that might be issues, things that might not be issues. And this is, this is what I think makes it a good test of the AI, is we're not just asking it, is there something wrong? Because we don't have all the information here to determine if there is something wrong. We're asking is this something that we ought to look into further or not? And that is a judgment call. That is a very valuable judgment call. Right? Like how you set that threshold of what's worth looking into and what isn't. This episode is sponsored in part by Live Flow. True Story was just shooting a Sunday YouTube video and whenever I write these, I'm like, what's a good, like ad partner that would fit in real nice and organically to this video? The video, I think, is going to be a banger. It's about how to better use your scheduling app to like automate and standardize everything that happens around meetings. And I thought, what is a great tool to like standardize the prep work for meetings? Oh my gosh, Live Flow. So many situations. Think about like a client discovery call. Being able to use Live Flow to pull a huge spreadsheet, Excel or Google Sheets workbook worth of data out of QBO or Xero. A whole bunch of tabs of information with the click of a button. Then literally select a different company that you're connected with, click of the button, pull all that data for that company too. Say you're headed into a client discovery meeting. The client books that on your calendar. You also book 30 minutes of prep time. You pull all that data from their accounting file into a spreadsheet. You've got an easy way to review it all, go through it with them in the meeting. Another cool example, LifeLow's got like over a hundred templates of like different dashboards and advisory stuff. One of them is like a four week cash flow dashboard, like this big elaborate thing. You click the button, it Pulls all the data from the accounting file. Let's say, as one of my accounting packages, I do like weekly or monthly advisory calls with them. I use my booking tool for the client to schedule that live advisory meeting. I then book a prep meeting on my calendar, say 15 minutes or something like that. To get ready for that meeting, I hop into my spreadsheet app. I hop into my spreadsheet app of choice. With the click of a button, I pulled that cash flow data into a spreadsheet. And so I got like everything. And just like that, 90% of my prep work buddy, it's. It's there. And I can customize that template to be whatever I want, use it across a bunch of different clients. Imagine making those types of meetings. So turnkey. Or you just use Live Flow to pull all the data that you need for that meeting in with the click of a button. Man. So cool. Really fun. Use case of Liflow. I often forget Lifelow also does consolidations. So where this is a real headache is when you got to pull stuff from a bunch of different company files. You can do that with a click of a button as well. Even if they have different chart of accounts set up the mapping. Once you could be heading into a big old board meeting with like a bunch of different companies involved. You can pull all the stuff that you need with a click of a button. Super cool. That's Live Flow. To learn more, check out the link down in the show notes. This episode is sponsored in part by Team up, who will help you find talented folks in the Philippines to work for your accounting firm. They have recruiters on the ground in the Philippines. We'll find you awesome people. I'm looking at a bunch of them right here. Rona, that's her name. She's got a bachelor's in information systems. She went to work for a US firm as a virtual assistant. They're paying her 1070amonth. She had four years experience. She's handling people's inboxes, calendars and scheduling their social media. A firm hired her through Team Up. How about Aaron? He's got a BS in accounting as a cpa. Six years income tax US Income tax prep experience. So he's doing everything. Business, personal hustle and ultra tax. US firm hired him. They're paying him 1965 per month. Farah. She's got an MBA and a CPA. Was last running an entire team of bookkeepers. Six years experience. US firm hired her. They're paying her just over three grand a month to run a team. A team Gang, whether these folks are in the us, whether they are abroad, they are capable of running any part of your accounting firm. Particularly as a lot of firms are moving towards like kind of a relationship manager model where you have an onshore person that is managing the client relationship. That's really like the one blocker for some folks is they won't put offshore folks in direct contact with contact with clients though even that is kind of evolving. But our worldview around offshore work is oftentimes misshaped by the fact that menial tasks in other contexts like support will be delegated to, you know, India or something like that. The reality is these are super talented people that are capable of learning absolutely anything your onshore team is learning. And you've got big four firms, you got regional firms, you got all sorts of firms employing tens of thousands of people in the Philippines that are absolute experts at their craft. I've shared before, my last four hires were in the Philippines and my firm completely changed the way we approach getting the work done. And that is again how I would build a new firm tomorrow. To learn more about Team up, check out the link down in the show notes. So I'm going to skim through this. In fact, I'm going to write these down and I will note down. Here are the things that I would potentially look into. So we've got a new savings account. It's got an $800 balance now. Didn't have a balance last month. I'm going to confirm that with a bank statement to ensure that the bookkeeper didn't accidentally just post something to like an old bank account. Right. Didn't have this savings account last month, they have it this month. Let's just double check that real quick. Undeposited funds have gone up from 687 to 2062. I don't really like to see a whole ton of that, but given that it changed increased by like 3.5x, let's at least double check that and make sure there isn't any old stuff hanging out there. We've got a new truck, 13,004 95 that we have this month and was not there last month. If I'm a senior, I'm probably just going to look at the docs and make sure that the purchase was booked correctly. They got a credit card balance, came in where it was last month. That's fine. They have a board of equalization payable under other current liabilities. I don't know what that is as a $209 balance, but had a negative $21 balance last month. Definitely looking into that. I don't know why it would have been a negative last month. And then there's under current liabilities, there's a new loan payable for $4,000 that wasn't there last month. I'm going to make sure that's legit and got booked correctly. But also that's just a crappy name for a loan account. Oh, we've also got a new long term liability notes payable for 25 grand that wasn't there the month before. That's just a bad name. But also double check the docs, what else? Oh baby, we got opening balance equity. This is the demo file into it. This ought to be a shining beacon for what a good file ought to be. Oh, even better. The opening balance equity is negative 9905. What does that mean? That they rolled a bunch of like loans in when they set up the company. So it's negative 9905 this month. But last month it was positive 5000. Definitely worth looking into that. Net income. Negative 1800. It was negative 127 last month. Whatever. We're limiting the scope of this to just looking at the balance sheet. Okay. So I jotted down seven items and so we are going in cold turkey just expecting this thing to be an expert without instructing it on how to do financial statement reviews at all. How good can it do this? Which is a really important question because I don't know that I would have said this maybe 12 months ago. And there are a ton of accounting things that we are not this close to a kind of different approach to, but reviewing a set of financial statements. I'm talking basic SMB stuff. That's something that I think in the next couple of years LLMs will be able to do pretty darn well. Like there's always blind spots. Right. And, and so part of the, part of quote doing it well is do you see something that ought to merit a follow up question? But I honestly think we're pretty close to it doing that stuff well, which is a quick thing to say and, and people will disagree with that or not or agree with it. But just, I mean to open your mind up and just consider what if in the next 12 months we had a language model where you could just pop that stuff in and it would do your review in 30 seconds to the level of a, of a staff person or an intern or a senior level accountant. Like just hypothetical. If that is the case, that's a pretty profound change to our workflows. Right? Like what does that mean? Does that mean, does that mean that I, I have a higher level person do the prep and then just rely on the language model to do the final check out the door? Does that mean I can have a lower level person do the prep because they have a really high level assistant to help and then maybe it still goes through a high level human for final review or not? Like this is significant because it potentially really changes the workflows. Right? So here's what I'm popping into the new model. O1 preview. You could try this yourself in ChatGPT. Just at the very top, you do the model selection, say O1 preview. Here's the exact prompt. Here is a draft balance sheet. You are the accountant doing the technical review on the financials to ensure their correctness. Identify any items that may be incorrect, then provide a list of items that should be looked into further. I'm not interested in insights about the company or its financial health. This is purely a technical review to ensure that the accounting has been completed correctly. Only list items if there's probably an error and then a copy pasta just the flat text of the balance sheet. So the new models right now they are not quote multimodal, you can't upload files to them or anything like that. They're text only. But language models are amazingly good at like inferring structure just from a big pile of text. So this thing that I copy pasted in, it's literally just me dragging and selecting the text of a two column comparative balance sheet and pasting it in and expecting it to figure everything out from there. And historically like it does, like language models can totally figure that out. So it thought about this for 23 seconds. And that's one thing with 01 is it's, it takes longer to get an output. It thinks about it and there's situations where that's unacceptable and that's not fast enough. But 99% of situations like, especially for knowledge workers like us, sure. I can wait 23 seconds for you to review a balance sheet. Like that's not a big deal. Right? Okay. So it looks like it came up with 10 potential issues. Interesting. I had seven. Let's run through this. It says after conducting a technical review that provide a balance sheet for Craig's Design Landscaping services as of August 31st. I've identified several items that may be incorrect or require further investigation. Number one, opening balance equity changes issue. Opening balance equity account shows a significant change from 5000 to negative 9905. Explanation the opening balance equity is typically used only during the initial setup of a company's books to offset opening balances. It should not change thereafter. Action Investigate why there are transactions affecting this account in August. Adjustments should be reclassified to appropriate equity accounts such as retained earnings or owner's capital. That's a pretty good answer. And even if you're a junior and you may not know why that's wrong, that gives you a taster of what you need to dig into further to remedy it. So that is one of the ones that I had written down. So it's ticked off one of the seven things that I identified. Number two Absence of retained earnings issue. The retained earnings account is listed but shows no balance in both periods. Interesting explanation. Retained earnings should reflect the cumulative net income or loss for both periods. The absence of a balance suggests the prior period's profit or losses have not been closed to the account. Action verified that the net income from previous periods has been closed to retain earnings if not perform the necessary closing entries. I don't think it's saying that it should have gotten closed out month over month because that wouldn't be correct. There is a for some reason, maybe the QuickBooks balance sheet always does this. It produced a retained earnings line and it's just empty. Like it doesn't even have zeros. So it's calling that out. I don't know that I have a problem with that because that for me I would know is this a new company or not? And if I know it wouldn't have a retained earning balance, I'm probably just going to dismiss that. So I wouldn't say it's wrong. I maybe wouldn't have called it out, but I wouldn't necessarily like give that a knock against it. Okay. Number three Negative liability balance on Board of Equalization payable. I had this one too. The issue the Board of Equalization Payable account shows a negative balance of 2160 in July. Explanation A negative liability may indicate an overpayment or misclassification application. Action Review this account to determine if an overpayment was made or if transactions were incorrectly recorded. Correct. Any errors found. Perfect. Great. That's one that I had too. Good. Explanation number four. High undeposited funds balance. Yeah baby. Undeposited funds increase from 687 to 2062. Explanation undeposited funds represent receipts not yet deposited. A higher increasing balance may indicate delays in depositing funds or recording errors. Action Ensure that all received payments have been deposited and recorded accurately. Investigate any discrepancies and clear out old items and undeposited funds count. That's great. Okay, so it's gotten three of the seven that I've noted. Number five, classification of loan payable. This is the one that was under current liabilities issue. A loan payable of 4000 is listed under other current liabilities. Explanation. Loan should be reclassified based on their repayment terms. Amounts due within one year are current liabilities. Amounts due after one year are long term liabilities. Action. Review the loan agreement to determine if the correct classification was made. If part of the loan is long term, reclassify the appropriate portion to long term liabilities. I had called that out as like review it as a new loan. That's arguably a better version than what I called out because I didn't even say like let's carve this up between long term and short term. That's totally fine by me. Number six, it's got note payable. Increase note payables under long term liabilities increased by 25,000 in August. Explanation. A new note payable suggests new debt incurred. Ensure it was recorded correctly. Action. Verify the details of the new note payable including repayment terms and interest rates. Confirmed liabilities properly classified as long term. That's great. Honestly these are all even like more detailed than the explanation that I got. So right now it's nailed five of the seven things that I called out. I don't, I don't make too big of a deal out of this, but imagine just being able to chuck a set of financial statements into a language model. Now like I'm not to the point of replacing a human here necessarily, but as a second set of eyes even like I didn't think about the loan payable and current liabilities potentially being a long term liability. Was that a useful call out? Like maybe number seven recording a fixed asset truck issue. A truck valued at 35 13, 495 is added a fixed assets explanation. The purchase of a fixed asset should be recorded at cost and any associated appreciation should be accounted for. Action. Ensure the truck purchase is supported by paper documentation if financed. Oh. Confirm that related liabilities are recorded. Set up appropriate depreciation schedules. It's gotten all but the new savings account for me. It's got six of my seven now. Issue number eight. Total equity is negative at 11775 in August. A negative equity balance may indicate accumulated losses or accounting errors. Investigate the components of equity to ensure all transactions are recorded correctly and eh, I don't know that I would have called. It's like it's calling out a total amount. That's definitely a system of a smaller issue. But here I would rather just see what the smaller issue is, which it called out. There's a negative $10,000 opening balance equity issue. And it already called that out. So I wouldn't have called that one out. It's not incorrect. Number nine. Boom. Introduction of savings account without prior balance issue. A Savings account with 800 bucks appears in August but not in July. Explanation. The new account should have an opening entry if it was open during the period. Action. Confirm the opening of the savings account and ensure the initial deposit and any transfers are properly recorded. Eight for eight. It's got one more here. Number 10. Missing amounts and retained earnings and other accounts. Some accounts, like retained earnings show no amounts which may indicate incomplete information. And so that's. That's saying more. In the formatting of the report, there are rows. There are accounts that have no balances. That's actually a good note because I usually don't like including accounts with no amounts on them. I didn't even have that in mind, gang. That was the 10 things that I it identified. I identified seven. It picked up the exact same seven that I identified. It picked up. That last one was a good note that I didn't make. I wouldn't say I missed it, but like it's a good recommendation. They gave us one note about total equity being negative. Eh, I don't care about that. I just care about the underlying issue. And then they gave a note about retained earnings not showing a balance, which is like eh, okay. Honestly, I'm probably not worried about that because accounting systems sort of do that automatically, right? And I trust QuickBooks to do that correctly. That. That's a killer review. Today's episode is sponsored in part by bookkeep, the tool that's gonna automate a whole bunch of your e commerce accounting workflows. You know, apps like to get all up in your face and be like, come use me. Bookkeep. What people love about bookkeep. This is an actual user quote. Bookkeep works so well I forget I am using it. Don't you want the thing that just works his thing with bookkeeping? Bunch of their team. Actually most of their team are accountants and bookkeepers. They're not software nerds that were looking to cash in on some sort of plugging a thing into another thing. Arbitrage. What's the difference with bookkeep? Well, they use sale data while other apps base their entries on payout data. This means that they're like inherently cash basis, right? Bookkeep uses sale data. Means you have way more granular control over what you can do with it. Seeing very detailed information from each sale and giving you greater power for handling very complex scenarios where, say you're selling across multiple e comm channels, taking various payment methods. You can map all that stuff so you're not just stuck with the payout data, right? To handle tricky stuff like booking Shopify revenue based on fulfillment date as opposed to the default, which is the order date. Somebody listening to this just got really excited, man. No other app does that, okay? Your payouts will always reconcile to your bank feeds to the penny. And you're never going to miss a payout because Bookkeep will literally tell you every single day how much money each selling platform owes you. That is the power of using the sale data, not just the payout data, right? If you're hustling e comm stuff, please just don't stop DIY and hacking together. You just, just don't, okay? Get in touch with the folks at Bookkeep. They're accountants and bookkeepers just like us. You will feel heard and understood. Check out the link down in the show notes. This episode is sponsored in part by Core v. True Story. I don't think I fully got my head around tax advisory work until maybe seven or eight years ago when I saw my first demo of corvi. What Corvety does really, really well is it visualizes the savings you provide by delivering tax advisory work. And this is something that our tax software just horrendous at. It doesn't do it at all. All it really does is report back on like outcomes. And you could use maybe say BNA and have like kind of comparison columns of before and after. But what Corvi does is they have over 1200 like quote unquote tax strategies, which are many things that we know and we take for granted, but we don't always present to our clients to be like, hey, look at this smart thing that I did for you. And it explicitly outlines the various strategies you're employing for the client in terms of like real dollar savings, dollar savings over time, year over year. Like, you know that selection I had you do two years ago, like, here's the rolling savings that we're getting you. And this is why some firms are able to charge like north of 10k for tax planning work, even very small, like 3 person firms, while other firms are doing quote unquote tax planning at the end of the year. And like charging 300 bucks. It is all about the framing of the value to the client. There is kind of a small category of tools growing around just tax advisory. Corvi is one of them. Admittedly they're not cheap tools, but the reason they exist and the reason they're finding success is because the firms that are using them are charging top, like top, top dollar for tax advisory work. If you tune into this pod, you know that my belief is if you run a tax firm, tax advisory work, getting people up that value ladder is arguably the best thing that you can do for the profitability of your firm. Now tech can be a massive help here. I honestly don't know how I would do that work without Corby or something very similar to it. Learn more about Corvey. Check out the link down in the show. Notes at a high level. Now we're just talking about a first glance. Balance sheet review at a high level. This is a hot take. I don't know. This is a. Let's keep in mind this is a SMB simple set of books. I would say that level of review is above junior accountant level. Would you agree? What are some, some of the things that it called out? Hey, there's a new new balance in a savings account. Make sure somebody didn't just mess up the entry. Undeposited funds went up by like 3x. We should probably double check that new truck. Check the docs. Two new loans. Check the docs. A current liability had a negative balance last month. It's positive this month. Why would it have been negative last month and positive this month and then opening balance equity. That's maybe an easy one. There are definitely things in there where you, you get a set of books from a junior accountant and they're going to have little stuff like that. Why was this negative? Something balanced last month, but it's positive this month and you end up going and having to look into that. If you have reviewed books like that in the past, I would argue this reviewer is a higher level reviewer perhaps than the person that prepared those books. I don't know that that's a reach to say that. Cool. Very cool. So what, what questions does that beg answering to this? We're gonna need some follow up episodes on this one. First question for me is how far could we take this? It's one thing to call out. It's one thing to call out the issue. It's another to come up with a resolution. And so maybe the next test is you give it the ledger detail and you see just how just how far this rabbit hole goes gonna go and then like track down what the issue is. If it has the general ledger detail, would it propose like an adjustment in the software that would fix the underlying issue? Right. Could you get it, give it the purchase docs for the truck or the loan docs and have it ensure that it actually did that stuff correctly. But then the, the bigger question is probably one of who does that impact and how does it impact your workflow? Do you actually remove anybody? Do you change the capabilities of the folks involved with getting the engagements done? At the very, very least it feels like an automated second set of eyeballs for your preparers. And I to say that if a preparer sends you a books with these issues, then this tool is higher level than that. Prepare to say that is is unfair because we all know when your head's down preparing something, it's very hard to bring your head up and then review the same thing that you just made. Right? Like that's kind of a hard thing about accounting 101 and just kind of the fact that the brain can't really context switch to then assess your own work. But I would argue at the very least this is a helpful second set of eyes for a preparer before it goes to review. Right? And some of the studies we've seen of AI models in the past, particularly around stuff like consulting and writing technical things. The one consistent thing we've seen throughout is that when they benchmark workers who are paired with AI, given some training with AI and they work closely with it and it kind of reviews their work and they bounce ideas back and forth. Workers that are paired with AI versus those that don't use AI, the groups that improved the quality of the output and, and work performance the most were lower level groups. It was the AI kind of bringing them up to a more common standard. And maybe this is actually what we're seeing now here is this could potentially bring lower level accountants up to a higher standard across the board, which is gold. I mean if I can get my, my interns, my junior folks, even my staff level folks to produce stuff at a higher level on the first attempt and take that work away from my most scarce resource, my technical folks, for, for my team to genuinely learn from the language model throughout doing this. And so that learning is like, you know, the source of that learning is totally scalable because it's not a senior person that has to sit down and teach that lesson to every single accountant that comes through your ranks. That's pretty cool, man. I Don't want to go like full tinfoil hat and make any statements that are too wild. Besides the fact that it's probably worth digging into this a little more because I am very impressed at the performance of the new OpenAI model here. Again, there's going to be better versions of this where they put it more on rails and they say, here's how you review this and review this specifically and go do a hundred prompts to check all of those things. And that's going to be a much more robust version of, of what we just did. But I mean, kind of the black box, the magic perfect solution is where you just chuck it in and it just knows everything that you know, right? Which is probably an unrealistic thing to project onto a language model. But the fact remains, for a very simple SMB balance sheet, it just did exactly all the same things that I did. Arguably a couple things maybe a little bit better. That's pretty cool. I don't really know how to feel about that, to be honest. Besides, I can start thinking about what do we do next? How do we, like, how do we put this to work and make this useful? Security things. Because if you're thinking, like, what if I roll this out tomorrow? Security is front of mind there, right? People are still on like totally different camps on this, and there's a lot of spookiness around the newness of AI and folks still getting to grips with, like, is that okay for me to use it for this or that or not? There's, there's a degree of this that you kind of got to make a decision for yourself. But do it from a place of being informed about how these different models work and retain data. Still, the easiest resource for that, in my opinion, is two things that I worked with writeworks to put together, not a sponsored thing. I went to them, I said, here's a framework I'm going to release. Do you want to work with me on it to make it better and to put a little more weight behind it? We'll link those two things below. One is a responsible use policy. If you're using AI in your accounting firm, it's worth having a responsible use policy to define the boundaries of what's. What are okay and not okay ways for your team to use language models. And then second, another resource they actually put together that is a roundup comparison of all of the different large language models. And like security retention, how that works right now, all the mainstream tools that we talk about here. Claude Chat GPT I've been talking More about Spark, which is rightworks is kind of chatgpt for accountants. If you're on a paid business plan, which is anywhere from 20 to 30amonth, none of your data is retained into the models. So the old issues we had of AI taking your prompts and training them into the model, that's no longer an issue if you're on these paid plans. So for me personally, honestly, I don't have a problem putting a set of client financials in there. I think what I would probably do. Chat GBT has a cool feature called temporary chat. You toggle it on and that conversation, as soon as you leave it, it disappears forever. All that, all that data is gone. I'm not worried about them using my data to train the models. The only potential risk you have here is you put something sensitive into the app and then somebody gets unauthorized access to your Chat GBT account. This is how most data breaches happen, is not actually the data being leaked on the back end, although certainly can happen. But at the end of the day, I don't have any control over whether that happens or not. Most data breaches do not happen that way. They happen from unauthorized access to your account, like they get your login. But ChatGPT, you can also enable two factor authentication, which eliminates 95% of unauthorized access cases. So for me, I personally, and again, you have to, you have to kind of decide this for yourself. Talk with your IT support. Don't take security advice from a YouTuber like at the end of the day, like you just can't. I'm not willing to bear that liability for you. For me, if I was running my firm right now, if I was on ChatGPT team, I have no concerns about that stuff getting trained into the model. We'd probably use temporary chats anytime we're working with client data, so that even if somebody did get into the account somehow, that data would still be there. And then I'd have two factor authentication enabled so that it's virtually impossible for an outsider to get access to my account. And through all of those things, it seems exceedingly unlikely to me that, that we could ever have a breach of client data. That would be my position today, but you got to choose your own eventually. They're super excited about this. Also remember, this is one preview. In a month's time, we should have a better model. In the fall we're supposed to get GPT 4.5 Q1. Next year we're supposed to get GPT 5, which I don't know the CEO of OpenAI Japan said the next one, which would presumably be 4.5, he said is is two orders of magnitude better than GPT 4, which is 100x. I don't know on what measurement spectrum that is, but if I told you next January you could check any set of financials into a language model and it would review it at a very, very high level, I would want to know that now, and I would be considering that in my workflows and the type of expertise that I hire. It's significant, right? Pretty cool. We'd love to know your thoughts down below. This went much, much better than I thought it would, and I'll start planning some follow up episodes to explore this a little further. So that's all I got today. Thanks for coming and hanging out, and I'll see you in the next exciting installment of what Can AI do for Me.