114 A Big ChatGPT Update for Accountants

ChatGPT has eyes now. Remember back when they announced GPT4 and they said it has image processing capabilities? How exciting. That was the stuff they showed in the demo. It's finally here, or at least it's. It's coming here over the next couple of weeks. I'm mega pumped for this. It's a huge deal. Come on in, let's talk about it. Let's talk about specifically how it impacts what we do and how it's about to make AI a bunch more useful. Oh, Okay. It's been a sweaty week for AI announcements as I'm recording this. The Windows update was supposed to go live today. That gave people Windows Copilot over the next couple of weeks. We're now getting a massive ChatGPT update. A week or two ago, we got the Google Update that now enables Bard, their chatbot, to see into all of your Google documents and your Gmail and stuff like that. So now you can hop into Bard if you use Gmail, even a free version of Gmail, and talk with Bart and ask it to summarize your unread emails, ask it to draft replies. Pretty wild that we've got all that stuff now. But let's Talk specifically about ChatGPT. OpenAI's big announcement that ChatGPT can now what do they say, see and hear or something like that. And it's because in addition to being able to upload an image to Chat GPT, you, you'll now be able to talk with it vocally and it will return. Like it will give you a generated voice in return. And if you've been paying attention over the last few months, GPT4. So the best, the most performant version of ChatGPT that you get on ChatGPT plus, it's gotten way faster. Like the immediacy of how quick it starts responding with something has gotten much, much faster, which is really important when it comes to voice, when it comes to talking with a thing, because not only does it now have to wait for it to return text, which can take some time, it also has run that through an additional step of generating the voice for that text. And if you remember back when we played with pie on this podcast, we can link to that one at the end of the episode. It's basically a voice AI assistant. That is really interesting. They've invested $1.3 billion into building this AI voice assistant. And while the model behind it isn't the most amazing model right now, isn't the most useful model, it is really jarring. The first time you have a somewhat intelligent conversation with that voice thing. Like, we're used to our voice assistants and Siri and Google Assistant and Alexa and all that stuff, we're used to giving those things commands. But, man, talking with pie, having an actual conversation where it will ask you questions in return and like, having that more natural interaction is really weird. It's like really bizarre. Like, it's a very, I don't know, like, it changed what I thought AI was actually going to be when I had that experience for myself. And now for ChatGPT to have something along those lines, I don't know yet if it's as good and as quick as pie, but for ChatGPT to have something along those lines, that's significant. We may find that it's problematic. People start using it for things that maybe they shouldn't, but either way, it's a big deal. And for all like the social issues and stuff like that that you're gonna have from AI, I'm gonna try to limit, I don't know, generally limit the scope of what we talk about to, like, how is this going to impact our work? So, main thing I want to talk about today is its ability to now see images or over the next couple of weeks, people on the paid plan will start getting this functionality. They plan to give it to the free folks eventually, but no, no expected date for that yet. Basically, you'll be able to upload an image to this thing on desktop or on either of the mobile apps and then talk with it about that image or that document. You can actually upload several things and have conversations with it. And so the example that they gave is with the mobile app, they take a picture of a bike and it asks it, how do I lower the seat for this bike? They actually even showed taking a picture of the bike and circling part of the image to emphasize that you're specifically talking about this part of the image they showed, taking a picture of the instructions for the bike in addition to the picture of the bike for it to give you a better answer. And man, this is gonna be, I feel like almost as big as original chatgpt when it comes to exploring what exactly the use cases for this thing, what exactly those use cases are, because it can now genuinely, you know, see. It can see. And it like, there's so much of what we go about in our day to day where we're making decisions visually based on, you know, maybe what your computer presents to you or this like actual physical thing that you're having struggles with and being able to bridge to the language model those physical things. That's really exciting. Now everybody, obviously this is real exciting news. OpenAI is said to now be selling equity at a 80 to $90 billion valuation. Remember a year ago when Microsoft invested $1 billion into OpenAI and that seemed unthinkable? Well, at that valuation, that is a 4x return for Microsoft in 12 months. Or does GPT4,5 and GPT5 are in the works? Oh yeah. And they're still rolling out their new image generation model also, which I don't have, but I've seen it do some unbelievable things like I can't remember what it was generate an image of it was like a couple different fruits or the second was an onion and give them each a speech bubble. And like the onion responds with a pun. And like in that onion's response was something about layers or something like that. So it's actually generating images and speech bubbles and text from a single prompt. So OpenAI's new image generation model, that's really interesting. That's going to like push the envelope of what you can just generate with a prompt. But getting back to image generation, obviously in real life, this has a ton of applications. The holy grail has always been take a picture of the ingredients in your fridge and tell it to give you a recipe for what to make for dinner. So now this could be take a pic what's in your fridge, of what's in your pantry, of what's in your freezer and give me a recipe. And we're talking about now like doing this with your phone as like a tee hee haha. Isn't this a cool thing? Like we're a couple cameras in your fridge and in your pantry. A way though of like having an app that can just like always be feeding you. Like what's the highest rated recipe for all of the ingredients that I have in my fridge or my pantry right now. Like, there's a lot of weird things about it, being able to infer stuff from what it sees, but something tells me that's going to impact like the technology we use day to day in a pretty compelling way. Now getting down to biz and how you use this stuff for business. So image to text models, and that's what this is, is you can give it an image and it will explain with text exactly what's happening in that image and be able to answer questions about it and that sort of thing. Image to text models have been a thing for a long time. I've seen people say that what ChatGPT. What they're building into ChatGPT now is the same as Bing's image processing, is the same as Bard's image processing, because you can also upload images to those services as well and then chat with them about that image. I don't know about you, though, if you've ever tried that. I've had really mixed results with it. And what was notable about the GPT4 announcement that OpenAI did earlier this year, that was the announcement where they prepared a tax return and there was all this drama around that. If you haven't seen that, I did a whole video on it on my main YouTube channel. It's actually the video you'll find on that, like, home channel page for me, I'll just. We'll link it at the end of this video if you're watching on YouTube. But with that GPT4 announcement, they said it can also understand images. And we thought we were going to get that functionality right away, but here we are, like six months later, just getting it now. And in that demo, what they showed was a couple different things. He, like, hand scrawled on a notepad a website design and took a picture of it, and then it wrote the code and then it created that website. But the other thing they did was they gave it a screenshot of what was on his computer. And at the time it was Discord, like the chat app Discord. And it returned this unbelievably detailed explanation of all of the things that were on his screen. In fact, let me pull that up, because this blew my mind when it happened and made me realize, oh, if a language model can see what's happening on my desktop, it's actually going to be able to, like, automatically navigate my desktop too, because it can totally, like, understand what's happening there. So in the demo, they gave the language model a screenshot of Discord and it returns four detailed paragraphs. The image is a screenshot of a Discord application interface. In the top left corner, there is a server icon named GPT4. On the left side, under the server name, there are two sections, Text Channels and Voice Channels. Under Text channels, there are two channels, General and GPT4 demo, the latter being the currently selected channel. Under Voice channel, there is a single channel named General. It goes on for three more paragraphs, identifying every single UI element on the screen. And where that becomes really wild is when that thing is watching what you are doing working on your computer, and you have this helpful assistant that can comprehend everything that's on your screen. You can make it even smarter by plugging that into the help documentation for the application you're using. Maybe that's Ultra tax, maybe that's QuickBooks. The idea that it can now see with that level of granularity of detail everything that is on your screen. If you know, for tax example, I've got a 1099 R and that language model can see, you know, the help documentation for Ultra Tax or Drake or whatever my tax software is, you better believe you're going to be able to give that thing a 1099 R and it enter that into the tax software now, like raw dogging it with vanilla GPT. Will it get that right? I don't know, but it's the kind of thing where you put it on rails a little bit and you like develop a domain specific application for this. And this is now seeing and comprehending everything that's happening on our desktop in a way that will then enable it to actually like do work in a very meaningful way. And this is why I've been saying agents are the huge like groundbreaking thing for accountants. Specifically, agents are where AI goes from being like a sort of novelty and it's like it's absolutely helpful and time saver, but it's kind of this other little silo thing. It's where it goes from that to like, oh, I could probably just plug this in and it will like reconcile this whole month end for me. Like it will actually go through QuickBooks and do all of that. Like we're talking about agents. Agents being an AI that can like click things on your screen and type in letters and stuff like that, we're talking about agents. They can see your desktop, understand all the options that are there, and then logic through what the next step ought to be for completing the task that you asked it to complete. And around the really mainstream use cases like reconciling a bank statement at month end in QuickBooks, a thing that happens millions of times every single month. Like, you better believe we're going to be able to build an agent that will do that. This episode is sponsored in part by the fine folks at Cloud Accountant Staffing. Do you hire accountants? Bless your little heart. Not the best part of the job, in my opinion. Not something I ever enjoyed. Well, listen, you can build your accounting dream team with talented offshore accountants in the Philippines that work 100% full time for your firm. Their accountants aren't freelancing or contracting for multiple firms. They're all yours. They work exclusively for you and are incentivized to stay with you and your team, long term, they're not gonna get swiped. Cloud account staffing is 100% dedicated to the accounting industry and founded by a former accounting firm owner that understands your business, knows your pain points. They had to hire some accountants and they said, you know what, we're gonna build our own pipeline in the Philippines, gonna pull in some super talented people and then open that up to other firms. Basically that's the story I've been talking a lot about staffing, building more resilient staffing pipelines for your firms. I had staff in the Philippines like totally red pilled me to like, oh geez, like we need to globalize the way that we get our work done. Check these folks out. Link in the show description cloud accountantstaffing.com this episode oh, this episode is sponsored in part by countests. You know what? I don't enjoy spinning the old wheel of does this person I just hired know what the hey they're doing? Because I can tell you that more times than not, the answer was no. Because how in the world do you actually figure out if this person can be productive based on, I don't know, a resume based on the firmness of their handshake? You can't. But Account Tests is gonna give you a little more information than you had otherwise. So Count Tests is a super duper simple way of doing skills testing for the people that you're going to hire. Basically generate a link, you can run them through these skills tests. It's nothing super sweaty, but you know what it is is information. Information that you wouldn't otherwise have unless you sent them this little quiz. They got tests for accounting stuff, for Gap stuff for tax stuff. Whatever you're hiring for, they probably got a test for you. Super cost effective, because let me tell you, it is a whole heck of a lot cheaper than hiring the wrong person and then training them for months and then realizing I have made a huge mistake and we just wasted months of our time with the wrong person. Boy, do I have a whole closet full of those T shirts. So that's something that keeps you up at night. Check out account test. We'll put a link in the show notes. Now, really interestingly, who's gonna build that agent? Is it gonna be Intuit? It's hard to believe that like Intuit's going to build something that will automate atop their platform. Is it going to be a practice management system? We saw from Pixie earlier this year, they're testing an agent that can go out to Xero and Fetch accounting info for you and include that in like a draft in email. Like pull financial results, export reports and attach those to an email. Is it going to be a practice management system so that. Let's say Keeper. Keeper is they're a tool for managing your month end close. Let's say they build an agent. They already have a Chrome extension that could help with this. Let's say they build an agent that can go out and reconcile that bank statement so that when you have your list of month end close tasks inside of Keeper, that reconciliation task, that's assigned to an agent, it's not assigned to a human user, it's actually assigned to an agent and it will actually go and do that work inside of QuickBooks. So while ChatGPT is not the first app to stand up the ability to upload an image and then chat with that image, the like, the level of comprehension it's shown in the demo makes me extremely excited because I've tried this exact same thing with Bard and with Bing right now actually what I'll do is I'll upload it a screenshot of ultratax and say tell me everything that you see on this screen and like it gets maybe a third of the stuff and then honestly like makes up a third of the other stuff as opposed to what GPT4 is doing here. It's literally every single user interface element, it's describing it one by one so that then in a follow up prompt you can say using Discord, for example, I want you to go to the text, I want you to go to the general text channel and post a message hello world. And because it knows where all those UI elements are, it can go and do that and it can infer based on what it sees what it's supposed to do next. Now that may all sound super pie in the sky abstract to round that out and bring it home. Ultimately what this enables is language models take like a chat GPT experience to see and comprehend everything that's happening on your screen. So if you're chatting with a assistant, you can just reference what does this mean? Or how do I do XYZ in this program that you're using? And I can see all of that stuff. It has that context it'll be interested to see. Interesting to see who all ultimately uses this. Right now Microsoft is actually developing its own internal language models to be less costly than those of OpenAI. Like they're these like GPT4 is a massive model that's pretty expensive to use at scale. And so Microsoft is already developing their own internal like alternative models that are like more lean and not as resource intensive. And so I like, I don't think we can any longer generalize and assume that Bing, for example, is using all of the OpenAI models or the same versions that you get in ChatGPT. Honestly, it hasn't been my experience to date. It seems like when I use chatgpt I get better results than when I use Bing. But all that is to say I don't know that Bing is just going to bolt in GPT4's image processing capabilities into Bing or into Windows Copilot that now lives on that right hand panel of Windows. Right? One thing we've seen for sure is like when you have a really impressive development, everybody else follows along. So Google rolls out the functionality and bard to be able to chat and etc. Into your Gmail and into your documents and stuff like that. And not a week later window gives a Microsoft gives us a launch date for the same thing inside Microsoft 365 of of November 1st. So if GPT4 is way ahead of the curve on this image processing, stuff like that'll only serve to encourage everyone else to cat the big things on the horizon for the rest of this year. OpenAI is having an event in November where presumably I suspect we'll get something GPT 4.5 maybe the big biggest possibility right now, kind of the biggest unknown, is Google launching its new language model Gemini, which supposedly is much bigger than GPT4, which is kind of unthinkable because that model is massive. But the big really interesting thing about Gemini is it will be trained on YouTube data. And if you think about just how much information is in YouTube and how good that information is and the metadata they have to score how good that YouTube data is, because that's really important. You can't just grab everything on the Internet because there's a whole ton of stuff on the Internet that's just garbage, right? But with YouTube, you have all this retention data and upvotes and down votes and all this metadata that tells you how high quality is that information and you even have it visually represented alongside those words. So this new model from Google, Gemini that's supposed to come, I think was it this fall, I don't know, sometime soon is gonna be trained on all that YouTube data, which could be like unbelievably powerful. So we still got that stuff kind of in the pipeline this year. And for developers, frankly right now it makes it a little bit hard to know who to hit your wagon to and like, who to build around. So right now I'm in New York at the AICPA's executive roundtable. Thing is this week, which is basically pulling in folks from different software companies. And the aicpa, you know, just kind of vocalizes what they see us needing as a profession to try to get, you know, folks in alignment with the needs of accountants. And I'm here talking on AI things and obviously AI is a really big part of it. Basically, in everybody's presentations, they're going to be specifically calling out like, how are we using AI? Or how do we plan to use AI? AICPA actually also said in their next startup accelerator, which they do each year, they're especially looking for AI companies, potentially even exclusively. I couldn't tell if that was all they're going to allow in this year or not. But a hard thing for these companies is knowing who you're going to. Who you're going to go in with. Like, are you going to go in with open AI? Amazon just invested. This is all stuff in the last week. It's wild. Amazon just invested $4 billion in Anthropic, which is like the leading competitor to OpenAI. So Amazon's kind of got into bed with them. Like, Google's doing their own stuff. Obviously, Microsoft is with OpenAI, there's sort of this arms race and if you're a software developer, it's hard to know who to build for because each language model has its own sort of little nuances. And we're now seeing, you know, chat being built into everything. Now Canopy has an interesting, like insights sort of functionality in their practice management system where you can type like, chat in and it'll pull relevant reports and that sort of thing. In fact, in the pitch video that's going on my main YouTube channel this Sunday, in a couple of days, there's a company that gives you a full chat interface to chat with your accounting file data. So that stuff is now, like, being implemented in meaningful ways in the tools that we use. And then now we get image stuff, right? And so, like, everything's happening so fast that it's really hard to know what to prioritize. Needless to say, it is an exciting time. I will keep running on that little hamster wheel to surface the most helpful stuff we can. I've been super active on social the last few weeks, trying to put out the most valuable stuff possible around. Like, here's just what you need to know about Windows Copilot, and we're working on a roundup on zoom AI and like all these different things that are coming out where it's like, good golly, how can somebody that actually has a job like reasonably keep up with all of this stuff? I don't know, but I'm gonna try to distill that down to the most concise, helpful stuff possible. I jotted down six specific applications that I think are especially interesting for image processing in the work that we do, just to kind of get the wheels turning. Number one, using image processing or image to text models, which are now going to be built into GPT4 and ChatGPT GPT using them to comprehend worksheets and tables, which really often come up in regulatory stuff and websites and things that the government puts out. So in the US here, the irs, for example, has worksheets for all sorts of things and tables for all sorts of things. And if we're thinking about how we build knowledge into language models, that is taking source material and using the model to reference that source material and then provide an answer just according to the source, you got to think about what that source material looks like. Because if you've ever, for example, uploaded a PDF to a service and then chatted with that PDF and you get like a really dumb answer, oftentimes it's because the answer you're looking for, tax table, for example, is in a table. And if you just copy and paste a tax table, it actually doesn't really mean anything. It's kind of like a tax return where if you like put that into a Word document, it's total garbage, right? Because there's like these visual boxes and all of these things that sequentially just don't make sense. So the ability for the language model to understand visually tables and worksheets and that sort of thing, that is going to make it much smarter. Like really just. It's going to be a big quality of life improvement. Because in the past that was all like, information that it fundamentally didn't even have because it couldn't understand it. Hey, let me just record this podcast. Not I can't because my computer's broken. It isn't really broken, but if it was, and I was in an accounting firm, you know who I call Tech Guru. That's right. This episode is sponsored in part by Tech Guru IT for accounting firms. Because you got better stuff to do than fix your broken computer. Now, if you don't have a partner yet for managing this stuff for you, buddy, I don't know what to tell you. You're doing it wrong. And the nice Thing about finding a IT group that works specifically with accounting firms is they understand like the actual tools that we use. When you show them your tax software, they're not like, what is this absolute poc? I spent myself a lot of money training non accounting firm IT groups on how to understand things like CCH engagement. Whereas if you go to a group like Tech Guru that actually works with accountants, they already understand this stuff. They know to leave UTF alone during busy season. And so like the cadence of how they plan and strategize with you is actually catered to your firm. And like the cycles that your firm goes through, check out Tech Guru. Put a link in the show notes because you got better stuff to do. Okay, this episode is sponsored in part by Copilot. Okay, everybody gather round. Take. Take one another's hands. Now we can all agree the way that we exchange information with clients. Very important. Can be a massive time waster if you don't nail it. But I get it. The fear of training clients on a portal system, right? What if you end up not liking it? What if you train a whole pile of people on this giant system and then you're like, turns out this actually isn't the one for me. Or this new super cool AI rocket ship platform just took off and now I got to get all my clients to move. Here's the thing that is the best argument for why your client portal experience ought to be separate from your workflow management tool. You getting what I'm saying, Copilot? Man, all they want to touch is that client experience. Just the portal giving you a mega flexible platform for how you want to work with your clients. So that if the workflow stuff changes, if you want to pull in a different tool for that, you can without changing the client experience. Pretty smart. Especially in these scary changing times of AI. Right? Right. Actually got a demo day coming up on the main YouTube channel from Copilot in the next week or two where we actually get hands on with it. You can see even, even more about it. If you're looking for a cool modern client portal experience, check out Copilot link in the show notes. Number two, Understanding tax forms. Again, because tax forms are not narrative. They're like laid out in a visual way. That's been a really hard thing for AI to comprehend. But imagine being like a junior tax Preparer in the US here and you have a weird W2 where you're like, why does this box. The difference between these two boxes are not the amount that I would expect and you're like troubleshooting and trying to figure that out. Imagine being able to check that into a language model and just be like, why is this number not that? And I can go out and I can pull the source documentation for that and it can say like, oh, that's like a San Diego local tax thing or housing allowance or whatever is causing that thing to be out of whack. That's something that a junior person really wouldn't know. That could be a massive help. And interestingly, in the studies that are now coming out and like, benchmarking professionals doing all sorts of various tasks, and they randomly give some professionals access to AI and they don't give professional, other professionals access to AI On a whole, like, virtually every study I'm seeing now is like, the folks that are getting comfortable and learning how to use the AI are being more productive, but the biggest difference is being made with the people who are previously underperforming. So it will help. In most of these studies, just about everyone but the people that make the biggest jump were actually the least productive people before. And in the context of accounting stuff, that's usually going to be junior people. So there's a lot of this that, like the old trope, which I think we can kind of put to bed, that this will automate the menial tasks. Like, it's. It's more nuanced than that, like what AI is doing because it's actually improving and replacing tasks on the entirety of the expertise spectrum. But it's led a lot of people to be like, oh, what is going to be the place for the interns and the junior workers and stuff like that? Honestly, what I'm saying is this actually is going to make them way more productive. And if you think about when you're super green, all the things that you waste a tremendous amount of time on because you're stuck and you don't know whether to ask for help. Like, this could be your best friend for that sort of thing, right? Number three, doing a comprehensive review of a stack of disorganized documents, taking that from something that is completely unstructured to structured. Now, if you think about what ChatGPT can do with a bank statement, you copy and paste the text on a bank statement. Literally just paste it in with no structure at all. It can give you a structured CSV in whatever format you want. And it does that for flawlessly. It's super, super cool. Because if you or I look at what just got pasted in there, like, the structure of it makes absolutely no sense. But it can figure that stuff out really well. Now if you expand the scope of that to, if you're a tax preparer, a pile, you know, 30 pages of documentation that's a mix of W2s and client emails and just all this random junk. A big part of tax preparation in the US is just organizing that stuff and annotating it in a helpful way, getting that stuff into, like, a standardized work paper or directly into the tax software. We have the tech now, and the image processing takes this a big step further to chuck that stuff into a language model en masse, obviously responsibly. But those language models are good enough now to where those things could turn around like a structured work paper and like an Excel workbook or something like that and give you something in return. That is a massive head start compared to having to build that whole file from scratch. And that is so exciting to me because that is such a big part of the tech tax preparation process. Just like the grunt work of organizing all of that stuff. Now, this gets more complex when you start talking about different file storage systems. Oh, I use CCH Engagement and I want my work papers to look just like that. And, and so, like, the more fragmented that gets, the harder it is. But like, man, surely there's a place to start there with like 1040s. And this is why a few months ago, I was actually pushing for, like, can we get to like, a. A standardized 1040 work paper that at least handles, like. And I understand it's not gonna handle every single thing in the universe, but even something that handles 80% of the things that most returns have, that's a huge step forward because we're all doing this work in completely different ways right now. But if you have a standardized format of what that work paper would look like, developers can, like, work with language models to get them to output those files in exactly the format that you want. And this is one thing where I think the profession is not doing enough, you know, around standardization engineers, as irritating as they are, they're very good at building shared infrastructure open source software so that they're not constantly reinventing the wheel over and over again. This is something where I think the profession could lean into that a little more to make it easier for developers to give us stuff in the standardized format that we want and stop getting stuck on the fact that it won't do 100% of it. Because I'm going to be super gosh darn happy if it'll do 70% of it for me, or frankly, 40% of it for me, like this is such an exciting development to me, the tech stuff it was capable of doing is really impressive. But when you layer in the image comprehension and being able to chuck a document at it and it's going to know what it is, even be able to go out to like a browser and find additional information about that thing if it needs to, that just seems like it is taking us to a fundamentally better way of getting that work done. Okay, number four, how about crawling a file system to make connections and generally be helpful? So we've been talking a bit about agents and how agents are like self guided. You set them out with a task and they just go and they do that task and they report back to you when they have questions or when they're done. And they're far from perfect now, but they've also gotten a lot better in the last couple months and age image comprehension will make agents much better. So if you think about all the stuff you have in your file system right now, what would you do if you had like an infinite amount of time from, I don't know, a junior level person, what could you task them with that would leave your system better organized or generate some insights or something like that, you know? And is that something that you would even have to define yourself or is it something that we could standardize better so that if for example, we document a month on close for our clients in a more standardized way, we can have a shared tool that then goes in and like audits that stuff on a more automated basis. Right. Right now that's really hard because we're doing stuff in very siloed ways and everyone has their own approach to that. But as we've talked about, the language model can now see and comprehend like everything that we look at. And so it's got to raise the bar for how it can contribute to that process. Number five, we talked about this a bit. Just comprehending what's happening on your screen. I don't know about you. There's the friction to me of having to switch to the app of ChatGPT and then go explain the problem that usually keeps me from going over there and doing it. As opposed to kind of like in the example of uploading the bike picture and circling the thing. That's the issue, man. They've already built that into the mobile app. I would love something like that on my desktop. Almost like your little screen snipper, like screenshot tool. Let me just draw a circle around this thing on my computer that I'm Mad at and tell it to help me solve the problem or something. Something like that. How cool would that be? Especially when we get into domain specific applications around like I'm troubleshooting this thing in QuickBooks or my balance sheet doesn't die out in the tax software because it never does. The ability for it to see what you're working on, that's gonna be really exciting. And then number six, just the seamlessness of doing that stuff over voice. Imagine sitting there working on your comp, you got your headphones on and doing like whatever cue you use for your voice assistant on your phone or whatever, something like that. Doing that to your computer and having a conversation with about what's on your screen like as you're working. Combining its ability to see what's on the screen with voice, that's really interesting. Like it may actually make work a little more conversational as you are typing and mousing away. Like is there a conversation that's then happening to as maybe it's an agent that's trying to do something in the background and it's reporting back and asking questions or having a dialogue about something on your screen that you're troubleshooting. Maybe it goes out and it fetches something from the web that's helpful to troubleshoot that problem as you're working on it. Stuff like that, man, isn't that interesting. Doing a little bit of research before the pod here. Mashable, the online publication, joked that now with the Voice functionality of ChatGPT, parents can now fully outsource bedtime stories to ChatGPT. I don't know if you've ever done this. I've actually done this. I did this for the first time with each of my bigger kids in the last week. I had Chat GPT write a bedtime story for them and I would give it kind of a general theme. And so like my four year old boy, I was like, he just got a new bed. And so the, the prompt that I gave it was something like create a bedtime story for my 4 year old son about a workshop that makes magical beds. He's into Lego Ninjago, the cars, movies and fart jokes. And it made this like unbelievable bedtime story where like this character from Lego Ninjago was like the head of the workshop and they were making these magical beds that make, you know, you have really cool adventurous dreams. And then Lightning McQueen and Mater showed up at one point and it just like blew his mind. And then I did something similar with my daughter that was more like princess mermaid themed kind of thing. And it's really amazing at that, doing it with voices. Like it'll now read it to them. I don't know, I still like actually like reading it myself. But it'll be interesting. I think the we'll see some of the mainstream stuff like that or a more compelling way to like make food because it can understand the ingredients that you have it. Like, I think we'll get that stuff first before the domain specific stuff because it is so mainstream and like that's where the first investment always goes is what are the really compelling general use cases. But the reason that we're here, the reason we're having this conversations, frankly, the reason I'm in New York this week is I love driving the conversation around. Hey, here's this cool new thing. Plug it in to make accountants more efficient stat. I need it yesterday. So the more that we can engage with these conversations and keep putting this stuff in front of our software vendors, the more they are going to develop cool new stuff for us. Because there's just an abundance of stuff to do. If you don't have to organize those tax documents yourself anymore, if your juniors can be much more productive, everybody's going to be just fine. Because like where the misnomer is that there's always this fixed body of work to do. And if you've done this stuff for very long, you just know that that's absolutely not the case. How many times has somebody said, oh, now that you've automated this task, you can focus on higher level tasks. And then what everybody actually does is they just go pick up one more client, right? They don't actually go on and do the advisory or whatever that higher level thing is. Like there's just way more work to be done than any of us can reasonably do right now. So this stuff can't come soon enough. That's it for this week. I know this stuff is confusing, so please keep the questions coming. We'll do a Q and A next week. I try not to get too abstract with the AI stuff because I know that we all actually have jobs and need to get a little more productive each day. So try to focus on the more practical like rubber to the road stuff. But this is a big time announcement that I'm really excited about. So keep those questions coming, especially the practical stuff where it's like, hey, what are some ideas for how I could use it for this problem or that problem? Or maybe you just want my bedtime story prompts, you know, could be as simple as that. That's it for this week. I'll see you next week. Toodaloo.