It’s happening! Chad’s running for mayor of AutomationTown! Not only that, he’s going to automate running for office. This can only end well.
And do you ChatGPT? Of course you do! It’s AI’s killer app, and we’re doubling down on it in this week’s ‘Oops All ChatGPT’ episode of AutomationPod.
Welcome to AutomationTown! A podcast about regular people, building automations for the problems we all share.
First time here? Start with S02E01 "For When You Need A Chatbot"
What are the things you do each day that you'd rather forget?
Chad & Jason explore common pain points for knowledge workers, and track down the people & tools necessary to automate trivial to-do's.
Previously on AutomationTown.
I dunno. Please tell me. It can't be.
I ran out of anxiety medication. Oh, I'm sorry. You're out of it. Yeah, it's called potatozone. There's been a shortage for whatever reason the past few weeks, a pharmacist hasn't been able to get any. Chad, you know, who would make a great mayor? Well, not Jake. You, me? Yeah. It would be a great way to learn about the capital underground.
Ooh. If you were the mayor, you get briefed on all that stuff. You're not wrong. Have either of you considered running for mayor in regards to running for Mayor Chad? I know you've talked a bit about it. Yeah, I'm, I'm thinking. Wow.
Come in. Oh, what have I stepped into? Just the Davis mayoral campaign, world headquarters. That's right. This RV has been through a lot. Not only that, but we're automating the campaign. That doesn't seem advisable. Hey, remember lob, the direct mailing api, we looked at a while back? Yep. We've been sending out a ton of mailers with lob.
Actually, Chad's been showing me how to generate texts with G P T. Eh. Might have been a bad idea. Yeah, and we're using slides to generate the flyers with the. That is disgusting in the best way.
Well, it's happening. It's happening. Chad's running for mayor of automation town because it just can't be Jake, right? Hold the cranberry. Not in this town. Okay, so there's that. But this week also we're talking chat, G P T, the biggest sexiest juiciest, juiciest AI application out there today. It's AI's killer app this week on automation.
Hey gang idea for the pod today. Okay, what if we just answer a whole bunch of Chad G p t questions? I like it. We've got more than enough queued up to fill up an episode. So Chad, we use G P T for the flyers, but is that the same as Chad g p t? You wanna save that one for the pod? Yeah, let's kick off the pod recording with that.
It's really a common point of confusion. Fusion what? The Thanks.
Is it potato zone? Yay. All my little friends, how I've missed you. Oh my gosh, that's a lot of, lot of pills. Chad, you making me so happy? Okay. Is arts and crafts time over here? We going to get on with the thing? Yep. Paul, can we do more mayor stuff after Paul? Hey, any updates from the love boat? I don't know.
Uhoh, I'm not sure it's really going anywhere. I just feel like she's trying to distance herself. That's okay. Ball. These things can have and flow. It doesn't work out, and she wasn't the right one for you. All right, let's do this. Let's do it. Mics are hot. Hot mics. What does that even mean In three, two.
Welcome in. It's a special, oops, all Chatt, G P T edition of Automation Pod. We've got a mountain of mail about chat, G p T, so we're gonna run through a whole pile of that stuff. How are using it today and how it can make you a little bit more productive. But first boy, what an exciting day. If you haven't heard.
Our very own Chad Charles Davis. It's not my middle name. He's officially running for mayor of Automation Town. That's right. The runoff elections are coming up after Mayor Goodway resigned, but my main opponent looks to be Interim Mayor Jake mc Kringle Barry at Mayor of Manuel Burke. You're in automation.
Really, even just anything from this century. Vote Chat. That's right. Vote chat. That is enough of your propaganda. For now though, let's introduce the third member of our team. Hello Pat, why did the politician run for Mayor Pat with the jokes. Okay. Why did the politician run for mayor so he could run the city?
Okay. Uh, was that G p T, Chad, pat, give me three more possible punchlines to that joke. Prioritize ones that are actually funny. Act as a comedian. Sure. One. He was looking for a new running mate. Oof. He was running out of things to do, so decided to run for mayor. That's a ground he wanted to put his running shoes to good.
That one wasn't too bad. No, it wasn't too bad. And frankly, it's a great way to kick off a discussion about G P T chat. So Pat, here's running a version of what's called G P T, right? Yes. Now chat, G p t has taken the world by storm. Can you run me through what G P T is versus what chat G P T is? Sure. You're making me do this one.
All right, fine. Yep. So think about it first. This is one company. This is a company called Open AI, who created a product. It's kind of like Kleenex. Like if you think of every other tissue out there, you still call it a Kleenex sometimes. I think that's what's happening with open AI right now and actually working through AI conversational chat box.
But G P T is. A family of language models. It was developed by OpenAI and that language model gets fine tuned, sitting on top of G B T. is a chatbot experience and that chatbot experience is chat. G P T. What else do you want to add to that explanation? Yeah, there's a whole, like I said, there's a whole bunch of different versions of G P T.
We're on three now. Who knows what we'll be on. Uh, when people listen to this podcast, even within three, there's different variations of three that are. Good at different things. Is that true or is it only two? I don't know. There's, there's different flavors of G P T that are good and bad at different things, have different costs.
Yeah. The older ones are cheaper, but you're right, we're on like this newest one. If you hear the words thrown out like Da Vinci three. like that is the newest, most advanced model that we have at the time of the recording. But like you said, there'll be new ones out and that's what things were there. Do you remember when, uh, everybody had been building on AI apps for the last couple years and then overnight they all got infinitely better?
Yeah. What did they do to get that? Yeah. Like a week or two before chat g p T was released, there was actually a G P T release where it got way, way better, and then you put on top of that. Version of it that was trained on a bunch of actual human conversations and what that looks like, the back and forth.
But to me, like the real novelty of chat G P T, yeah, everybody throws out the metric at hit a million users in five days. Like what I think was important about chat G P T was it was the first time that anybody could just sit down and have a play with it, and in a very rudimentary way. even train it to do like their own kind of custom little stuff.
So like they can tell it, Hey, anytime I do this, I want you to do this other thing or process this data this way. And it's really the first time that normal people could sit down and like get an AI to do its bidding. And so maybe that's like not a novel thing for developers, but for Normies who could just spin up that chat app.
Like that was significant. Yeah, I think there's like the flex right now on, on social media is people that we're using. The API to do what chat g p t is doing, sort of pre-chat G P T. And now we're in this world where just millions of users have played around with that conversation and it's opened up their eyes to songwriting, joke writing, , all kinds of, all kinds of fun stuff.
But look, fundamentally chat, G P T is using a version of G P T called chat, G P t three point. and you know, it just gets stronger and stronger as more people use it because it has this sort of supervised learning and sort of reinforcement that happens the more people use it. Yeah. I've heard the analogy that it's AI's killer app, kind of in the sense that, you know, the camera was the iPhone's original killer app is you've got this really cool te.
but what is an application for it that will fundamentally change how people work, how people will make decisions, that sort of thing. Time will tell if that's the case, but it definitely feels like we've kind of gone past an inflection point. Yeah, and remember, this is just one company. This is open ai, who partnered up with Microsoft and is running this.
Just kind of first to market system. When you've got the others that are gonna be releasing theirs and there's other companies that are doing this, we're gonna live in this world where just like you're choosing the version of the car that you want to drive, you are gonna be able to choose which model of AI you want to use in your application as well.
Okay. That's enough of us. We got a ton of Chad G p t questions. Let's cruise through 'em. How many unanswered questions do we have in that base? Questions? The base questions currently has 173 records. Holy smokes. Pat, create a new view for the base called chat, G p t for any questions that include the word chat, G P T.
I've created a new view with the name chat, G P t. It is 21 records. Wow. We, I'm not sure we'll get through all those. Chad, how do you make it start the thing? Let's start with the first question from Daniel. When I ask Chad g p t questions, the answers often aren't right, but I see everyone raving about it.
Am I missing something here? Oh, boy. I mean, how many times do you write something you're like, ah, yeah, that's on me. I didn't give it enough, uh, information to answer that one. How many times has that happened to you? A lot. I think where people get stuck is, is it on me? Like should it just be that good? But like there is a very real element of when it gets something wrong, it is oftentimes a reflection of ambiguity in the prompt that you gave it.
So it's actually made me more aware of like trying to be more literal and careful about the words that I choose. But unfortunately, A lot of people's experience with it right now is they come in, I call these YOLO prompters. They try it out for the first time and they're like, what's the capital of New Mexico?
Or what's 135,000 times 12? And it gets it wrong. And they're like, ah, this is dumb. And then they move on with their life, and that's like it. , which I think kind of misses the point, but that's what Daniel's getting at here. There's a very clear relation between what we're dealing with here and like the relationships in our own life.
If you think about the strongest like relationships you have, they're probably because you invested time, you asked better questions, and you invested in the relationship. If you just are in passing and talk to somebody and you know, ask a cursory question and you move on, you're probably not gonna have a pretty deep or effective relationship with them.
It's kind of like what AI is like, right? , the more you put into it, the better questions you ask. So in the hierarchy of your personal relationships right now, where would you say chat e p t slots in . Right. So, okay. Outside of like the relationship analogy, there are ways you can look at chat, e p T, and one of the ways is through an assistant.
So not necessarily. Having it be the perfect answer, the perfect solution to something, but something to help you along with your day. How do you think about chat G P t from the lens of an assistant? Part of the problem is people come to chat g p t with so many different intents. I want an answer to a question.
I want to help with a thing. I want to build a personal relationship with it so I don't have to do that with other people, . So it's like, it kind of comes down to how you're planning to use it. And honestly, some of the. AI models we're seeing are a little more intentional in that regard. It's specifically for things you would normally type in Google or specifically to give you factual answers with citations, whereas chat GPTs like Chuck, anything in there, we'll see what happens, which in some ways is like part of the fun of it too.
But Google searching is just one use case, and so there's a bunch of different ways to use it, I think where people miss the point. They find something that it doesn't do well when in reality, I think you have to approach it from the opposite angle is it only has to do one thing well for you, for it to make your life easier.
So like let's just double down on those couple few specific use cases where it's like, wow, this is actually super handy for me right now. I'm gonna use it for this. Yeah, absolutely, and, and don't be afraid to give it information. I think that's the biggest unlock right now with the versions that we have at our disposal, is that we assume that it knows all these answers, but if you can point some data at it and then ask questions about that data, you'll be far better off.
And to sort of put this one home, say you're working through issues with like a Google Sheets formula or something, or an Excel formula. Give it the formula and then. Hey, I'd like to add a third column that does this instead of just making it redo the full formula. So again, giving it that information is going to completely make your prompt and your experience a little bit more accurate and a little bit more enjoyable.
Yeah, put it a little more on rails. Pat, can you change that question Status to answered and read out the next one, and in the future, do this each time we ask you to go to the next one from Angelica. I often see chat G P T referenced as a writing assistant, but I don't want to be among the people spitting out AI generated articles.
What's the happy balance? Yeah, there's a lot of bad PR right now around AI generated writing as there probably should be. Honestly, that's gonna be such a big theme this entire year as this. Aha. We caught you. Like the balance between people trying to use it to be more productive versus people going out and trying to track down the, you know, the big organizations who lean on this heavily.
Yeah. I think you're gonna see a lot more ai. Text generated checkers. Mm-hmm. , almost like plagiarizer checkers. You're not gonna know which ones to choose or which ones to trust. You're gonna have a lot more, uh, skepticism when somebody writes something that doesn't sound incredibly specific or like how you wrote or sounded, you know, three months ago.
And people are gonna always be wondering, was this written with AI or. Do I care? I don't know how I feel about it. The most important thing for me right now is to understand what it can do so that I can make a better and more informed decision of like how I want to use it in the world. You're always going to have people that want to game the systems, and there will always be people trying to stop people from gaming the systems.
Think about Google's entire existence right now. Seo. How many times has the algorithm changed over the last 10, 15 years for what gets shown on the first page? We're just gonna go through that secondary revolution now when it comes to text and ai. Yeah. The way I think about it is just assume that someday AI generated text recognition will be built into your browser.
so that if it detects something is written by ai, the moment somebody comes to that page, they're probably gonna bounce. So I think just about everybody's in agreement. Nobody should be just sitting down and publishing articles that are just vanilla AI generated. But then like what's the happy balance?
And that's really, I think the heart of Angelica question a few things for. Where I find it helpful, I think most people are like me when I say I can't just sit down and write something cohesive in one sitting, like I can't just sit down. It doesn't just flow. So for me, ideation is helpful. Or if I have kind of a core idea.
Great example. I was talking with some people the other day about writing a blog post around a very technical subject matter. It was how to deduct under your business, like the business use of your office and home. And we kind of iterated on that idea a few different ways and said, How can we combine this with like pop culture ideas to make this into something funny?
And so we iterated with Chad g PT on a few funny ideas and landed on like doing it through the lens of the office. Like characters from the office are going home, they're gonna work remotely for the first time, and can they deduct the stuff that they have at home? Or if you're a business owner and you do this, can you, and it actually gave us a bunch of great ideas for how to make that really funny, even through the lens of specific characters from the.
When we're writing something technical, adding a prompt at the end to say, talk like a pirate, or talk like Jerry Seinfeld or talk like a used car salesman from the eighties or something. It's really funny. I tried one recently where it was like in the voice of Rick and Morty explain. A very sensitive and like important, important topic, and it was like, hey, so it was just, it was really funny the way that it wrote it and I don't know, like there's gonna be so many use cases for just being fun, but also being super practical.
And I, I like what you. talked about it with ideation. For me, it's like, okay, what are the 10 ideas I should talk about when we're talking about, you know, learning about Google app scripts or something, and then it like fundamentally breaks things down. You're like, oh yeah, I forgot about number three. So, While we can use it for ideation, we can also use it for our own learnings, and then we can write whatever we want.
I also like it for editing too, like change the tone of what I've written into something more casual or more professional, or add a bit of humor to this that always seems to like make it better. So I don't see this as like a doom and gloom situation. I just feel like we're moving into a new era of quality and content that.
It's just generated in a different way. I mean, mm-hmm. , I don't know if I'm going to be so anti, like, was this written by ai? If it's super helpful. So I don't know if there's gonna be some revolution around helpfulness versus effort, but right now I'm in the camp of I'm okay with AI generated content if it helps me because I am using these tools to help me right now.
And I don't mind that it's AI generated. So we'll see where this goes, but it's only gonna get more exciting or. Contemporaries using it as an editor. That was a big unlock for me. You touched on this before. The more you can put it on rails, the better. So you give it that bit of writing and you say, give me feedback as an editor, but then like clarify, what do you want that editor to be considering?
Is this a professional voice? Is this a playful voice? Is this look for ideas to make this funnier? Another phenomenal hack I saw on Twitter as somebody had kind of designed. A one paragraph prompt that instructs chat g p t to, based on a block of writing of yours that you give it, write a descriptive paragraph about your writing voice that an AI can understand.
So the idea is I've got this blog post and I want AI to describe exactly what my voice is. In a way that's reusable so that the AI gives me a block of text that is a descriptive paragraph of what my voice is like, so that I can then reuse that paragraph, that descriptor in another chat conversation, and it immediately understands how to write something in my voice.
So I also love the idea of creating this kind of reusable assistant, or maybe you're iterating on that editor to say, no, edit a little more this way, and identify situations where my voice isn't in alignment on this or that. And then you keep kind of like iterating on that assistant to make it as helpful as possible for you.
Oh, you know, just Right. Why even use AI just right. For me, there's some benefits to using chat G P T to help. What else have you got for us, pat, from Tanya? I'm just overwhelmed with where to start. What are your favorite use cases? That's kind of a, a almost a hard thing with chat GBTs. It's not on rails, so there's a lot of people that use it for the things they would normally check into Google.
We talked writing and kind of all the different ways that that can help with your writer's block and generating more creative ideas. Something I've toyed with a bit that's actually really interesting is data transformations. So you've got data in a certain format, you give it a cs. and you say, I want you to apply this set of eight transformations to the data.
Remove this column, add a new column that concatenates these two bits of texts. The one thing you gotta look out for is math. It's really bad at math right now, so don't try to use it to like add things together or do arithmetic. Sounds pretty complicated. Maybe we just shouldn't use this stuff. We're just not gonna use it.
Now you're gonna give up this personal relationship you've built with chat G P T. Yeah. I don't even know why we bother. . Okay. I think Chad needs his mid-afternoon oj. Uh, any other use cases? Uh, it's really good at like poems, haiku's, when it hit G two, 3.5. It just got really good at those things. I've seen people make entire games out of it where you're like, here's the rules of the game.
And when somebody submits something they can pick from these menu options. I've seen some interesting things. Ever see the movie? Terminator? Terminator? I'm, yes, Chad. I'm sure that's where this is all gonna end. Uh, I run a weekly newsletter. I've even gone as far as like feeding it as CSV of different data and saying like, all the bits that I wanna include in the newsletter.
And saying, give that back to me now in mark down format with like the bold and the headings and all of that stuff. And then being able to pull that like into your email newsletter builder. So I don't know, maybe this question is, is hard to answer because it's so open-ended. I would think more about like what are the things I'm doing day-to-day specifically through the lens of those things.
Is there a way for chat G P T to make that easier for me? Anything you'd add to that, Chad? Yeah, like the matrix. iRobot Chad, her. Chad, let me, we're gonna wrap it up here. Thanks for tuning into Automation Pod and be sure to keep those questions coming. Chad, come back to me. Can you get him some water? Jack?
Can you hear me? No. Or
what did he say? Just give up. Just give up. The old way is better. Something's wrong. Is it the potato zone? It could be. Why don't call an ambulance. Uh, I'll get 'em laid up over on the couch here. Come on Chap.
Does it seem like he's been back there a long time now? . Paul, did you see this article? Davis Mayor Campaign, off to a bumpy start. Residents of automation town have reported receiving flyers from the Davis campaign with unusual messaging, uhoh, unusual messaging, like, is that the G P T generated text? Mm mm Cindy, I've sent in a picture of a postcard that said, I'm a language model and do not have the ability to generate flattering copy about a middle-aged male who loves automation and is running for, oh my gosh, that's all that was on the postcard.
That's all that was on the postcard. Oh, they've even got a quote from Jake here. It's unfortunate to hear Mr. Davis is struggling with what looks to be a runaway ai, yet another example that the world is simply a better place without it. Well, that's one way to kick off a. Yeah, that probably wasn't a good idea.
Boys. What? Gil Boys or Jason? Hey Gil. You met Paul the other day, right? Him. Really? So what are you in for? I went out with the team the other night to celebrate the promotion. Right. I came back with something Uhoh. It's okay. They'll get it sorted. Hey, great news though. Here we go. So all my coworkers want to learn live flow now that it's kind of.
Secret weapon. You haven't been on quite the journey since you discovered live flow. I have. It's been incredible. So check this out. I've now set up budgets for all the people who work under me. I've never been super into budgets. It just seems really arbitrary and you never know where you're at on the budget throughout the month.
Okay, but check this out. Since live flow syncs QuickBooks data to Google Sheets live, that is what the live bit means. Yep. That's the live part. I've been able to set up live updating budgets for each of my team members in Google Sheets. Wow. Okay. So I know some fancy new age credit cards do this now.
Kind of budgeting on the fly, but honestly, in my experience, Since that doesn't include activity from the checking accounts and invoices you receive and all that, you still have to wait until the accounting's done to get a true budget versus actual. That's right. So since this is connected to the QuickBooks file, it syncs all of that out live to the budget versus actual report.
That's pretty cool. Your last podcast gave me another idea. I could then use some other app builder to set up a mobile app for my. On top of that Google Sheet data. Oh, Gil. Then each of them can see their budgets updating live in the mobile app. Beauty, all because Live flow's syncing that data from my QuickBooks file automatically.
That's great, Gil. That might be some of your best work yet. Appreciate it. Jason, I owe it all to you and Chad. Hey, good luck with whatever you're doing here. Automated job. Sorry. How'd you get connected with that guy? Ooh, nurse is coming. Let's go. He's back here. Thank you. Wow. Look at all this. He's just, he's like a zombie.
Chad, can you hear us? Amelia Paul. Whoa. I came as soon as I heard. I, how is he? I don't know. They just let us back. He's just, he has this distant look on his face. Hello. Are you the family? Something like that? Yes. They had us waiting in the lobby for so long. We thought maybe something had happened. Is there an update?
Well, we lost him. What? Oh my God. No. We brought him in outta the ambulance and logged him in. But when I came into the room, he was gone. You mean. Eventually we found him. He was just shuffling around. Okay, that's a, uh, relief. So what do you think is going on? Looks like he's a vegetable now. Oh, oh my, oh my gosh, dad, gosh, no tuber really Look at him
But guys in another place right now, what your friend takes an anti-anxiety medication called potato zone. He was prescribed a 20 milligram dose, but what's in his system right now is, More. There was a potato on. The drug is made with enzymes from potatoes, so right now he's quite literally a potato, so he's not technically a vegetable.
In a few hours it'll start wearing off and he'll come too. But right now what he doesn't know could fill a book. What? Give him a few hours. He'll be back to normal. He'll probably be out of here by the end of the night. In the meantime, let me know if you need anything. The cafeteria will be open for another couple of hours.
Thanks. Thanks.
Was he just me or no? That that doctor was? I don't know. I'm just glad. Chance? Okay. Well, not much to do until he comes around. Don't feel like the two of you need to. I can let you know when he starts coming too. Are you sure? Yeah. Don't worry about it. Okay. Get outta here. Bye. Chad.
You gave me quite scared tonight, Davis. Oh, he should have seen Paul. Amelia turned up. I think they're gonna be just fine. Oh, also, Bad news about the flyers. So we were sitting out in the waiting room. I came across this article about the campaign being off to...
AutomationTown is written and by Chad Davis and Jason Staats edited by Paul O'Mara. Keep up with the characters of Automation Town on Twitter at @AutomationTown.