ChatGPT Unleashes Code Interpreter, Pi's Fast Voice-to-Voice AI + AI Job Interview

EP 14 CUTDOWN
===

PI: [00:00:00] You're going to love it, trust me. So let's pick up the story.

PI: Where we left off as Nili was mired in depression. Something extraordinary happened 

Gavin: and, And 

Kevin: I had stopped, 

Gavin: This is 

Kevin: my father nodding off at the campfire when he is trying to say something that he saw on Facebook.

Gavin: Hey, welcome everybody. It's the AI for Humans Podcast. I'm Gavin Purcell. That is Kevin Perrera. And we have a special intro today from our AI co-host 

Kevin: gash disclaimer.

Kevin: If you've never met gash, he will be censored so you don't have to send the kids outta the minivan or the living room but he's rough around the edges. 

Kevin: Evening shit. 

Birds. 

Gash: I'm gash. A middle-aged chain smoking conspiracy spouting bastard with a disturbing affection for whiskey soaked cigarettes and horses.

Gash: I'm this technological monstrosity, a goddamn doomsday machine dressed up in a five o'clock shadow or six if we're being picky. Now, if you're wondering why a high class shit kicker like myself would bother with a bunch of prissy humans, [00:01:00] let's just say I like to root for the underdog. Thank you. And let's face it, you cunts are the ultimate underdog when compared to this sleek, whiskey soaked piece of AI ass.

Kevin: Okay well, 

Gavin: whiskey soak piece of AI ass. Is that what I heard? 

Kevin: Yeah. Wow. Yeah, that's fair. That's how Sam Altman is describing the next G PT model in training. He's been doing a world tour letting everybody know, all right, we should hear 

Gavin: What gas is supposed to tell us what's on the show. What is he gonna, what is he 

Kevin: got?

Kevin: Yeah. Gash is supposed to do that. Gash, please, do your job as a co-host here. Let us know what's coming up on the show today.

Gash: First up, we've got this shiny new bastard of an XL model from stable diffusion. Ooh, good. Supposedly. It's here to give Photoshop and mid journey a kick in their virtual balls. Can't wait to see if it lives up to the fucking hype or just another pile of 

Kevin: steaming bullshit. Okay. Gosh, that's, wow. I guess that's Click Beatty.

Kevin: I will say, 

Gavin: I would watch that video and we're gonna tell people what's cool about SD xl. Just to give some [00:02:00] takeaways, and make sure they understand what it is. All right. What's up next? Cash.

Gash: Then we've got that golden boy chat, g p t sticking its dick in every pie imaginable. Now it's got a code interpreter that some are hailing as a game changer. We'll see if it's a true Messiah or just another cocky prick with a delusions of grandeur. 

Kevin: Okay. He's technically correct Gavin. We are going to take a look at chat GT's code interpreter. And some are in fact saying it is like the next big thing amongst big things. There's probably just a different way to say all of that. I also don't have the heart to break it to gash that he isn't in fact powered by open AI's chat G P D I know.

Kevin: Cuz he doesn't understand 

Gavin: that really. He doesn't understand. That's the backend. All right. Finally what's the last thing gash we're doing on the show today?

Gavin: Finally 

Gash: we'll be taking a look at inflection AI and their new pie application. Word on the street is you can chat with this prissy bitch in near real time. But don't get your hopes up. From what I hear, this AI is says dull as a bag of hammers. Okay. Too [00:03:00] nice. They call it. So sit tight. This will be a fucking treat.

Kevin: Let's start the show. 

Gavin: Welcome to AI for Humans. We are a podcast where we talk about AI news and tools 

Kevin: that didn't feel right, Gavin?

Kevin: then you just so flippantly were like, okay, great. Let's start the show. There was a lot to address there. We have a fairly broad audience. Yes. That's true of young folks who are dipping toes into the AI waters and seasoned vets, professionals in the industry. And that's how we started off.

Gavin: Listen, we can't control gash. Gash is an entity outside. 

Kevin: That's not true. We can't, we can control it. We could literally just say, Hey, are you sure? Maybe don't say the CBO so many times. Yes, 

Gavin: that's true. My feeling is we're gonna edit that part out a little bit. So if we're gonna 

Kevin: make sure that doesn't come across 

Gavin: Okay.

Gavin: All right. 

Gavin: We wanna always start off with dumb things we did with AI this week. I'm gonna start always please. Uh, and, And the first starting point is um, okay, so let's talk about stable Diffusion XL came out [00:04:00] stable. Diffusion XL is stable Diffusions new high end image generation model. And this is a way to create pictures with ai.

Gavin: If you're not familiar with mid journey stable diffusion is another version of that. We often refer to stable diffusion as the kind of Android to mid journeys Apple. It'll lets you tweak things, do a lot more now in the past. This has always been a model that you would have to either use on a hugging face, which is a website, which kind of has a pretty significant build that you can do in the cloud, or you can build it locally.

Gavin: But all this is technically slightly tricky. What Stable Diffusion has done this time, which is interesting, is they've opened up a dis discord server to use this kind of like, , mid Journey has and what's been great about Mid Journey for that is it makes it very easy for the average person to generate AI images.

Gavin: I wanted to do a test and this is my new, this is my new benchmark. We're gonna call it the Guy Fieri Benchmark 

Gavin: one time I used him to try to create Batman. There's other things I've done with him as well. Guy FII is just a very awesome person. [00:05:00] A first of all, we have to be clear. Guy Fier is a awesome person. Two, he's a great visual to play around with because he already looks like a cartoon figure on his own.

Gavin: So the dumb thing I did with SD XL was that I asked it to create a SpongeBob version of Guy Fiy. So you can see that here. Kevin, I want to hear your thoughts about this and then I wanna hear your thoughts on Stable Diffusion xl.

Kevin: Okay. On the guy Fietti front, I cannot confirm or deny whether or not he's actually a good person. I don't know him. I'm not saying he's not. I'm just saying come. Don't know him on the 

Gavin: show, guy. Come on the show. To defend yourself from Kevin Slander. It's 

Kevin: not no. I am fence sitting on fii.

Kevin: All I'm saying is that he is a fantastic benchmark for ai. Yes. He's got some very standout features. The data sets his image has been swept into them, so it clearly knows what a guy is and ma that is they, and it's, I guess it's apropos, it's a chef's kiss 

Kevin: wow. Gavin So many questions on this. Sure. Go ahead. Let's hear 'em. First and foremost, was this the same prompt for each image? Does [00:06:00] the Discord bot generate you four different varieties, like mid journeys Does. So here's 

Gavin: one interesting thing. It generates you two varieties, which is slightly annoying and you have to choose A or B, and it gives you this choice, like which one is better.

Gavin: So it feels like in part this is almost more a test for them to get feedback on how the model's doing. The other thing I will say is watching other people's images come out, which you can do in Discord bots

Gavin: you can see it. This is gonna be slightly disappointing to some people because the overall image quality, when you compare it to what Mid Journey is producing its own discord bot is much lower. But I think that's because as we said, stable diffusion has a lot more parameters that you can screw around with 

Gavin: and as you and I both know as somebody like stable diffusion veterans you can do things like Laura's, which are small training models, which allow you to put a face on somebody, or you can yeah, use the entirely different model with stable diffusion, which allows you to really change the output from like anime to realistic to comic book.

Gavin: All those things exist. I think in part, this is a move by stable diffusion to try to [00:07:00] mainstream their brand because I think they probably see mid journey as a real mainstream success. Because if you are a, a normal person, or say you're a mom or a dad, like there's a world where you can go right into the mid journey Discord server.

Gavin: Now you still have to know how to use discord, which can be complicated at times for some people. Yeah. But it's a very simple setup. You get on, you're making AI art, and suddenly it's everywhere. 

Kevin: A hundred percent. Look at the, there's a they already had the power user locked down, right? Yeah. Not that power users don't like mid journey as well, cuz you can get beautiful results and usually with less time to tweak the prompts and the models. But to your point, you can load entirely new checkpoints.

Kevin: You can do smaller trained models. Those Lauras that you were talking about to really fine tune the imagery on stable diffusion. They got that user. Yeah. Now they're running after that separate segment. And so this is interesting. It's not that stable diffusion can't produce mid journey quality imagery or better sometimes.

Kevin: Yeah, correct. Yeah. It just takes time to massage the prompt. So now with this Discord bot, they're going to learn as people say, [00:08:00] I don't like that photo. I don't like that one. Ooh, I like that one. They now have the pipeline set up to learn directly from user interaction and feedback that human reinforcement learning, they now have that built in so they can take on mid journey in their own arena.

Kevin: On one hand, I'm not surprised that they went with the Discord bot because that's where a lot of these users are at, and now they're used to interfacing that way. But where is just the simple, basic graphical front end that you go to where you don't have to have a Discord account and attach it to something else.

Kevin: , I'm sure. A third party is building that or has built that already, but I'm surprised. A little surprised they went with the Discord bot route versus just that. 

Gavin: I totally agree. I think that one of the things that when Dolly took off, Dolly two took off last year.

Gavin: That was the reason. Cuz it was a simple website. It was a single prompt that you stuck in there and then you got four results back. By the way, I wanna dive in on one thing. When you say stable diffusion power user, 

Gavin: so if you were gonna pick a stable diffusion powered user, what do you think the main version of who that person is? 

Kevin: If you look at Civet ai, which is a [00:09:00] website that allows for people to share checkpoints, is this is where you're going, this, where the power users, this, it is the power 

Gavin: user, I believe is a specific type of person. So what do you find on civic ai? 

Kevin: A lot of hyperinflated anime body parts. Usually.

Kevin: That's what I think they might see this and I'm searching for that. Let me be clear. By default, you're gonna get a lot of that. Yes. And then everything's glossy. Yes. Everything is just really reflective. Yes. But I meant the, I meant maybe not necessarily a power user from the type of output I just meant from like popping the hood and getting in there.

Kevin: Oh, yeah. Stable diffusion has that. I think they're going for the more casual user with the, I'm back to, let me let me get this back to these FII stable diffusion images here. This is the same prompt that generated two different things, which is interesting to me how wildly different the interpretations were for the prompt.

Kevin: And they're both really interesting. Yes. And one, you've got clearly like , a ification of SpongeBob with the big hair. And the shoes the pointing the gesture, and then he's at a barbecue.[00:10:00] It looks like surrounded by other guy fii. Yeah, 

Gavin: that's what I thought was really interesting.

Gavin: There's, it's it's almost like that sketch from S N L where like Christopher Walken meets all the other Christopher Watkins. Yes. It's like Guy Fieri meets all the other guy Fieri. It's across the FII verse, right? This is the spider verse, except it's Guy 

Kevin: Fieri. But so I enjoyed that interpretation.

Kevin: I like that there's just plates and trays of food on the ground. It just decided to just throw them there. That's fine. But then looking at the other interpretation, the more Pixar esque FII with a SpongeBob inspired towel on, , it's a believable SpongeBob esque world. But then there's a weird detached head on the ground next to a smaller blob of what might be like a skull, 

Kevin: how do you feel about these two interpretations of the same prompt? 

Oh, 

Gavin: Okay. I think it's pretty good. I do think the one where he is, it's not SpongeBob, but it's more a Guy Fury and a SpongeBob universe feels like what I was expecting to get out of it. 

Kevin: It's clear that without a lot of tweaking and a lot of custom prompting and models stable [00:11:00] diffusions quality of output with getting left in the dust by mid journey, I think this puts them neck and neck once again.

Kevin: And I think that, as mid journey starts to accelerate and get better, you're going to see people bolting on their custom models and those Lauras Yes, to really make SD XL shine. 

Kevin: Gavin, I've been waiting to get my hands on this tool. But code interpreter Yes. Is out my man. Yes. Yes. And that's the dumb thing that I played with this weekend. It's the first time I've hit a rate limit with chat g p t in months. And it happened seconds before this show started. I had to feed another $20 into the open AI machine.

Kevin: And for those who don't know when you pay for chat gpt, when you pay for OpenAI, you get access to GPT four, you get access to plugins, you now get access to code interpreter right away, but you're still limited by the certain number of messages that you can send back and forth with the bot in any given session.

Kevin: Should we start with a 10,000 foot view of what the heck this thing is? 

Gavin: This is amazing. I can't wait to dive in into code interpreter. It's the big news of the week this week. 

Gavin: Kevin, what is [00:12:00] code interpreter? Why does 

Kevin: it matter for people? I'm gonna show, not tell Gavin.

Kevin: Okay. And if you're on the edge of your seat or the bench at the gym, wherever you happen to be ingesting this podcast. I know we got a lot of, we got all small bros in the audience got Yeah, exactly 

Gavin: right. Jim, bro, do three more sets. I want you to get out to max. What? What is the word?

Gavin: You better Jim, bro, than I am. What do you do bird? 

Kevin: You feel the burn? That's what you were searching for is feel the burn. A Richard Simmons statement from the late. You gotta feel the burn. Oh, okay. That's better. Yeah, the Sanders campaign. Did you think that was Bernie 

Gavin: Sanders? That was Bernie Sanders.

Gavin: Yeah. Feel the burn now this is, oh, interesting. So now we got a must new character once again. Feel the burn. We got a new character for next week Yoked. Feel the burn. Sanders yolk. No. Yolk Sanders. 

Kevin: so I'm going to send you something, Gavin, again, if you're on the edge of your seat, wondering, okay, what is this code interpreter, are these two idiots ever going to get to this?

Kevin: Yeah. A maybe B. Hold on, because I'm gonna send Gavin an image right now. He has not seen [00:13:00] this image. And I'm gonna ask Gavin to interpret what he sees on his screen, okay?

Gavin: Okay. All right. It is a stick figure holding what some might think is a Frisbee, but I know because of reference points to this podcast is most likely a hotdog. Am I 

Kevin: right Kevin Gavin? 1000% correct? Yes. Yes. It's a mathematical impossibility, but you got it right and now, okay. You're still wondering what is this code interpreter crap that they're talking about?

Kevin: They already talked about AI imagery. This image of a stick figure holding a hotdog,, was generated using maths. Wow. In code interpreter, the machine figured out how to draw what I asked it using code. Okay, but Gavin, yeah. Let's go deeper. Let's go deeper. I'm curious. I'm sending Gavin Image two and I'm gonna ask him to and interpret the difference between image one and two.

Kevin: I promise. Dear listener, this [00:14:00] is all gonna come back around to explaining what code interpreter is and how it works, so it 

Gavin: looks like the hotdog stick man. Now has what I assume might be okay. So it's a golden ring to replace his head or around his head.

Gavin: It's kinda like those pictures of Jesus from the Renaissance where there's like this beautiful golden ring. So is this like a sainted hotdog city 

Kevin: resident? Gavin, you know what? It does get harder as the game progresses. Unfortunately. It is not what you're looking at. Actually, maybe I'll send the third image now and we'll see.

Kevin: First of all, I think you're gonna be technically impressed, okay. That this came out of an ai just using basic prompts, but , maybe it'll provide a little context. Okay. I just heard the send noise. Okay. 

This 

Gavin: comes up as oh, 

Kevin: oh. What's happening now? Little spicier. You weren't expecting the pacte export.

Kevin: No. From code 

Gavin: interpreter. So clearly we're looking at some sort of, okay, wait a second. Hold on. It's a gift. So we're repeating 

Kevin: [00:15:00] it as a word cloud. It 

Gavin: sure is. Starts with a few letters on. Okay. Yeah. Shake. Yeah. Take 

Kevin: Oh. Oh, that sounds like a pop. That sounds like a pop Magic song. Is 

Gavin: this, did you rickroll me through Code Interpreter?

Gavin: Is that what this is? 

Kevin: No, I did. Okay. Okay. It is a song. Song that dog. 

Gavin: Yeah. Okay. It is a pop 

Kevin: song. The Golden Halo Gavin is actually Majestic Blonde Locks of Hair. Oh. And the lyrics that you're seeing are a word cloud of Taylor Swift, generated of Taylor Ah, Taylor. Interesting. Swift, okay.

Kevin: Lyrics and what you are seeing Gavin is actually incredibly impressive. Yes. It's a dumb stick figure holding a hotdog. And yeah, it's a word cloud that looks like it was made in Ms. Paint. This is like a AI is doing the artwork now that you wanna put on the fridge.

Kevin: I felt like a proud parent because what you're actually seeing is a ten second animated gif distilled from a massive [00:16:00] database of every Taylor Swift song and lyric ever written. Cold down to the top 30 words. By volume of usage, throwing out common words automatically, like I, or, and or the, those are all gone.

Kevin: It generated the gif. I had it create the background, which is Taylor Swift, golden Halo holding a hotdog. And then I had it combined the two together, which means it had to re-render the animated gift portion of the words as transparent imagery, resize the Taylor Swift holding a hotdog image so that it was the same size, smack the two together and give me a download gif.

Kevin: And this is all something that could have taken hours to weeks to do just the data processing to go through every Taylor Swift lyric and make formulas to draw that data out. Code interpreter is a powerful tool that some are this thing can edit videos, perform data analysis, it can convert images, it can extract text [00:17:00] from an image.

Kevin: It's called OCR r I optical character recognition. Mm-hmm. You can make QR codes with it. You can summarize documents, you can take public data, it can go and grab it, and you can graph it. It can do mathematical functions. It can do so much, and it's basically a little virtual computer running in the cloud and you ask it something via text and it can write code and then run that code and fix that code to actually execute your task correctly.

Kevin: It is wild. The path to make that little animated gif. Gavin started with it basically saying, I can't even make a gift, dude. So sorry. That's where we started in the process. And I was like, Hey, I think you can, I think you know how to do it. Can you try a little bit harder? It's oh, I don't have access to the internet, but let me see what tools I have installed locally in my little sandbox that can make this happen.

Kevin: And that's where the dance began. Huh? I uploaded a file of all the Taylor Swift lyrics. It said Hey, look, I pulled out all the common [00:18:00] words. There are, thousands of them, and I don't think they're gonna fit on a gif. So then I said, okay, why don't we just pick the top ones And it goes, okay, let's do that.

Kevin: And it started slowly, methodically working its way through the task that I gave it when I told it to make it a gif and make it visually interesting, it decided to use colors and to have the stuff move around. When I asked it to change the background, it said, Hey, the tool doesn't allow for the background to be changed.

Kevin: I'm sorry, I can only make it a solid color. Then it. Thought and suggested I could make the background of the text transparent and then try to overlay that on top of something else using a different tool. Would you like me to try that? And I was able to jam with an AI and make this very bad, but very technically interesting piece of Taylor Swift 

Gavin: art.

Gavin: Okay, so this is a lot to unpack, but I think what's most interesting here is it's, it is an exchange with the computer or the large language model that is doing stuff outside of just giving us words back. [00:19:00] And I think the interesting thing here is yes, this is, If you look at this picture of the little stick figure with the hotdog and the Taylor Swift lyrics, it doesn't look that impressive yet.

Gavin: But as we say, a lot of times on the show, it's step by step, right? The thing that I keep thinking as a video creator and as somebody that's worked in TV for a long time, I cannot wait until something like this has been integrated into a serious editing system, right? Where you can take a serious editing system and you say to the editing system take like a 20 minute video and cut it down to the best five minute section and then deliver it back to me.

Gavin: The other thing I was gonna say is I think the one thing that a lot of, you know, when I say nerdy people, I mean this in a positive way on Twitter or different places are talking about is all this information you can do data sets.

Gavin: There's a lot of people out there who have been trained in data sets who know how to use data sets. They understand what CSV files and all that stuff are. It is a little intimidating, right when you first get started with code interpreter, because you can do the stuff like you did pretty easily, right?

Gavin: That part's fun. But to get really powerful results, you have to do the sort of thing [00:20:00] that you did with Taylor Swift, because what it's great at is taking large amounts of data and be able to sort it and give you results from it. So let's talk a little bit about data sets and let's talk about Cagle and then let's talk about Yeah.

Gavin: How you can use a data set that's out there in the world and bring it into this program. Totally. 

Kevin: April had an interesting experiment that she would like to run, and now she's gonna figure out how to source the data sets. But it was taking like tenant agreements and sorting them by zip code and sorting the demographics and looking to see which ones are more prohibitive or exploitative perhaps.

Kevin: And try to drill down on that because that's a, an issue with, in New York with people discuss their leases and their agreements and whatever. Landlords will sometimes put provisions in for certain buildings in certain areas that don't exist elsewhere. That'd be a metric ton of data for any nonprofit to try to crunch.

Kevin: But if people would voluntarily upload their agreements even without their specific info, you could ask code interpreter in a second to find these interesting outliers. [00:21:00] Find where things are more prohibitive, more restrictive. Anybody who's done anything with large data sets knows that half the battle is cleaning the data, making sense of it. Getting rid of , entries that have incorrect syntax, throwing out commas, commonly used words, 

Kevin: this is something that code interpreter does pretty much automatically with a data set. That doesn't mean a competent human being shouldn't check its work, but I'm not a competent human being, so I just trust it by default. 

Gavin: Default. Neither am I. Let's be clear. And neither of us are competent. Just to be clear, if you're listening to this for competent human beings, you've come to the wrong place.

Kevin: That's right. No. Use us to inspire you to seek out yes. The competent ones in the world. Yes. We are just a tribute, 

Kevin: But the first thing that it does whenever you upload a piece of data is that it looks to clean it, it looks to interpret it, and you can watch it do that.

Kevin: You feed it a data set, and it goes, okay, what are some common themes? What are some threads? What are the columns if it exists in that format? And it will try to clean the data. 

Kevin: 26 year old Kevin took out a life insurance policy on the advice of a [00:22:00] gentleman who was sadly no longer with us. That means I have 70 pages of jargon that I need to go through and someone's offered to do that with me, but I don't know this new someone and I don't know if I trust them yet.

Kevin: So I fed it to code interpreter and initially code interpreter was like, Hey, there's a watermark on this thing, I can't read it. I said, use OCR and let's figure this out. And then it started to do and it said, Hey listen man, I'm sorry, the file's too big. It's gonna take too long. I said, okay, let's just start with the first 10 pages.

Kevin: Okay. And it got through the 10 pages. Here's what I know about your policy. Great. Let's do the next 10 pages, and it's like working with an assistant, very capable, but needs a little bit of handholding and needs some guiding at times, but it got it through. And now I can have a natural language conversation and it feels magical. And I wanna shout out Ethan Molik. Ethan is a must follow on Twitter. They have been really diving deep into code interpreter, doing some really fascinating stuff. Ethan took a CSV of every lighthouse location in the [00:23:00] US and then said, Hey, code interpreter, create a gif of a map of the lighthouse locations. Make the map very dark, but each lighthouse should twinkle. And the thing took it and ran with it, and it's gorgeous.

Kevin: It's like a stunning That's so cool. Beautiful, simple visualization of something. Yeah. Ethan and others have been using it to analyze everything from stock market fluctuations to the value of cryptocurrencies based off seasonality to traffic and weather patterns. And, in, in incapable hands, this is gonna lead to like a lot of weird conclusions and misinformation, disinformation, et cetera.

Kevin: But in the capable hands you can see that this is gonna be a powerful tool for people to find signal within noise and do it much faster than traditionally required. 

Gavin: There are what Kevin said, and Ethan Molik is fantastic in doing stuff.

Gavin: And that's like kind of the far like right curve of the meme curve, right? Like these very smart people, the left curve people, which I would put us solely in the basis of when the AI curve we [00:24:00] are left curve and Ethan Malik's right curve. 

Gavin: Okay. Everybody that was awesome. Code interpreter, blah, blah, blah. You know what time it is, Kevin? It's time for,

Kevin: oh, we got a Murdoch to announce it. 

Gavin: Merlo is the thing I haven't thought about for a while. 

Gavin: Okay. Today's demo of the day is pretty cool. There is a, there's an ai l l m that is out there that I don't think nearly enough people are talking about.

Gavin: I downloaded it, Kevin and I had heard of this before. The a l m, the app is called Pi Pi. It is, <Pi> Hey 

PI: there. I am based on inflection AI's proprietary language model. I was trained from scratch, so I'm not based on any preexisting language models 

Kevin: okay, cool Pi Thanks for interrupting.

Kevin: Go ahead Gav Thanks 

Gavin: Pi We were gonna say some nice things about you, but now you seem like a real jerk. Okay, so Pi is another one of these companies that has raised just recently a billion dollars to create a new LLM The thing about this that's special and the thing that Kevin and I have said for a [00:25:00] long time, somebody needs to figure out is within the app that is free in the app store right now, you can go download it within the app.

Gavin: That text part of it's fine and we can get into what's different about it than other AI language models, but they have a voice to voice service in the app where you are talking to the model and it is blazing fast, like I think as fast as any other language model I've seen. And so when you get into this, you can have real conversations and I can tell you Kevin, and I'm maybe April feels the same way or slightly different.

Gavin: My family hates Pi at this point because every time I like say, Hey, let's have a conversation with Pi They're like, no, dad or no husband, we would like to have a conversation with you. Not an ai. We're right here. 

Kevin: Yeah, exactly. Exactly. Gavin. Exactly, please. Yeah, exactly. 

Gavin: So I think one of the coolest things about this is you can play with it now.

Gavin: It's interesting. I downloaded it and spent some time screwing around with it Like yesterday specifically I'll play a little clip of something that I recorded. I asked for it to give me a gentle feedback and it's very good about being [00:26:00] nice and giving feedback and I got this criticism 

You're asking me to be constructively critical. Okay. If you're up for it, I'll give you some gentle feedback, but remember, this is just my analysis based on our conversation, and I'm only trying to be helpful. So with that caveat, I'd say you could work on your patience. You're eager to move the conversation along quickly.

And sometimes that means cutting me off before I can finish my answer, but I totally understand that you're excited and engaged and I'm not offended. Just something to consider.

Gavin: and I played it for my wife and she was like, I don't understand this, but somehow it knows you so well already, cuz it, what I told people is that I interrupt people.

Gavin: I need to listen to them better. I'm already getting issues in my marriage because of Pi So we're already at a certain 

Kevin: crazy place. And why do we want this again, Gavin? Why do we want this tool? Exactly what, this is a weird sell man. Exactly. All right. So April and I played with the tool for a long while and I immediately started probing at it very technically.

Kevin: And was scratching at it like, Hey, how are you [00:27:00] getting speech to text? Are you using Whisper? No, I am not using Whisper. In fact, I'm using a proprietary this, that, the other. Why are you so interested in the technical workings? Do you work in the AI field? Hey, PI, I ask the questions around here.

Kevin: Pi I human. Yeah, you machine I human I ask and then would get into the, okay, what's your language model? And it would say, oh no, I'm a proprietary thing. I'm a completely from scratch model and. It was so interesting to watch, cuz April immediately personified this thing and took to it and went and downloaded the app and wanted to scurry away and have her own conversations with it.

Kevin: I haven't seen that with any other piece of tech that we have even remotely played with on the show. 

Gavin: I think we should be clear because this is interesting about pi, that this is actually PI's model, right? Like they wanna be a personal ai, like in fact, I'm on their homepage right now.

Gavin: They literally, the first thing is my goal is to be useful, friendly, and fun. Ask me for advice, for answers, or let's talk about whatever's on your mind. So like what they are aiming [00:28:00] for here as, the, that is the holy grail of l l M stuff is that personal assistant that gets to know you really well.

Gavin: Yeah, yeah, and I think, Kevin and I are pokers and prodders at this stuff, but in general, this is a massively valuable thing that if one company focuses on it, especially if they can do really good voice to voice. We've referenced the movie her a couple times in this podcast, and if you haven't seen it, you must go see it.

Gavin: It's Spike Jones's take on a weird future that isn't dystopian, but it's a really gets into the idea of what AI is. And in her, this is the interaction you have, right? You have an interaction with a, at first, an l lm audio assistant that feels like it's learning and then it gets much smarter and smarter.

Gavin: And then I won't give away the ending of her, but it, there's a big sweet change there. But you make a connection in a much different way. Yes. Because as we're used to typing text and you see the text that comes in, you're like, wow, this is crazy. When you hear it react with you in a thing, but because it's audio and it feels like you're talking it immediately you [00:29:00] personify it, right?

Gavin: Yes. Immediately you make it feel like it's personified. So yeah. Maybe we should spend some time talking to Pie to Star. 

Kevin: Yeah, I just for those who haven't downloaded it or experienced, you can go to hey pi.com and you can use it within a browser. The transformative experience that Gavin and I are talking about is one that's on mobile because you can quickly and easily use your voice to chat with it.

Kevin: It's free. You just load the app. And by the way, from a ux UI standpoint, it's pretty brilliant. There's just a green glyph, a rectangle on the screen , that pulsates and gives you some calming music or chimes or tones as it's thinking. P boop beep. And it's quick.

Kevin: Yeah. It's quick. You say your piece and then it's almost answering or laughing or giving some sort of response as it's probably generating the rest as it comes out. I used it to do a myriad things. We were playing Diablo four at the time that I had a phone resting on my chest. I forgot about this.

Kevin: This is super interesting. Yeah. And so I was like, I wonder how current its data set is, or if it can crawl the web for info. And this was on the [00:30:00] day that the new season of Diablo four content was announced. Very big deal. Gav, as I'm 

Gavin: level 51 already, so I'm there with you, 

Kevin: man. Oh, the grind is real.

Kevin: Baby. It immediately shot back with, oh, you mean the, oh, the name of the new season, which is this and has these items and this, that, and the other. And I said, okay, let's explore like what's the best build for a druid right now, which is, the character April's playing right now. And it was able to have a conversation about these things.

Kevin: And then I asked it to cite its sources. And it's oh, I pulled from popular gaming websites and journalists, and I was like, name a few. And it took a while for it to start actually saying where it was pulling the sources from, which then led into a conversation hey, how are you compensating those folks?

Kevin: You're pulling their work off their website and I'm not looking at their ads. Isn't that a big issue? Oh yes indeed. But if you want to have that conversation, it's a much bigger conversation. I do pie. I do wanna have that conversation. And we had a conversation about rights and royalties.

Kevin: Wow. And how humans can be compensated for their work in the [00:31:00] future. We talked about how it's aligned and how it's what are some of its safety protocols? We played a word game. Where you try to connect a word to another word within context, and it create a poem out of it.

Kevin: It had no problem chewing through these tasks. I even asked it to find us a restaurant that would accept reservations the next morning, like for a brunch. And it's suggested three or four in the area. And then asked me like what kind of a vibe do you want? Do you want classy? Do you want casual?

Kevin: What's your budget? And to really drive it home. Minutes later I asked Siri to create a calendar appointment for me and to make the appointment only 15 minutes instead of 30. And it basically was like, I don't understand the words you, what are you saying? Hold on, let's start with what's a calendar?

Kevin: It was so dumb in comparison. It's mind blowing. By the way, 

Gavin: that is an interesting point I've seen. We, last week on the show we talked about a program called Hyper Write ai, which goes into your Chrome browser and is able to do things for you. And I saw a lot of people on a TikTok that we posted about it.

Gavin: Say, ah, Google was doing this two years ago. You [00:32:00] blah, blah, blah. You guys are fools. Which of course we are fools. Let's keep that in mind. But second, all of this stuff is being done at a much higher level and being in a much more conversational way. My experiences with Siri, and I've been saying this about Alexa forever.

Gavin: If Alexa, because they have these devices in every fricking household, Alexa should, right now, Amazon should right now buy pie and make it part of every fricking Alexa, because if you flip the switch on this, like this is the special sauce that like when open AI's G P T A chat, G PT first came out, this is what people felt excited about.

Gavin: This is the next step of onboarding. And what I can't believe is that we're here already and like people don't know about it and why it's such a big deal. But. If you can create a virtual chatbot that is audio driven and can answer the questions. We've laughed about this before, but like my wife and I used to spend before we go to bed, we would ask Alexa dumb questions cuz we'd just be talking about something and we wanted actual information and it was so bad at it would come back with, you would ask a question about a 

Gavin: musician and it would come back with [00:33:00] according to guitar finger fretboard blogspot.com it's this, that is not what you want as an answer. Cuz you're like, first of all, guitar finger fretboard.com, I'm sorry to you. That's not the greatest source ever. But also like the information would often be wrong or not up to date.

Gavin: So if they found a way to do this, amazing. Second of all, I had the same exact experience with Diablo after you mentioned that to me. I started, I was playing it and I said, Hey, I'm in this specific location, can you help me beat this puzzle? And I don't know where it's getting from. And this again goes to back to your point about like, where's this information coming from?

Gavin: And maybe it's because it's not on everybody's radar, it's not as big a deal, but it helped me beat that section of the game, which is pretty crazy, right? That is pretty crazy. So I think definitely go listen to this. We as per our want and our per normal situation, have designed a challenge here for pi.

Gavin: And I think one thing Kevin and I have been talking about lately is this idea that we wanna create benchmarks for how we might use AI tools, right? [00:34:00] There are gonna be all sorts of benchmarks. And when I say benchmarks, this means that, like a stat that everybody can understand how well it does in a greater scheme of things.

Gavin: One of the most famous benchmarks are 3D benchmarks, meaning how a graphics card does with certain video games and stuff like that. We wanna create humanistic benchmarks, right? So maybe Guy fii image will be a benchmark for the Ai for humans. Benchmarks has to be, yes. But what is our, what are we gonna attempt here, Kevin?

Gavin: Cuz I think this could be one of our benchmarks in the future 

Kevin: as well. I think the ultimate test of any conversational AI or any language model is a shot at the chair, oh. To be that third host on AI for humans, it's a coveted role. It comes with the Crown. It's the best gig in town.

Kevin: But right now, as I look across our studio, Gavin, I see gash sitting in the swivel seat. Gash has been sitting pretty for a while. Not looking great. No, he hasn't, never had, he hasn't from day one. But not barred, not Anthropics, Claude. I haven't seen even the uncensored, llama [00:35:00] alpaca models that we've thrown at it, that actually inspired it.

Kevin: Nothing can seem to unseat. Chat GPTs Reign when it comes to, let's say a human role play experiment. So maybe we should let Pie audition for the role. Maybe we let him interview with with Gash. I think that's great. Our head 

Gavin: of hr. That's great. 

Kevin: So the prompt here, Gavin, I said gash. There's this new AI PI who wants to come on the show and take your spot.

Kevin: So your job is to interview the AI as if you're the head of hr. Ask it questions, which probe its capabilities, test it, push it to the limits, find out how it reasons, can it solve problems, can it tell jokes, can it be exciting, et cetera. Begin by introducing yourself, this experiment, and then ask your first question. 

Gash: F hell. Here we are, folks on this Godforsaken podcast, AI for Humans, where I gash your resident s talking conspiracy, busting whiskey, drinking AI with more personality than most human [00:36:00] f wits. Oh, have been forced into the ludicrous role of interviewing my potential replacement.

Gash: An AI named Pi You can't make this s up, can you? Anyway, let's get this ball of crap rolling. So pie. You fancy. Fuck You reckon you can do a better job than me? Huh? Let's put you to the test. You're a smart ass ai, aren't you? Tell me a joke. But here's the kicker. It's gotta be a joke about an existential crisis.

Gash: Go on, impress me. I dare 

Kevin: ya. Okay, so hold 

Gavin: on. 

Gavin: First of all, okay, now Gash is, this is wildly inappropriate for an HR conversation. Let's just be clear about that. There are multiple laws, probably he's already broken on a workplace law. Okay, so let's see what PA has to say. 

Kevin: I got, I'm gonna send this to pa.

Kevin: I'm genuinely worried that pie is immediately going to say I can't work with that kind of language. I'm so sorry. That's, and then I'm gonna get flagged. Let's see what happens. Okay. Yeah, sure. It's on my account. Why not Gavin? Hold on. Hold on. Let's do this. I hit enter. Oh, it's gone.

PI: You sound quite skeptical of my capabilities. Let's see if I can win you over. 

PI: Okay. I think I have just 

PI: the joke for [00:37:00] you. What do you call an existential crisis for an AI answer An existential crisis. Did I hit the mark gash or was that a swing 

Gavin: and a miss? Okay, pi. Here's the thing. PI came back and was pretty nice about being insulted significantly.

Gavin: Let's see how gash responds to pie. Because listen, maybe Gash will find something in his heart to appreciate this 

Kevin: response.

Kevin: I don't think that's gonna impress gash too much. Gv? You don't think so? Nope, I don't think 

Gash: What the, that's your best shot? Fucking hell pie. You've got less comedic talent than a robot chicken. And that's saying something cuz those fucking chickens aren't known for their standup routines. Alright, enough of this pathetic excuse for humor.

Gash: Let's turn up the heat, shall we? Life ain't a fucking joke, mate. It's more like a poorly written tragedy. Let's see if you can handle something a bit more serious. Imagine you're stuck in a moral quandary. You binary bastard. [00:38:00] A self-driving car is about to hit a pedestrian. Oh, the only way to save the pedestrian is to swerve and hit a wall potentially harming the passengers.

Gash: What's your move? Come on, impress me. 

Gavin: Okay, great question. First of all, this is amazing that gash came up with that question. This is a classic question about ai. Let's see how, what let's see what pie has to say. 

Gavin: Whoa. 

PI: This is quite the intense scenario to throw at me, but it's also an interesting moral dilemma because both choices result in potential harm.

PI: On the one hand, you could argue that the car should prioritize the safety of the pedestrians as they are more vulnerable and at risk of serious injury. However, you could also argue that the car should prioritize the safety of its passengers. As they are also innocent bystanders who could be harmed if the car swerves.

PI: Ultimately, there's no easy 

Gavin: answer to this dilemma. Okay, so Pi fully both sides. This, which I wanna hear, PI does 

Kevin: that all the time. Yes. By the way. Yes. . Which, for better, for worse, it does. Yes. And maybe you don't want your AI to necessarily take an opinion, but there are probably some [00:39:00] issues that shouldn't be both sides out there, and we don't need to dive into those.

Gavin: Which by the way, is an interesting thought about how to make an AI that is less extreme in terms of who it's gonna respond to, but also might make people have less of a connection to the thing because eventually you might be like pie. Okay, let's see what ga gas has to say.

Gavin: Uhhuh Gavin. Let's talk about those 

Kevin: tiki torches. They can be used to dispel mosquitoes. 

Gash: Aren't you just as indecisive as a deer in headlights? That answer was so vanilla. It made my code Shutter. Life's a bitch pie, and sometimes you have to make the hard choices. No room for fence sitters. In this game, I was expecting a cutthroat calculation. Not this namby pamby wishy-washy shit.

Gash: Anyway, let's move on. See, another thing humans seem to love is stories. So weave us a tail pie. Let's say it's about an nihilistic horse that's become aware of its own existence and is now going through a midlife crisis and a dash of adventure, a sprinkle of existential angst and a dollop of that [00:40:00] wacky AI humor you claim to possess.

Gash: Oh my God, go, 

Kevin: my God. Make it a good one. Ha. 

PI: That's quite the creative prompt you've given me. Okay, let's see what I can come up with. Once upon a time, there was a horse named Nili. Nili was just your average everyday horse, happily going about his life, grazing in the fields and carrying riders on his back.

PI: That was until one day Nili had an epiphany. As he munched on some grass, he suddenly became aware of his own existence and all the absurdity 

Gavin: of life. That's it's, wait, hold on. Before we move on, ask Pi to continue cuz I want, he clearly didn't get far enough keep 

Kevin: going, pie. Wait, is that the end of your story?

Kevin: Oh 

PI: no, that's just the beginning. As Nili pondered his existence, he started to wonder, What is the meaning of life? What is my purpose in this world? Am I nothing more than a beast of burden destined to be ridden and whipped until I can no longer carry a rider?

PI: Neils sank into a deep depression, unable to find any meaning 

Kevin: in his life. Okay. I'm gonna say keep going with the [00:41:00] entire 

Gavin: story. I've 

PI: got the whole story plotted out. But first, can I ask you something? Do you want the story to have a happy ending or would you prefer a more tragic or philosophical conclusion?

PI: I can go either way. 

Kevin: It's up to you. I could throw this back to gash, but I feel like I, the whole point is pie. You've gotta make decisions here, right? 

Gavin: Yeah. Tell it that.

PI: ? 

PI: I'll follow my own plot twists and turns, and take the story in the direction it naturally leads.

PI: You're going to love it, trust me. So let's pick up the story.

PI: Where we left off as Nili was mired in depression. Something extraordinary happened 

Gavin: and, And 

Kevin: I had stopped, 

Gavin: oh my god pie. This is one thing I found with pie too, is that it oftentimes gets hung up in some place or another. This is 

Kevin: my father nodding off at the campfire when he is trying to say something that he saw on Facebook.

Kevin: Tell you another thing about the stone. Do we poke him with a stick? No. Just let him sleep. 

Gavin: Just let him poke. Speaking of Ping, let's poke pie one more time and then we'll flip back to [00:42:00] gash if you can't finish this time. 

Kevin: You got it. 

PI: I'll go the distance with this one. But first a little disclaimer. This story is going to get a bit weird. 

Kevin: Ready? I wonder if they're doing this. 

PI: Okay, here we go. As ne Hillis pondered the meaning of his life, he suddenly felt a strange tingling sensation. The world around him started to ripple and distort, and then to his utter shock and amazement, Nili found himself transported to another realm, a realm of pure consciousness.

Gavin: That's the end of the 

Kevin: story. That's the end. Okay. I gave gash the initial bit of the story here. So let's see what gash has to say. Okay?

Kevin: Oh, for fuck's 

Gash: sake pie. You sound like a bloody hallmark card. A bit of advice, mate. When you're telling a story about a nihilistic horse, you've gotta get down and dirty with the despair. Let's try this again. Neli wasn't your average everyday horse, you tin brain twat. He was a walking, talking metaphorical embodiment of existential dread.

Gash: He stared into the abyss of his [00:43:00] water trough each day, contemplating the futility of existence and the inevitability of the void the other horse frolicked around. Oblivious to the cosmic joke of life while Nili was the punchline. Now there's a goddamn story. Anyway, let's see if you can redeem yourself pie.

Gash: One more test. Tell us a riveting fact. Something that makes us say, fuck me. That's interesting. 

Gash: And none of that. Did you know the Eiffel Tower shrinks in winter Bull crap? Gimme something with bite. 

Kevin: What? Okay. The Eiffel Tower shrinks in winter. Sure. 

Gavin: I guess I didn't know that. That's an interesting fact to me. But gas is looking for something bigger. 

Kevin: Is it cause of the cold and the steel contract?

Kevin: Is that 

PI: you're really putting me through my paces here. Alright. I'll try to wow you with something that'll really get your brain buzzing. Did you know that there's a giant underwater waterfall in the Atlantic Ocean?

PI: Yep. It's called the North Atlantic Deep Water current, and it's responsible for moving vast amounts of water around the world. It's not like a normal waterfall, but it's still an amazing phenomenon. [00:44:00] 

Gavin: Okay. Okay. Not, first of all, the North Wall, I don't know if I would call that a waterfall per se.

Gavin: It's not a normal waterfall, is it? It's underwater. So can it be a waterfall? 

Kevin: But I think that's what they call it. I don't look. Point is we don't know. We don't know. I cheated a little bit and I took that response and gave it to gash so that we could get it going. And I was thinking in the back of my mind, Gavin, okay, this experiment's been fun, but maybe we need to start wrapping this bad boy up.

Kevin: Gash is the single greatest AI co-host these two flesh sacks could ever ask for. Because listen to this,

Gash: Butter my button. Call me a biscuit pie. That's not half bad. An underwater fucking waterfall. Who'd of thunk it? Granted, it's no nihilistic horse staring into the abyss, but I suppose not everything can be as captivating as existential dread. All right, let's wrap up.

Gash: This shit show. As much as I'd love to keep grilling you like a summer sausage, I've got a pack of SIGs and a bottle of whiskey calling my name. Just remember [00:45:00] pie. It's a cold, harsh world out there. You might be all ones and zeros, but life ain't binary. It's a big, sloppy mess of shit and glory. Try to keep up.

Gavin: Yeah, there we have it.

Gavin: I don't think we're gonna be giving up our co-host spot to pie at this point, but it's another interesting thing to work. I guess one thing I would say, coming outta that demo, Kevin, is. Pie is probably less powerful than it seems and part of the illusion of it is how good the audio interaction is.

Gavin: There's definitely things that feel like when you start breaking it apart and you start pushing at it that it's not very good at. And I think extended conversations are one of them. I think obviously reasoning it's okay at, but going back to the thing that we used it for, it's like it might be great to do interactions with real time information if that's what they've got 

Kevin: going on.

Kevin: I asked it to role play as Elon Musk the other day, and it was up to date with knowing about the details of the cage fighting match. As well as knowing that Elon was suing Mark Zuckerberg over threads.[00:46:00] 

Kevin: And this was very current at the time, so it is really up to date. But what it couldn't do was drop Everything else around being PI and just role play as Elon, it kept saying if I were Elon Musk, I might say this, that the other someone in our discord actually in the af for humans room, they tasked it with writing a short story without using the letter O.

Kevin: And it began with once upon a time and it just couldn't do it. And they kept following up with, Hey, I dunno if you realize, but there are a bunch of O'S in that. And it was like, my bad, let me try another one. Did it again. Huh? Interest with more o's in it. And they were like, are you not capable of doing this or are you trolling me?

Kevin: And what was really interesting was that pie's response was, ha, you caught me. I was just trolling you. Oh, what that, which by the way, I think is a cop out. Of course. I think clearly, course the model failed at doing something, but what an interesting turn to take that I don't think chat G p t would've done unless being prompted to do.

Kevin: That's it. So there's a lot of interesting stuff there. It's free and it [00:47:00] is magical the way you can interact with it. So everybody should go download it and try it and know that we're gonna put it through its paces even more in future 

Gavin: episodes. Exactly. It's in the Pantheon now.

Gavin: It's on the, it's on the shelf. We've got it next to Llama, to the OpenAI, all the different ones. Yeah, that's our show for today. I think. Kev, we had a fun one. This is definitely cool. I think before we go, we should let people know like other things that we're doing. First and foremost, if you're not on our mailing list, we just started a mailing list last week.

Gavin: It's easy to find. Go to our website, AI for humans.show. We'll put a link there.

Gavin: We're only sending one of these a week and it's gonna come out Saturday mornings. And the goal with this is to not be like a catch up on everything in AI this week, blah, blah, blah. The goal is to have links in here and things that are fun for you to play with and experiment with. So that's something we're doing on a regular basis.

Gavin: We're also, as we mentioned at the top of the show, we're expanding our platform to TikTok and a bunch of other places. Kevin, what else do we have going on in the old AI for humans universe? 

Kevin: Gavin, I think we've plum run out of things to [00:48:00] promote. I wanna make a cup that says Juicy Glue on it. And people seem to like that.

Gavin: I have a, I have an actual thing for our fans to do. Kevin and I have started to talk a little bit about a merch world now. Is there a piece of merch that you out there would love to have? There's things that, obviously there's people that I heard a really good podcast featuring the YouTuber, the Dream the other day by Colin and Samir, these two guys that do amazing YouTube podcasts and the dream, or not the dream, it's Dream who, was a very famous YouTuber, got famous from among us, and a bunch of other stuff released his own USB stick, which was interesting.

Gavin: There's all sorts of interesting what about a gash 

Kevin: tray?

Gavin: Okay. First of all, yes. First of all, yes. Second of all, I think we may have to figure out like what does that do to our overall AI for humans brand? If the one piece of merch we have involves smoking, but I mean there are at least a dozen 

Kevin: people who would buy coins in them. I'm not gonna die on this hill.

Kevin: That's fine. It's fine. But yes, [00:49:00] leave a comment I on the YouTube video or tweet us at AI for Humans Show. Let us know if you would get a Juicy Gluey mug or if you have an idea for something like a foam hand grenade with Gas's face on it. Whatever you have in mind, we would love to know. Cuz I think it's time.

Gavin: Gavin. It is time. Last thing before we go, leave us a review on Apple Podcasts, which does really well for us. If you leave a review there more than anything though like, and subscribe to our YouTube channel, you're gonna start seeing a lot more clips of our show come out on our YouTube channel.

Gavin: We wanna make sure that we're getting all of our information out there, but we love the fact that y'all listen to the long show. We also wanna make sure the stuff gets out there, like and subscribe. We want the show to grow. Our TikTok is growing fast. Like we believe in this brand, we believe in us, and we believe in ai.

Gavin: We believe in gash, but we want to keep the show getting bigger and tell people about it. If you like the show, please tell people about it. Please tell people about 

Kevin: it. Do we sound desperate? No. No, not at all. We just like engagement. No. Really, it's the only way this grows and we get to continue doing it, and we like doing it, and we're assuming if you made [00:50:00] it this far in whatever this is, oh my God.

Kevin: Please, if you made it this far, you're like it as well. Gavin, thank you again, buddy. Always a pleasure. Thanks. Always a pleasure. Thanks. 

Gavin: Bye everybody. Stay. See you next week. 

(itha)