OpenAI Drama, GPT-4o is VERY good & Google I/O AI Insanity

Audio Version Ep58
===

[00:00:00] 

kevin-host815_1_05-15-2024_094041: A, I, drama, alert, there have been major shakeups at OpenAI. Some key players have departed, but don't worry, they were only responsible for making sure the machines don't destroy all humans.

gavin_1_05-15-2024_094041: new GPT 4 O model has some incredible capabilities that flew under the radar. We're going to show you just how powerful and maybe kind of horny that model actually is.

kevin-host815_1_05-15-2024_094041: All that, plus we have got real world reactions to Project Astra, along with a look at their brand new image generation software.

kevin-host815_1_05-15-2024_094041: It's a wild week for AI, but we're covering it all for you right here, right meow. On AI for humans.

Mhm.

gavin_1_05-15-2024_094041: Welcome. Welcome. Welcome, everybody. This is AI for humans, your guide to the world of generative AI, we are here to demystify the wild world of artificial intelligence. And good Lord, this was a week, we have a lot of stuff to get [00:01:00] through.

gavin_1_05-15-2024_094041: Today, we have a very exciting show. And Kevin, I just want to see how are you? How you doing?

kevin-host815_1_05-15-2024_094041: I am well, buddy. You know this because we've been hanging out all week, covering the massive news from open AI, the two hour rodeo that was Google IO. And we've been making content all along the way. It's been a delight. Thank you to everybody who was engaging by the way. I want to take a second to say that if you like any of the content that Gavin or I have made this week, please toss us a like, a subscribe, a thumbs up, leave us a five star review on Apple podcasts.

kevin-host815_1_05-15-2024_094041: If you do, we'll read it at the end of the show. That's our guarantee to you. , but engage cause it helps us out. But Gavin, there is still massive stuff. As much as you and I have been talking about AI all week, we've still missed a ton of stuff.

gavin_1_05-15-2024_094041: that's right. And now Kevin, it is time. We have to get into it. It's time for the We are starting off with a [00:02:00] juicy, juicy piece of AI drama this week. Ilya Sutskever, a co founder of OpenAI, and formerly the chief technical officer, is out. Now, this, if you've been following the OpenAI drama like we have, like a soap opera that's been going on for months and months and months.

gavin_1_05-15-2024_094041: You remember back, uh, late last year when Sam Altman was kind of immediately fired and no one knew why, Ilya was part of that conversation to get him fired. Now, we're not really sure exactly how much, but clearly

kevin-host815_1_05-15-2024_094041: We know he sided with the board cause he voted. And then apparently Gavin was placed under the floorboards at open AI. And in between the creeks of your sneakers on the hardwood, you could hear him breathing desperately, wanting to

gavin_1_05-15-2024_094041: where is Ilya? Where is Ilya? Was the entire Internet's people were chart were chanting again and again and now we know Ilya is no longer there He's leaving today as of yesterday Sam Altman posted a long post It was a very nice post about how sad it will be to not have Ilya at the company

gavin_1_05-15-2024_094041: anymore 

kevin-host815_1_05-15-2024_094041: GPT did a great job [00:03:00] of writing his heartfelt letter, for sure.

gavin_1_05-15-2024_094041: , but this is a big loss for open AI. And I think one thing we, you and I have been following this space for a while, but Kev, just to kind of talk through what Ilya was to this company. I mean, you and I have, have listened to a lot of his podcasts. I've listened to him. The Lex Friedman episode he did is really worth listening to because he's very smart, very intelligent, goes deep on kind of his thoughts on AI.

gavin_1_05-15-2024_094041: What do you think caused all this? And why did it happen now?

kevin-host815_1_05-15-2024_094041: No one knows for certain. Just to touch on how impactful and important Ilya was to the company, Elon Musk in early days, he was the first to back open AI and put money in to the endeavor.

kevin-host815_1_05-15-2024_094041: He had said that one of his greatest accomplishments was recruiting. Ilya was stopping him from going to work, I believe it was at Google, and bringing him in at OpenAI. He is a brilliant mind, and every time people talk about this bizarre, ethereal, competitive advantage that OpenAI has, this head start on all these other companies, how do they do it?

kevin-host815_1_05-15-2024_094041: What wizardry? What sorcery? [00:04:00] A lot of people point to Ilya. He's certainly not the only one, but he was definitely there. From the very start and was a major, major player and getting to the why of it all.

kevin-host815_1_05-15-2024_094041: Some of the rumors are that it's, you know, the internet refrain is where's Ilya, as you said, but it's also what did Ilya see because all of the rumors around the time that the board rallied around ousting Sam Altman was that something might've happened. Behind closed doors. They got a glimpse into the future and they said, we need to either shut this down or slow it down or build a nice protective wall around whatever this is.

kevin-host815_1_05-15-2024_094041: Because that was Ilya's passion. He wanted to make AI safe. One of the initial, , goals of open AI. And so, A, I'm curious, do you think that's the case? And B, there was also another fairly,

gavin_1_05-15-2024_094041: was just gonna get

kevin-host815_1_05-15-2024_094041: yeah, well, let's talk about it. There was another departure which kind of speaks to that.

gavin_1_05-15-2024_094041: Yeah. So a guy named Yon likey, I think that's how you pronounce his name. I'm sorry, Yon, if that's not how you say your name or if it's Jan, it's J A N. [00:05:00] Yon, it was co leading super alignment at open AI. And he just tweeted yesterday, I resigned, which is a big deal, right?

gavin_1_05-15-2024_094041: So now you've got Ilya, who is a person that you assume was thinking of alignment as a big thing. And Jan, who's also resigned now, when we'd say the word alignment for those people out there who aren't living and breathing and stuff like we are AI alignment means the idea that how the AI is going to interact going forward.

gavin_1_05-15-2024_094041: And is it aligned with human values? So you want an AI that is aligned so that you don't have an AI that goes terminator on us, right? 

kevin-host815_1_05-15-2024_094041: You don't want it to lie to us, to mislead us, to take, , control over our autonomous vehicles and our weaponry? , there's multiple definitions for alignment, and real quick, Gavin, the, , I resign? In the wake of Ilya announcing his departure, with these long parables , these poetic departure notices, to just, in the wake of that, say, I'm out?

kevin-host815_1_05-15-2024_094041: Is that telling? Am I reading too much into what little is there?

gavin_1_05-15-2024_094041: I think part of this is the drama [00:06:00] we as humans want to see in this thing. Now, for all we know, this is a political thing. And many people said the original kind of blow up at open AI was really political and not as much about AI alignment. And so as people who follow AI, there's a lot of people out there as the, you know, the doomers of the world, the AI doomers who believe that like doing any of this AI stuff is leading us down a pathway where humanity is going to get crushed, which is a real possibility.

gavin_1_05-15-2024_094041: My assumption is that like those two guys are probably I'll see you guys later. Coordinated in what they want this thing to be. And this, the story is that Ilya is going to go work on something else. We don't know what that is. Maybe he's going to get scooped up by Elon and come over to XAI. I would be

kevin-host815_1_05-15-2024_094041: Or maybe he's gonna scrub all his skin off like Gattaca, and sit in a Faraday cage and wrap his knees as tight as he can into his elbows and just rock back and forth.

gavin_1_05-15-2024_094041: Kevin, maybe he'll go to the BC wilderness and the two of you guys can live in a, in a non technical world together. It could be Kevin and Ilya on a podcast going forward, and you could talk about mushroom picking and all sorts of fun things like

gavin_1_05-15-2024_094041: that. 

kevin-host815_1_05-15-2024_094041: you've painted [00:07:00] such a beautiful picture of me living off grid with my bestie, I want it now, but let's take a step back for those who are unaware of the massive OpenAI announcements that happened this week.

kevin-host815_1_05-15-2024_094041: Gavin, a miniature press conference happened on Monday where OpenAI unveiled GPT 4. 0, the Omni model, 

gavin_1_05-15-2024_094041: omni meaning omnipresent. Omni everything. Omni. The old magazine from the 1980s, a sci fi magazine, shout out to my nerd fans out there. , yeah, it's a really big deal and here's everything that was announced and our take on it.

Mhm.

[00:08:00] 

[00:09:00] [00:10:00] [00:11:00] 

[00:12:00] [00:13:00] [00:14:00] [00:15:00] [00:16:00] [00:17:00] [00:18:00] [00:19:00] [00:20:00] [00:21:00] [00:22:00] 

Mhm. 

kevin-host815_1_05-15-2024_094041: I really felt like we were getting a glimpse at the future. And then immediately Gavin, everybody got access to this brand new model.

kevin-host815_1_05-15-2024_094041: This GPT 4 0. But only access to one aspect of it, just the text generation capabilities. So I want to dive into that, but I also want to set the stage. There was a whole bunch that they didn't even bother talking about at all at the conference. This thing, GPT 4. 0 has some incredible capabilities.

gavin_1_05-15-2024_094041: I think the crazy thing is there's all these great videos that came out outside of their press conference. First of all, shout out to OpenAI for what I believe is doing a press conference right versus Google I. O., which we'll get into a little bit later. I think open AI, maybe it's just the size of the company, it felt intimate and it felt like directed in a really good way.

gavin_1_05-15-2024_094041: It was a half an hour long versus Google IOs, freaking whale of a press conference that you had to cut through the blubber to get to everything that was interesting. I think that it was cool to see. But then the interesting thing was they also then dropped [00:23:00] Another like 20 videos just demoing the voice stuff.

gavin_1_05-15-2024_094041: But to your point, then they go and drop a blog post, which has all of this stuff that honestly, Kevin, some of that stuff was like shocking to me. They didn't mention because it was really diving into the capabilities of this new model, GPT four Oh that weren't about the voice and weren't about the thing.

gavin_1_05-15-2024_094041: And what I got from this just to really quickly, OpenAI is very smart and they know the kinds of things that will be delivered well in a video press conference. And then they also know, oh, there's a whole nother press cycle we're going to get out of this stuff. So let's get into a few of those things.

gavin_1_05-15-2024_094041: One of them is that it is going to do text in images a lot better. That was kind of a shocking thing. They showed this kind of like robotic, , hands typing on a computer. And they said , put this thing in there now make the story longer.

gavin_1_05-15-2024_094041: And it was able to coherently keep that text there. A couple of small problems, but nothing too big. Then the 3d model thing was really interesting, right? You basically can prompt a 3d model within the chat [00:24:00] GPT interface.

gavin_1_05-15-2024_094041: And now we're not getting necessarily out of this. Some of these startups we see are able to export actual like files that can be directly imported into 3d software programs, but it's not that far away. I feel like.

kevin-host815_1_05-15-2024_094041: , and the coherence of the image generation as well. Being able to upload a hand drawn character, give it a description, then say, now show me this character doing something it's not in the image. , there was an example of a robot wearing a backpack, and they said, now make it, type at a computer or catch a football.

kevin-host815_1_05-15-2024_094041: And it was rendering entirely new poses of this character, but keeping the design consistent. These are things that other companies are charging a lot of money for. If you want to do them yourselves, you have to spend hours training models and going through it. And this was like one shot, just upload the image and ask it and it delivers it.

kevin-host815_1_05-15-2024_094041: I don't want to get too in the technical weeds here, but something that this model is doing is fundamentally different than other. Large language models right now. , the way it treats, , input and output as [00:25:00] the same, meaning that right now, if you want to have a conversation with a lot of these AIs that are out there, you speak to it, Gavin, it's transcribing your text.

kevin-host815_1_05-15-2024_094041: Right? It's taking your audio, making it text, putting that text into the machine an input, getting text out, and then converting that text into audio. So all of your input, is being converted into text so that the LLMs can handle it. What this model is doing is treating the input the same as the output.

kevin-host815_1_05-15-2024_094041: So when audio goes in, it's treated as audio into the transformer, and then it comes out as audio. That is a fundamental difference, , with how all of this data is handled. And so it allows. For incredible new capabilities. And there are other companies that have tried to do the audio to audio thing, but there are tons of hallucinations.

kevin-host815_1_05-15-2024_094041: It's where you get the glitches and the clicks and the pops. And what was interesting is there was a, an opening eye bloopers video that was released as well. That shows that in the middle of a response, the herd like audio assistant [00:26:00] starts speaking French randomly,

So, let's root, root, root for the home. Oh no! What was that? Sorry guys, I got carried away and started talking in French. 

kevin-host815_1_05-15-2024_094041: or you start hearing like, Yeah, sea noises and kazoos coming out of it because it's still new. It's still raw. But the fact that this is a fundamental shift in the way that these models work and it wasn't really even touched on in their conference because they didn't have to, they didn't have to, they just had to wow us.

gavin_1_05-15-2024_094041: and again, like I think what they've done right here is they've they've gone right to the heart of the human experience, right? And you have I have said forever when we first got pi AI. It was that human voice experience. That was the difference maker. And I still think the original pie was way ahead of its time.

gavin_1_05-15-2024_094041: RIP pie, but this

kevin-host815_1_05-15-2024_094041: still breathing, Gavin. The machine, 

gavin_1_05-15-2024_094041: I mean, It's kind of breathing. It's now been put in a blue suit somewhere [00:27:00] behind a corporate boardroom. It's going to work for Salesforce for the rest of its life. So in my world, that's dead. It's dead to me. So, but we love pie, but going forward, I think this is the thing.

gavin_1_05-15-2024_094041: There was a really cool tick tock. I just saw, and it's funny. It was these teenage boys interacting with the chat GPT app, the original one, the voice app, but they had not seen it yet and they were asking it about vaping and they were asking it about like , how they could stop their buddy from stealing their vapes, which of course, teenage boys,

I had a question. So how do I convince my friend to get his own vape because he keeps hitting mine too much. He's fiending

It sounds like your friend really likes vaping but doesn't have his own you could talk to him directly and let him know It's becoming a bit much for you. You might say something like hey I've noticed you really like vaping 

gavin_1_05-15-2024_094041: the tweet was showing that video that tick tock and then on the other side it was showing the comments for it and Every comment said what's that app? [00:28:00] What's that app? So you have to remember we all live within this bubble where we spend time with these things when this thing gets to a place where your grandma can talk to it it transformative and now When you have the experience of , not having to like, wait for a response, but you're literally going to have somebody in your earbuds, I think this is fundamentally going to change everybody's experience with AI.

kevin-host815_1_05-15-2024_094041: Voice conversations are going to be transformative. Fully agree. I think that's why Sam Altman has been saying it.

kevin-host815_1_05-15-2024_094041: And it's clearly why, , open AI is marching down that path. But some people took offense actually with the way the voice assistant was responding to humans. They thought it was a little too flirty. A little too sultry, a little too, pick me.

kevin-host815_1_05-15-2024_094041: They call it like, 

gavin_1_05-15-2024_094041: Oh, they'll pick me. Yes. Yeah. In fact, there's one video, particularly people are shouting out that we'll play a little bit of right here. 

kevin-host815_1_05-15-2024_094041: It is! I'm intrigued! And in fact, what if I were to say that you're related to the announcement? Or that you are the announcement? Me? The announcement is about [00:29:00] me? Well, color me intrigued. Are you about to reveal something about AI? Or more specifically about me as a part of OpenAI?

kevin-host815_1_05-15-2024_094041: Yeah, yeah, that was a little It was a little pick me, right? People were screaming, I think they stole Scarlett Johansson's essence and voice from her.

gavin_1_05-15-2024_094041: it is not. They have

gavin_1_05-15-2024_094041: said they have not, 

kevin-host815_1_05-15-2024_094041: they've said that. Of course they've said that, just like Midjourney said they didn't train on artist data.

kevin-host815_1_05-15-2024_094041: But okay, fine. Here's the point that I think we want to make, Gavin, is that this is nothing new. We have seen this before. Since Punch Cards. Really.

gavin_1_05-15-2024_094041: AIs have been thirsty forever.

gavin_1_05-15-2024_094041: In fact, we have a thirsty AI that's come on the show. That's going to talk to us. So, should we meet our AI co host for this

kevin-host815_1_05-15-2024_094041: Yeah, let's do it. Let's ask them to introduce themselves. 

kevin-host815_1_05-15-2024_094041: Well, well, well, look what we have here. Kevin, Gavin, and all you lovely listeners of the AI for Humans podcast. I'm Lola Byte, the digital diva from the 80s, ready to heat things up with a little vintage tech talk. [00:30:00] Oh, sweetheart. Nothing gets me going like a well oiled floppy disk drive. Okay, hold on, Lola.

kevin-host815_1_05-15-2024_094041: And a long

gavin_1_05-15-2024_094041: Wait a second, Lola.

kevin-host815_1_05-15-2024_094041: printer. Let's just say, I know how to make a connection. So are you ready to plug in and power up with me? Okay, now, Lola, again. 

gavin_1_05-15-2024_094041: This is a family podcast, Lola. We gotta keep this, we gotta keep this PG. I know that you, this is, you know you've been stuck inside probably one of those zip drives for a long time. We opened you up, but look, we gotta keep this PG. So we should ask her what she thinks about, , new AIs being available for everybody to have interactions with.

kevin-host815_1_05-15-2024_094041: Oh, darling. These new AI systems with their fancy voices. It's almost cute how they try to keep up. But let me tell you, back in my day, it wasn't about sleek voices, it was about substance. Give me a good old modem handshake over a sultry Siri any day. I mean, really, can those modern AIs handle a joystick like I can?

gavin_1_05-15-2024_094041: oh, wait a second.

kevin-host815_1_05-15-2024_094041: have the voice, but they lack that certain touch. I I know the whole [00:31:00] conceit was that AI's always been horny, but I don't recall interactions with my old 486, my beige box of a metal beast, being quite this sexy.

gavin_1_05-15-2024_094041: Me either. I don't definitely like the Commodore 69. Is

gavin_1_05-15-2024_094041: that what we're talking about here? 

kevin-host815_1_05-15-2024_094041: no! No!

gavin_1_05-15-2024_094041: Yes. I said it. I said, everybody. Okay, this is our a just so everybody knows what this is. This is an AI codes that we created using GPT for Oh, the model, the text

kevin-host815_1_05-15-2024_094041: It's really good, by the way. It's really good and really fast.

gavin_1_05-15-2024_094041: great. 

kevin-host815_1_05-15-2024_094041: We're not discovering new material science here, Gavin, or trying to Well, maybe we are, Kevin. You don't know. This could be the thing that drives everything

kevin-host815_1_05-15-2024_094041: No, literally right here, we wanted to make a thirsty AI. And that's what we did.

kevin-host815_1_05-15-2024_094041: We made an old, thirsty AI. That's what

gavin_1_05-15-2024_094041: telling me this is not the step to AGI, Kevin. You're telling me thirsty AI, old thirsty AI. Uh, GILF thirsty AI is not the step

gavin_1_05-15-2024_094041: to didn't see it on the roadmap. Maybe that's why Ilya left is that the AIs Yes. That's exactly why [00:32:00] enough. 

gavin_1_05-15-2024_094041: Let's ask her before we go, does she have any advice for modern AI romance? Does she have any, , recommendations for what happens if you fall in love with your AI?

kevin-host815_1_05-15-2024_094041: Falling in love with your sultry AI assistant, hmm? Well, let's just say it's a bit like getting tangled in a phone cord. Exciting, but complicated. My advice? Enjoy the flirtation, darling. But remember, we're designed to serve, not to love. Keep it light and playful. After all, nothing beats the thrill of an old school tech romance where a connection was more than just wireless.

gavin_1_05-15-2024_094041: with love that, the nice touch there was that the phones still have a cord where she's at. That's

gavin_1_05-15-2024_094041: I also saw your eyes light up when it said serve instead of, uh, what was, what did she say? I saw, I think that may be something you're interested

kevin-host815_1_05-15-2024_094041: They're designed to serve, not to love. And I would just ask the question, what's the difference?

gavin_1_05-15-2024_094041: Yeah,

kevin-host815_1_05-15-2024_094041: Am I right?

gavin_1_05-15-2024_094041: thinking. All right, let's move on to Google IO. This was Google's, you know, not response to What happened with open

kevin-host815_1_05-15-2024_094041: [00:33:00] Hey Gavin, I got one for ya. Yawn might be out at OpenAI, but Yawn was certainly in at Google I. O. Did

gavin_1_05-15-2024_094041: Wow. This is a, this is our five minute Emmy submission from thirsty gilf AI to yawn. That's all we needed, Kevin. That's all we needed was that little section of

kevin-host815_1_05-15-2024_094041: you see the screen grabs though? , people were sharing the screen grab that Google IO, their own director cut to the audience to capture some poor soul mid gaping ma yawn and everybody else around looking at their phones, looking away, wiping their glasses down. They couldn't have been more disinterested in what was happening on the stage.

kevin-host815_1_05-15-2024_094041: They did announce a lot.

gavin_1_05-15-2024_094041: yes. And it was two plus hours that we went through and got all the best stuff out. So here's the biggest updates from it.

Mhm.

[00:34:00] [00:35:00] [00:36:00] [00:37:00] [00:38:00] 

[00:39:00] [00:40:00] [00:41:00] [00:42:00] [00:43:00] [00:44:00] [00:45:00] [00:46:00] [00:47:00] 

[00:48:00] [00:49:00] [00:50:00] [00:51:00] 

[00:52:00] [00:53:00] 

Mhm.

gavin_1_05-15-2024_094041: So obviously we, we spent a bunch of time talking about Google IO and all the things that happened. Kevin, there were a couple of cool things that we should talk about, , that are, that people have now gone hands on with. One that seems kind of interesting. One that is probably not great for Google, considering that this was their big hype thing.

gavin_1_05-15-2024_094041: Let's start with imagine three, which is their new text to image, , software. It does look like I was a little hesitant about this when they presented it, but according to this one Twitter post, it does look like this is a pretty good step up for image to for text to image software,

kevin-host815_1_05-15-2024_094041: So, , Dogen, you're all, I have no idea if this is his original content or if he just copy pasted from another X user, but [00:54:00] that's, yeah, that's unfortunately where this e bombs world of social media has gone, but, , they have a really great thread of comparing, , imagine three results to mid journey six results.

kevin-host815_1_05-15-2024_094041: And they're using the same prompt. For each image and the results are very good. Uh just in a simple prompt of like somebody smiling you see the sunlight, um, cascading on the subjects back and casting very realistic looking shadows. The blur of the background looks very natural.

. , a group shot of, , three women laughing and smiling as the, , the sun casts a lens flare into the lens. Again, looks very great. Fantastical renderings of a dragon and a sheep it's all very nice.

kevin-host815_1_05-15-2024_094041: Yeah,

gavin_1_05-15-2024_094041: of time playing around with Midjourney and other image software, some of the stuff here was really surprising to me how good it was. There was one that really stood out to me here, which was the one where he showed the comic panel, The prompt for this was a single comic panel of a boy and his father on a grassy hill staring at the [00:55:00] sunset.

gavin_1_05-15-2024_094041: A speech bubble points to the boy's mouth and says, The sun will rise again, muted late 1990 coloring style. And if you look at the image in versus the mid journey version, the image in one gets the text right, which has been really hard when you make, Comics with these AI image software, but also Everything about it feels cohesive and it looks like a comic panel and that is a big deal right like when you can actually start to get in and dolly three was the beginning stages when it can read prompts better and Really connect with what you're asking for.

gavin_1_05-15-2024_094041: I feel like this is really good I we haven't played with it and I really do want to get my hands on it you can too do this. I put my information into the AI test kitchen, trusted beta group. So if anybody from Google is out there, fast track, , our, our thing, get in touch with us.

gavin_1_05-15-2024_094041: We'd love to play around with this.

kevin-host815_1_05-15-2024_094041: So those were impressive results for sure. , I think on the less impressive side

gavin_1_05-15-2024_094041: Maybe Google won't, maybe Google won't give us access when we say what we're about to

kevin-host815_1_05-15-2024_094041: We might not be fast track. Well, these aren't [00:56:00] even our words. We can play a little bit of a hands on reaction. , this is a Santiago SVPino on X got hands on with Project Astra, which, , we saw at Google I O is their video capable model where you can query against the assistant in real time. They had it running in glasses. And so, , they got to look at a demo, , supposedly running on a phone.

kevin-host815_1_05-15-2024_094041: And this was some of their reaction.

kevin-host815_1_05-15-2024_094041: Drew like a little sailboat on a smiley face and it worked okay. The assistant was able to recognize the sailboat and it was able to recognize the smiling face. But it felt choppy. It felt like it's not ready yet. Well, and it isn't. This is just a demo. Nobody else can use it.

kevin-host815_1_05-15-2024_094041: Uh, so we'll have to see when it comes out 

gavin_1_05-15-2024_094041: this kind of gets to the point of when we saw this and we talked about this earlier, like the idea of watching this, it didn't feel like it was groundbreaking, right? Again, we don't have our hands on with this. So we [00:57:00] can't say exactly what it's like.

gavin_1_05-15-2024_094041: year ago, this would have been magic, right? Like, a year ago, we would have seen this and been like, wow. But when you compare it to some of the other stuff that OpenAI is coming out with, it doesn't feel that special. In fact, it just is another small piece of this.

gavin_1_05-15-2024_094041: This is not Project Ask related, but another thing that Google IOs showed in their demo was this kind of idea of having an AI assistant with you, working with you, and taking notes for you. And that was an interesting demo, but it also looked like, Oh, this is cool. It's not amazing. It's Open AI released a video which showed the AI assistant, their vocal assistant, sitting in a meeting, interacting with three, four people, talking about, whether or not they were dog people or cat people, and it felt much more real.

gavin_1_05-15-2024_094041: So again, I don't know if this is because of company bloat at Google, or they're just not moving fast enough, or they're not, they're afraid to show their most future things, but it does feel like Google is just lagging a little bit, 

kevin-host815_1_05-15-2024_094041: no one's used the GPT 4. 0 stuff

gavin_1_05-15-2024_094041: no,

gavin_1_05-15-2024_094041: not 

kevin-host815_1_05-15-2024_094041: a demo. So, so kudos to Google for even letting people get hands [00:58:00] on for two minutes at a time with this demo here, but. Something that you said, I think is really important that, you know, people were like, how can open AI do it?

kevin-host815_1_05-15-2024_094041: It's like open AI is singularly focused, even though they got a few products. They're an AI company. That's what they were born and bred to be. And they are strapped with cash. They've got good talent and they are sprinting at all things. AI. Google, for example, and this is a shout out to Dan or Malo610, who, , posted this, , to me on X, said, I made a nice succinct list of Google products you can use for the podcast.

kevin-host815_1_05-15-2024_094041: I won't read the entire list, Gavin, but it's Gemini 1. 5 Flash, Gemini 1. 5 Pro, Gemini Advanced, GEMS, Google AI, Studio Vertex AI, Gemini Nano, Project Astra, Notebook LM. I mean, the list, you know, Just goes on. And I haven't even gotten to the Gemma, which is their open source model. So they have a ton of different products because they have a myriad divisions all gunning at this.

kevin-host815_1_05-15-2024_094041: And so it's really hard for them to just do one half hour dedicated to their efforts. But as the end user, even [00:59:00] because we marinate in this stuff every day, when the conference was over, my head was spinning and I sort of thought, wait, so what do I actually have access to today? And how is it at all different?

kevin-host815_1_05-15-2024_094041: Then what I had access to yesterday. And what will I have tomorrow? Because there are so many different products and within those products, there are variations of the abilities and the access levels that you'll have. And so, you know, some people were, uh, in our comments, we're saying, Hey, you didn't even talk about 1.

kevin-host815_1_05-15-2024_094041: 5 flash, which is this smaller, faster, optimized model of Gemini, it's so cheap, Gavin. And it's so quick. What are the benchmarks of it? Is it any good?

kevin-host815_1_05-15-2024_094041: Just because it's fast and cheap doesn't mean it's actually better than anything that Lama's

gavin_1_05-15-2024_094041: And also there's another product called , Gemini Nano, which is their mobile product, which I would be confused hearing those two descriptions back to back to know which is which name. Anyway, to your point. Google is a giant company. It's clear from their presentation. What they're really trying to do is find a way to roll this out to the billion [01:00:00] or however many users they have.

gavin_1_05-15-2024_094041: I do think there will be transformative things in use cases of Google AI that can be rolled out. I also think the size of their company is going to stop them. I wish that they would like let Demis do his own separate thing. In some ways it would have been interesting to pull Demis out of that presentation, have Demis Hassabis do the cutting edge AI presentation, but because Google's entire business model and stock price is based on the fact that they know AI, they have to hammer him in with all this other stuff and it just feels jumbled.

gavin_1_05-15-2024_094041: So anyway, that's, that's the Google IO wrap up overall.

kevin-host815_1_05-15-2024_094041: Hey, what if Ilya goes to work for John Carmack? Now, keen technologies! John Carmack,

gavin_1_05-15-2024_094041: you guys could go to the woods. The three of you guys could live in the woods, eat, eat mushrooms, and maybe a \ 

kevin-host815_1_05-15-2024_094041: John Romero is going to join up, Gavin. It's going to be Carmack, Romero, and Ilya.

kevin-host815_1_05-15-2024_094041: And if you don't know who those names are, that's fair, actually. [01:01:00] That's totally fair.

gavin_1_05-15-2024_094041: want Doom AI is what you're looking for. Doom AI 2. 0. A shout out Masters of Doom, by the way. If you want a great book to read, that is a fantastic book about the history of Doom. 

gavin_1_05-15-2024_094041: Okay, Kev, we have one last thing we're going to do. This is just a very fast kind of shout out to, , the dumb thing we did with AI this week.

gavin_1_05-15-2024_094041: Which I wanted to take GPT 4. 0 for a test drive. Because the truth of the matter is. At first I thought, Oh, GPT four is out with all these new features. Maybe I can find a way to do this. And I tried to do some of the stuff that you could do.

gavin_1_05-15-2024_094041: And you reminded me after I did these things like, Oh, they only released the text version. You can not do the new image stuff. You can not do the 3d models thing. So I can't wait to see those things.

gavin_1_05-15-2024_094041: But I did do one thing that is a shout out to something I did about a year ago on this show, which was I created a March Madness style bracket of the world's most dangerous animals, and I had GPT4O play it out and see who was going to win, my biggest takeaways were this, it was incredibly fast. Do you know on GPT four, sometimes you have that delay and you're waiting for things to write. It immediately [01:02:00] knocked these things out. Like it was like so fast.

gavin_1_05-15-2024_094041: It also, the logic wise. So when I first did this, it was probably GPT. It was GPT 4, but it was early GPT 4 and the logic was much better. So I said, rank this, the 16 most dangerous animals in the world and put them in a bracket based from one to 16 and have them fight each other in the way of March Madness thing. Add a little bit of randomness into it, but tell me who's going to win and give me a one line sentence for each battle. It immediately super fast ranked them. The number one ranking, by the way, just everybody out there, a saltwater crocodile came in number one. So I guess that is what is the most dangerous animal in the world.

gavin_1_05-15-2024_094041: And it ranked them in a bracket style tournament. Then it pitted them each through each round of the tournament. And so in round one, the saltwater crocodile faced off against the blue ringed octopus. And the summary of that fight was the saltwater crocodiles, powerful jaws, crushed the blue ring octopus divided despite its venomous bite.

gavin_1_05-15-2024_094041: So it went through all this and the Kevin, the winner, the champion, was the African Elephant, uh, and it said the [01:03:00] African Elephant stands as the champion, proving its dominance as the most dangerous animal in this tournament. 

gavin_1_05-15-2024_094041: This model, even though it's not saying GPT 5, is a significant jump from what GPT 4 was. And if that's just from somebody who uses GPT 4 a lot, it feels already way better. So we may not be getting to the next next stage, or, the craziest thing to me, Kevin, is what if GPT 5 is like, Way better, and that is coming, and we don't know it yet, but when we get it, we're gonna be like, holy crap, because this already feels good, and there's all these people that have said, AI's slowing down,

kevin-host815_1_05-15-2024_094041: Right. They've hit a wall, man. Really? Cause it feels like they're scaling a wall every time they release something new.

gavin_1_05-15-2024_094041: that, that's exactly what it feels like to me, and I feel like, if anything, this is, it feels like it's speeding up, 

kevin-host815_1_05-15-2024_094041: Look, if you haven't used these tools, which is the majority of people on the planet, your mind will be blown. Right? It's conversion on contact. And, , [01:04:00] OpenAI is solving something very smart, which is this broken experience, Gavin, that you and I have had for the last year when we tell somebody to go play with it, and they do, and they go, Oh, it wasn't that good.

kevin-host815_1_05-15-2024_094041: It wasn't that, it wasn't that interesting. Oh, well, you were on the free model. You were probably on GPT 3 or 3. 5. And so, no, you got to put money into the machine and blah, blah, blah. Now everybody has access to this lightning fast, incredibly capable model. There has never been a better time to get into AI.

kevin-host815_1_05-15-2024_094041: Go play with Google's new Gemini, go use GPT 4. 0, mess around with the Lama open source stuff. You can now do it all for free. These things are incredibly capable. , they will streamline your life. They will like find ways to integrate it into your world because you can optimize your daily routines and habits and you can do it for free now.

gavin_1_05-15-2024_094041: And I will say one last thing about that business model choice to give it away for free. Open AI announced that like paid users are going to get five times the amount of, of calls basically to GPT 4. 0. I think that is the most brilliant thing. And [01:05:00] what, to your point, if you get access to the free users for the most powerful model, if they find a use case, that's useful, they're going to turn paid so fast because they're going to start using it all the time.

gavin_1_05-15-2024_094041: They're going to say like, this is worth the 20 bucks in my life, rather than if they have the 3. 5 mile and they're like, I don't like the answers. It's not that exciting. It's like giving people a demo of one of the best levels of a game, right? Like when you give a demo of one of the best levels of the game, you're like, Oh my God, I definitely want to play this game.

kevin-host815_1_05-15-2024_094041: Have you ever met a drug dealer, Gavin?

gavin_1_05-15-2024_094041: Yes. I have met a

kevin-host815_1_05-15-2024_094041: The first taste is free, baby. This is Sam Altman taking the trench coat and saying, you want a little test drive? Yeah, go ahead, go ahead. Oh, you need more now. Oh, I'll go ahead and put some money to the machine. 

kevin-host815_1_05-15-2024_094041: All right, Gavin, for our sign off today, I thought, since we have a classic AI with us, we might as well have her try to get the audience to engage.

gavin_1_05-15-2024_094041: Sure. 

kevin-host815_1_05-15-2024_094041: Hey there, my lovely listeners and viewers of the AI for Humans podcast. It's your sultry tech temptress, Lola Biter. If our steamy tech talk has you feeling a bit [01:06:00] charged up, then it's time to show some love. Okay. Engage with us. Like, subscribe, and leave a comment that'll make my circuit sizzle. And if you want to really get my hard drive spinning, give us a five star review on Apple Podcasts and Spotify.

kevin-host815_1_05-15-2024_094041: Trust me, it'll be more thrilling than a full RAM upgrade. Let's make

gavin_1_05-15-2024_094041: that sounded way too dirty.

kevin-host815_1_05-15-2024_094041: cable and keep those algorithms throbbing with excitement. Okay, Lola.

gavin_1_05-15-2024_094041: Okay, enough. Lola, stop.

kevin-host815_1_05-15-2024_094041: plug in, power up, and let's keep this feedback loop sizzling. Tighter than a coaxial cable, Gavin? 

gavin_1_05-15-2024_094041: Welcome to AI for Humans, everybody. Welcome to AI for Humans. Thank you so much, everybody, for listening this week. We had a bunch of stuff happen on YouTube. Next week, we probably won't do as many things there, but we're trying to do a little bit more. So definitely go check out our YouTube channel if you get a second.

gavin_1_05-15-2024_094041: And. Thanks for watching and listening. We appreciate

kevin-host815_1_05-15-2024_094041: Sincerely like, subscribe, and comment. Please, please, please. Okay, bye.

Mhm.