Waveform: The MKBHD Podcast - Hey ChatGPT, Summarize Google I/O

Episode Date: May 17, 2024

This was a week full of AI events! First, Marques gives a few thoughts on the new iPads since he missed last week and then Andrew and David bring him up to speed with all the weirdness that happened during Google I/O and the OpenAI event. Then we finish it all up with trivia. Enjoy! Links:  MKBHD iPad Impressions: https://bit.ly/3WzFFWk MacStories iPadOS: https://bit.ly/3V1G0Qq The Keyword: https://bit.ly/4blfFm5 OpenAI GPT-4o Announcements: https://bit.ly/3V3Sabv 9to5Google I/O 2024 Article: https://bit.ly/3V2rDLv Merch tweet: https://bit.ly/4bnhNcV Shop products mentioned: Apple iPad Air: https://geni.us/SsXTRLt Apple iPad Pro M4: https://geni.us/HXDlXo Shop the merch: https://shop.mkbhd.com Socials: Waveform: https://twitter.com/WVFRM Waveform: https://www.threads.net/@waveformpodcast Marques: https://www.threads.net/@mkbhd Andrew: https://www.threads.net/@andrew_manganelli David Imel: https://www.threads.net/@davidimel Adam: https://www.threads.net/@parmesanpapi17 Ellis: https://twitter.com/EllisRovin TikTok:  https://www.tiktok.com/@waveformpodcast Join the Discord: https://discord.gg/mkbhd Music by 20syl: https://bit.ly/2S53xlC Waveform is part of the Vox Media Podcast Network. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Support for this podcast comes from Huntress. Keeping your data safe is important. However, if you're a small business owner, then protecting the information of yourself, your company, and your workers is vital. In comes Huntress. Huntress is where fully managed cybersecurity meets human expertise.

They offer a revolutionary approach to managed security that isn't all about tech. It's about real people providing real defense. When threats arise or issues occur, their team of seasoned cyber experts is ready 24 hours a day, 365 days a year for support. Visit Huntress.com slash Vox to start a free trial or learn more. Support for this show is brought to you by Nissan Kicks. It's never too late to try new things, and it's never too late to reinvent yourself.

The all-new reimagined Nissan Kicks is the city-sized crossover vehicle that's been completely revamped for urban adventure. From the design and styling to the performance, all the way to features like the Bose Personal Plus sound system, The screen is pretty dope. Available feature, Bose is a is. That's the intro. Can you guys... Wow, this is really freaking thin. Yo, what is up people of the internet? Welcome back to the Waveform Podcast.

We're your hosts. I'm Marques. I'm Andrew. And I'm David. And we're all actually three AI, AIs. AI. And we're going to be doing this podcast by training it on all previous Waveform episodes. So whatever we say will just be a remix of things we've already said. And it'll be great. Thanks for tuning in. Just kidding. I wish we have two episodes to record today that would take a lot of no we're we got a bunch of AI to talk about today there's a lot of acronyms that I'm that I'm mixing up but IO happened open AI had an event a lot of AI things were talked about and we should

we should summarize it yeah because I didn't watch we can just use Gemini to summarize it oh yeah true yeah that would actually be perfect. But first, wait. I have the iPad here. It was iPad week last week. Oh, right. I was away, so I didn't get to fully do my review yet, but I have been using it. And, well, this is the 13th. Ugh.

I just hit the mic really hard. This is the 13th. It would have been harder with the old heavy one, though. True. Yeah. This is thinner and lighter than ever. This is the first time you guys are seeing it in person. Because you left immediately after it came in.

Basically, yeah. I took it out of the box and then packed up and flew to Columbia. So you guys didn't get to see it. Guys, look at this cool iPad I have. Okay, bye. But getting it back, I'm one of those people who has been using an M1 iPad Pro for years. However long it's been since it came out, and set this one up, mirrored it to the old one, so it literally has that old wallpaper. I don't know if you guys remember this wallpaper from M1.

It's like years old. And I've been using it the exact same way and doing all the exact same things. Which is? Which is emails, web browsing, YouTube, some Netflix, offline Netflix on the plane. Which is? Which is emails, web browsing, YouTube. Oh. Some Netflix, offline Netflix on the plane. Wait, what were you watching?

Drive to Survive. I'm sorry, I'm behind. Don't judge me. And yeah, Twitter, buying things on Amazon. Like I got some widgets going. It's just the same thing as before. Nice, powerful. And it's just thinner. And that's mainly the difference. Are you crushing a bunch of artistic tools with it oh of course yeah everything just shoved

into this ipad no it's that's here i'll let you guys hold it if you guys don't want to actually experience it i know david wanted to see it i still have a huge issue this is the most oh you weren't here last week when we talked about it right it is the most like non-uniform camera bump i feel like i've ever seen out of apple and it feels so unappalach not a single circle on there is the same diameter as a different one yeah i never thought i didn't think about it that much so they got rid of the ultra wide camera didn't really talk about it yeah they didn't even mention that at all whenever they get rid of something though they they never want to say it out loud and then so now we just have the regular the lidar and the flash and then whatever that tiny cutout is maybe that's a there is a microphone oh the mic other

one this is lidar i think yeah big one is lidar what is this big flash and then i don't even know what that is yeah none of them are the same size though correct it feels like the models that we get yeah there's usually at least one that's the same it could have at least gone with the lg velvet thing and made it like big medium small that was cool but instead it's like medium large small it's damn it just doesn't feel like an apple camera bump yeah it's un-aesthetic the screen does look dope though yeah okay so the actual biggest differences so if you've seen the impressions video already you know the new differences but there are some interesting yeah there's some interesting choices with this ipad one of them being that it is substantially

thinner and it's the thinnest apple device ever made with a screen that's my oh yeah because no the apple card or yeah they've made thinner objects before. Yeah, objects. But this is the thinnest device. Yeah, the thinnest device they've ever made. Dang. And it is 5.1 millimeters thin. Do you know how much thinner it is than the last one?

Probably about 2 millimeters thinner. 2 millimeters? I don't know. I'm American, so that means nothing to me. It's not a lot, but it is visually noticeable. What is that in Subway sandwiches? I'm American, so I can't do. It's one piece of lettuce.

One piece of lettuce. I have the actual answer. It's hilarious. It's a 20th of an inch thinner. A 20th of an inch. I have a... Do you know they keep comparing it to the iPad Nano Clip, I think it was? And they're saying it is thinner. It's not the Clip one. They're comparing it to the iPad Nano clip, I think it was, and they're saying it is thinner?

It's not the clip one. They're comparing it to the long one, the long Nano. The long Nano. I thought I saw a different one where they were comparing it to the clip. They're comparing it to the 7th Gen. The clip, I believe, was the 6th Gen. Yeah, the 7th Gen. I saw something the other day,

but are they including the camera bump in the uh on the ipad or are they only counting the sides i don't think they're including the camera bump yeah wait here ipod shuffle third gen oh it does have a clip this has a clip but are they counting the clip yeah no i don't think you well there's no screen on it but that's not what they said i don't think they counted the clip in all all the videos, it's just the aluminum housing up against it. That's the body of it. I just think it's an interesting choice because they decided to actually make it thinner again. Like, we got years and years of iPads that were totally thin enough.

Right. And then this year coming out and making a big deal about it being thinner was really fascinating to me because I thought we were done with that. Yeah, Johnny Ives gone. Johnny Ives gone, but here we are. Yeah. to me because I thought we were done with that. Johnny Ives gone. Johnny Ives gone, but here we are. To me, there's a lot of choices about this iPad that feel Tim Cook. I'm going to try to explain this in the review, but here's maybe the most Tim Cook choice of them all. There's a new Apple Pencil Pro, right?

The new Apple Pencil Pro is only compatible with the newest iPad Pro that you're holding or iPad Air. Yeah. Now, why is that true? Because they moved the webcam, the front-facing webcam, from the narrow side to the long side, which is where it belongs. But now that's right exactly where that Pencil is supposed to charge. So now there's both the webcam and the Apple Pencil.

What are you trying to do? Intuitive, baby. I'm trying to lower the brightness. Oh, it's on the right. You now there's both the webcam and the Apple Pencil. What are you trying to do? Intuitive, baby. I'm trying to learn the brightness. Oh, it's on the right. You have to pull from the right. Pull from the battery. Oh. Yeah.

Okay. So, okay. So the Apple Pencil now charges in the same exact spot, but there's also now a webcam there. So they had to arrange the magnets and the way it charges those components differently on that side so it doesn't work the same way as the old ones.

Why didn't they just move the Apple Pencil to charge on the other side of the iPad? Like the bottom? Yeah. Because of the folio. The Poco pens for the... Why just move it? Why not put one to the left and one to the right on the top

magnets actually it'd be pretty yeah that's good i think it's a sneaky great smart way of me making sure if you want to use any features of this new apple pencil you need to buy the newest ipad i think that's true so yeah that happened it's like one step more than at least the new Pixel coming out and being only available on Pixel, and then six months later being like, okay, it's actually available on all the Pixels again. At least this has somewhat of a hardware difference. Which is why it strikes me as very Tim Cook,

because supply chain guy would find a way to make sure nothing changes in the other ones, and now it just works with the new ones. It's sort of like how the M2 MacBook Pro is sold really, really badly because the M1 was so good. It's similar in that the iPad is so iPad. And what more iPad could you add to the iPad that they have to like have a better accessory that you have to have the new iPad to be able to use? Yeah. Because, you know,

it's thinner. Cool. But no one's going to buy it because it's thinner. Exactly. And the good display on the last generation was almost as good as this display. Yeah. So I can talk about the display in a second. The other Tim Cook thing is the new iPad Air is just a bunch of newer parts that used to be an iPad Pro. It's like it has the M2 now. It has the webcam on the side again and all these things that are like, oh, we now have a better iPad Air full of things that we already saw in iPads. So that's very efficient.

Very efficient. But yeah, the new display, it's just brighter. That's the main thing you'll notice. It is obviously a really interesting tech to make it happen. It's this tandem OLED display, which is not something I've ever seen in a shipping gadget before, which is cool. It's in cars.

It's in cars? Yeah, it's in car displays. Yeah, we went down this rabbit hole a little bit ago because Apple really wanted everyone to believe like this. We just invented the tandem OLED display. a tandem OLED display. But if you go on display makers' websites and display industry journals, the first concept

displays by LG were coming out in 2017-2018. They entered mass production with LG in 2019. Of a specific stacked OLED? Yeah. Referred to as tandem OLED. And it's the same thing and then

samsung just began as far as i could tell samsung just began their mass production for ipad this year um and i couldn't find any like leaks or like confirmation like this product uses an lg tandem oled display but But in LG's marketing materials on the industrial side of things, they're very clear, like we are shipping these things like hotcakes in our automotive departments. We also had a viewer, sorry, Zach reach out and say that Honor has a product from a few months ago

with a durable double layer OLED screen. Yeah, it's so in the industry that LG actually makes a durable double layer OLED screen. Yeah, it's it's so in the industry that LG actually makes a 17 inch tandem OLED folding touchscreen that I wouldn't be surprised if is in products like the the okay, I could not find any sources for this. But there's only one thing out there with a 17 inch folding touch OLED screen. Yeah. The Asus ZenBook Fold, whatever. And I'm like, is that a tandem OLED? That is interesting.

Anyway, this is all like Ellis's tinfoil hat conspiracy display industry nonsense. So take it all in the grave. Can that be a new segment of the podcast? Yeah, tinfoil hat segment. I will say the purpose of the tandem OLED display is to give you both the benefits of oled and super high brightness so in tablets we straight up have not had this combination of brightness and contrast ratios yet so i'll give them credit for that like the we've had oleds in that huge also super thin what is it called galaxy tab pro 14 or whatever just like gigantic tablet

we reviewed and it's awesome inside it's this super huge bright OLED and it feels like you're holding this magical piece of glass but it has like 500 600 nits max brightness because it's not as bright because I was okay to spray so here we are we have this new awesome thing that is better at max brightness better better outdoors, better on a plane with the window open. Right. That type of thing. Great. Yeah. It's the exact same brightness as the mini LED display

on the last iPad Pro 12.9 inch model. And now, deeper blacks. Yes. Sick. And they're deeper because they're off. Love that. Because they're pixels that are off. So infinite contrast ratio.

Yeah, so I've been using it. I don't know. I'm going to put my review together. I don't think it's going to be a standard review. I think it's more just me diving into my maybe tinfoil hat theories about some of the weird choices. Why is it thinner this year? Why is the Apple Pencil only compatible with this one this year? Why do they do all these things?

Why did they get rid of the ultra-wide camera? There's a bunch of interesting choices that they made. I think that is a better angle to go with for the review because what else are you going to say? Exactly. It's the same. I saw someone did a review that wasn't at all about the device really. It was like, what does iPadOS need for me to actually want to buy the M4 iPad Pro?

I would also be into that. Which is a better video. Yeah. And it's also such an evergreen that's been like the question on our minds for like four years now it's like this is a really powerful thing that's still an ipad so what do we want it to do okay sick that's the ipad i just wanted to throw one more thing in from last week? I did because last week I very dismissively was like stem splitter is boring and I

got drinks with a friend of mine who is you know working in the upper echelons of the music industry which I am not and he graciously informed me that I was pretty wrong and that stem splitter works way better than

previous attempts and it's integrated really well into logic and as like a utility it's actually like a really cool thing to have built into logic maybe not as a creative tool for making things necessarily but if you're a professional, he was like, this is very, very cool. So thank you. Unnamed friend. Yeah. But yeah, so sorry, I wrote it off. If you're working in music, maybe you should check this out and not listen to me. Yeah. Also, for those that maybe didn't watch the podcast last week or or forgot about it, it's basically that during the iPad event, they announced logic pro 2 for ipad yeah and also an update to logic i think it's called logic 11 now something like that but they're uh they're

introducing a stem splitter feature where you can add any music file and it will separate the tracks i got a live demo of it and this is one of those things where the demo was so good that i was like this is definitely tailored exactly for this feature. I don't know how well it will work. But in that demo, it blew my mind at how well it worked. And it was like, here, we gave it this song and we added, we did the stem splitter thing and it separated it out. And here's the drums and here's the vocals and here's the bass and here's the different instruments all split out through the power of ai and apple silicon and you play back just the vocals and it's like it sounds like i'm just

listening to a recording of just the vocals and yeah super cool tool to use and to me it always struck me like i don't use logic but it struck me as just like getting rid of some plugins that were doing that in the past it's just like yeah we just built it in now it's like when final cut does a better tracking tool you're like oh well that's better than the plugin i was using so great so i just i i would love to see it on other tracks also a question could you just remove one of the stems and then content id wouldn't like go off i think well content id is kind of like sophisticatedly trended trained on more than just like the exact set of sounds i think it would still go off i think it would just because of some of the examples i've seen in the past of like a car driving by and

like the radio is playing and they've gotten hit with stuff like i've heard people have hummed songs and it's gotten copyrighted really yeah that's so i think it's trained i think it's a great question but i have a feeling it goes on the it's a little more cautious when it comes to it and we'll probably still hit it okay that could be a short how many parts of a song can we delete before we get maybe we don't want to like destroy the channel testing to make a new channel to test that yeah cool jury is still out on uh chroma glow uh i haven't i haven't used like i don't have any projects here or i'm like doing stuff like that so i haven't used, like, I don't have any projects here where I'm, like, doing stuff like that, so I haven't used it.

Everyone I've talked to who is using it for work says the CPU usage is higher than they expected for, like, an Apple native thing, which might mean there actually is some AI stuff happening. Yeah, that would be my guess, because it's probably really AI-focused, and M4 is specifically tailored to that because they have all the neural cores in M4.

There's an email in Apple's inbox that's me being like, please explain this to me. Okay, yeah. And Chromaglow was the AI feature where they basically sampled a bunch of old, really famous instruments and tried to extract the vibes from it

so that you could add the vibes. Allegedly. Straight vibes. Vibe extractor. Vibe extractor. That's what they should have called it, dude. That's a way better name. Much more telling.

Well, speaking of an event that was lacking a lot of vibes, OpenAI had their event on Tuesday. So I didn't get to. Monday. Monday. It's Wednesday. Thank you. So I didn't get to watch it.

No. How did it go? Well. It was interesting. It was interesting. Yeah. I would say the high-level stuff, they announced a new version of GPT-4. There was this big thing where everyone was thinking they were going to unveil their own search engine the day before Google I.O.

That ended up not being true. That would have been wild. Sam also tweeted, that's not true, like the day before google io that ended up not being true wild um sam also tweeted that's not true like the day before which is funny but they did unveil something called gpt4o which i think is a terrible name okay uh also looks weird does the o stand for something yes it stands for omni i had to do a lot of digging to find this on the day of okay and it's because it's now a multimodal model so omni is supposed to mean everything it does everything omni yeah cool but it's kind of weird because it looks like 40 it's a lowercase it's a lowercase so that feels weird it can't like smell it's not omni yet well uh that's true they just didn't go over that part yet yeah so it is

natively multimodal which is good because the that part yet yeah so it is natively multi-modal which is good because the only other model i believe that was natively multi-modal was gemini it is much faster than gpt4 they say that it can respond in as fast as 232 milliseconds which is probably like the most ideal possible case during the event they were sort of talking to it naturally and talking back and forth and you know had that scarlett johansson voice when and when asked when they asked mira marty like is this scarlett johansson's voice she's like what are you talking about we what no did you see the tweet that was like we know know this isn't Scarlett Johansson's voice because none of her movies are on YouTube for it to rip from? Yeah.

And so they did like this demo where they were talking back and forth to it pretty naturally. And it works pretty well. It definitely feels a lot more like the way that a human would speak to you. Oh, a bedtime story about robots and love. I got you covered. Gather round, Barrett. They did have to like interrupt it multiple times because it has the same problem that all assistants have

where it doesn't just tell you your answer and move on and stop. It just tells you your answer and then keeps talking a bunch. I would say the like the interrupting thing was also something that was like a feature that they were describing, which I think at the core is pretty cool. Like in a general conversation. It's a good band-aid yeah people get interrupted or like sometimes you chime in on things people say we interrupt each other all the time so you know that's just part of natural i just wanted to say real quick disagree disagree and it did a pretty

good job at it except for sometimes i felt like when they were talking to it was a little too cautious of thinking it was getting interrupted where there were legit parts where it sounded like the audio was cutting out because they would ask it a question it would start answering and then i don't know if they like moved or like somebody in the audience said something but it would like cut out and then stop and they would have to ask it again or it would cut out then nothing would happen and it would start talking again. It felt like it was trying to be courteous of the person asking it and be too conversational to the detriment of itself, which was cool but not great.

This all comes back to the super high level original theory that I've said many times before about these AI chat models is they don't know what they're saying. They are saying things because they are trained on things and can say things that are similar that look like they should be correct. But once they have the sentence that they spit out, they have no way of knowing. They don't have a way of understanding if they just said a true fact or not. They don't have any way of ver understanding if they just said a true fact or, or not. They don't have any way of verifying. They don't know what they're saying.

So if you were conversational and you could actually look at what you're about to say and then trim it because you realize, Oh, this is like extra and we're in a conversation, but it doesn't have any of that ability. So I feel like the best, the biggest advantage to these, like,

okay, four Oh is cool and it's like faster and it's multimodal and all that. But the, like, okay, 4.0 is cool and it's, like, faster and it's multimodal and all that. But the advance that I'm looking forward to the most is an LLM that can actually fact-check itself and understand what it's saying and do something about that. So the funny thing is in Gemini, there's a button that you can press, like a Google button, that it gives you an output and then you can press the Google thing. And then you fact-check it. No, it fact checks itself. Does it? Yeah.

Like, cause it's, it goes straight from what the model thinks from the internet. Yeah. And there's like a double check button where it references more sources and then gives you a bunch of links. And then can verify if it was true or not. Well,

it's, I think it's going to still be on the person. I think it is too. Cause like when you have Gemini on the phone now, there's a Google it button. So if I go like, what is the Marques Brownlee's podcast called? And it says it's called the Verge cast.

The web show. It could just tell you that and then give you a Google button and then you hit Google and then the results show up and it's like, oh, he has been on the Verge cast, but it's called Waveform.

It's the human that still has to do the fact check yeah so if it could if it could fact check itself and be like oh my bad i remember i just told you that it was way for one thing it's actually the other thing that that's what i want but i don't think any of them do that yeah and i really want i don't know if they're ever gonna layer i mean maybe eventually it feels like it's just a layer on top of the LLM. The LLM is already doing a lot of work and it's super cool what it can do, but I just want that like, okay, now take your output, feed it back into another layer where I can go, okay, I told you the truth.

Yeah. Like how high can you get that confidence interval? Right. Yeah. Some other very weird things that it can do. There's like a vision version where you can talk to it while it uses your front facing camera. And it's supposed to be able to be able to understand your facial expressions as well as the intonation in your voice to better understand what you're asking in the context of what you're asking.

And they did a bunch of demos that were very like, what is my facial expression? Which is like, you know, it's kind of like pushing. Yeah. What do I look like the borders of it i feel like a lot of those examples were like it did a pretty good job of doing what they wanted but they always had to ask it specifically like if we were having a conversation mid-conversation i wouldn't be like david look at my facial expression and get the hint here like instead they're like it can see your facial expression while we're talking and the guy goes look at me and look at my facial expression what is it and it's like that's not

conversational i think that so we saw this issue with like the rabbit r1 as well right like all of the reviews are like what is this what is this and what is that like and perhaps what is this and nobody's gonna do that in the real world but the way that everyone thinks that you need to test it is like if this doesn't work then the whole thing falls apart so you have to test like the least common denominator at first so i think that's why they were showing it is they're proving it can understand your facial expressions but i still want to see more of a real world use case where my facial expression changing changes the output of the model yeah is that like it uses it as context or something that's what it

says but they never showed any demos it should and its whole point one of the big things they talked about which i actually think it did a pretty good job with was like inflection in the voice and so like through quicker responses interrupting an inflection that's supposed to make it feel more conversationally based and one thing i was thinking about was like sometimes it did a really good job but the problem with trying to replicate a human it just reminds me of like walking in video games like we have been doing that for so long and graphics have gotten so good and it's still so obvious when like a video game character is walking and it doesn't look anything like a human.

So like this thing will hit a couple little marks where it's like, oh, that kind of did sound like less of a robot, but more, and then it'll just do something that sounds exactly like an AI. And I think we're so far away from it being a true, like to really confuse someone.

I'm picturing it like you ask it for the weather and it sees that you're happy and it goes it's raining outside bring an umbrella and it says see someone who's like kind of sad and says what's the weather and it goes it's a little worse than normal it's totally fine like it tries to comfort you yeah i'll say one thing i thought that was impressive was like the way it fixed its mistakes sometimes or like the way it had i'll use an example so like normally when you ask an llm or whatever something and it's wrong you'll just go like my mistake or just like the most can responses of like my bad it's actually the same question it gets it wrong again yeah yeah so

there's a point in this where he asks it to look at his facial expression. And its response was, uh... Seems like I'm looking at a picture of a wooden surface. Oh, you know what? That was the thing I sent you before. Don't worry, I'm not actually a table. Um, okay, so take another look. Uh, that makes more sense.

Look again and tell me my facial expression where normally it would just be like my mistake and instead but instead it responded with like oh that makes more that makes more sense like i get it now and then answered so like it is semi using the context of what's going on and making up these little things that it doesn't need to but that is a little tiny step that does make it feel more conversation i feel like they didn't harp on that because every time it did, that was when it was screwing something up, which they probably didn't want to linger on.

But like that actually to me was the most impressive. That said, like, I feel like people are so used to them screwing things up that if you can screw things up in a more natural human way, then that's kind of impressive. It does feel more like the conversation and not just like ask you a question,

you respond kind of crap. I also think that they that they took like i don't think they took this from rabbit because they probably don't give a crap about rabbit at all i don't even think about something the r1 did that the humane ai pen didn't do was it would go like hmm i guess let me look that up for you and it saying words the filler time makes it so what's happening in the back end doesn't seem like it takes that long and it did that a lot every time they would ask it a question it would be like so they'd be like how tall is the empire state building whatever and big oh you're asking about the empire state building sure it's blah blah blah blah blah and it's like if you just said nothing and just had a waiting time that would feel

that tension there would be like ugh but because it's repeating it back to you you don't feel any delay but then so you feel less delay but it feels more AI like that's one that's that to me is like a giveaway if I'm talking to a human and I'm like bro how tall is the Empire State Building and he goes you're asking about

the Empire State Building yeah the Empire State Building is and I'm like why are you saying all this? Just say the number. Yeah. Oh, you're stalling. I see. I think if you, sorry. Well, yeah, that's more human.

There's a part in there that is doing what you're saying, but I think they did an even better job, which was anytime they asked a question that had multiple questions or like points of reference in the question, you could almost tell it was thinking of how to respond and then could respond respond while i was thinking of the next response it was going to tack on to that so this wasn't in the actual event but they did a bunch of test videos on their youtube channel right and one of them was a guy saying he says like also they all say hey chat gpt which i hate i would like a name or something like that's a long wake word um but all say hey chat gpt which i hate i would like a name or something like that's a long wake word um but he said hey chat gpt i'm about to become a father and i need some dad jokes

for like the future can i tell you some and you can tell me if it's funny and you're asking me i know i know jokes are funny bad example but you could tell the two responses it had cooked up ready to go which made it feel quicker so after he said that it responded with oh that's so exciting congratulations on becoming a new parent and you could tell that was one response and then the next response was like sure i'm happy to listen to a joke and tell you if it's funny so you could tell that while it was waiting to figure out response two it had response one loaded already yeah and i think that that is how they're able to claim like such faster models is that it seems like is is that they just use filler content like

they take the the important information that they actually have to parse through the the model and they crunch that while they take the very simple things and can fire that off to you rapidly with like a conversational that's way better filler though than just like hmm you want to know what the brooklyn bridge is like i can tell you what that is like it is blah blah blah i'm becoming a father even though i don't really want to hear that from an ai yeah it's a little weird i'm proud of you man thanks dad yeah yeah so yeah i mean i actually i thought it was uh the event in general was pretty good i'll give them a short yes nice huge points for that it was 26 minutes i think yeah it was way shorter they didn't try to just like keep it going

um they did a bunch of math with it which was kind of cool because that's a big thing that you know they've been trying to do like if you're just predicting then you're not actually doing reasoning right and i'm sure it's still just doing that but with higher accuracy it does seem like every example is like most like childlike task ever like yeah name this one thing or just like read all of my code and tell me exactly what it does there doesn't seem to be the in between i think that mostly comes down to poor examples that they always show but like it really feels like all of them i think you said like all the demos are like hey can you show this by looking like a total moron and asking it the most basic question ever?

I was thinking about this during the Google I.O. conference yesterday. And I think this is every AI demo that you do. It's like you make your engineers look like total idiots because they have to ask the most basic questions just because they have to show that the AI can do it. it they're like oh so like for example that they had it work it they had gpt4o work it through a math problem that was what is 1x plus 1 equals 4 it's like 3x plus 3x plus 1 equals 4 and it was like how do i solve this can you work that through me step by step but it was like well first what do we do when we want to figure out an exponent move all the exponents to one side and so it basically like just took it through the basic math and it's funny because google at io did a lot of the same stuff where it was like this could be useful for kids this could be useful

for kids which reminded me a lot of the the rabbit thing that we talked about yeah um it's not just that it's like because i've this has been on my mind like it seems like as we get deeper and deeper into this ai assistant nightmare the the use case examples are getting worse yeah like yeah like they're like not only they're making their own employees look kind of dumb sometimes but then they'll try to like balance it with a sort of like human example like the hey can you teach me some dad jokes like i know i say this before on the pot but it's like what a sad reality when you're like assuming you you i'm assuming you have a spouse you know you're about to welcome your first child that's who i guess like you want to impress with it's like but it's like what's the like oh no my spouse like isn't funny at all like like they they can't help

me like workshop these jokes i need this stupid computer or like i actually kind of hate them and would rather talk to them yeah it's like what are you talking about one of the gemini examples that they showed in the original gemini demo that they showed again yesterday in like a sizzle reel was write a cute social media post for baxter and it which is their dog and it was like baxter's having a great saturday hashtag bark i bark. I'm like, are you that dense? Are you really outsourcing that to an AI? I know! What is wrong with you?

I think these individual examples have to be bad because there's no one example you can give that will apply to everyone watching. So they're trying to give a Halo example that can apply. Individually example that applies to no one individually it applies to no one but the concepts of it are maybe applicable because that's when you see all the linkedin posts about like oh this changed my like when you see write a cute social

media post about my dog on one half of linkedin people are going it can do copywriting oh cool and on the other half it's people going it can write my school essay for me oh cool and on the other half it's people going it can write my school essay for me oh cool and on the other half there's people going it can write a letter for me like when i need a condolences letter they're all they're about i feel like you would outsource a condolences like people would bro i just i just got back from a tournament i have a teammate who is a ghost writer for executives at large corporations. So I'm not going to say their name or the corporations, but many of these presidents of schools and CEOs have to write letters and

things all the time to congratulate someone on something or to write back to an alumni or to convince them to donate or all these other things and they just don't have time to have all that bandwidth and so they hire a person to do it or there's someone going oh the ai can do that now yes that's super useful i like when i wrote to obama in 2012 and i got a letter back that was signed barack obama yeah i was like what yeah that's somebody's job right now and guess how many letters that person's gonna that's that's one thing but being like yo i'm sorry this person you like died like having the computer marquez was like no i have a teammate i thought he was gonna be like who just recently lost something and i

really didn't feel like writing this letter too but no that's true i mean it's all the examples are are very specific that i'm giving but it's like they're trying to give you like one ultra generic example that applies to no one so that everyone else's more specific examples that they wouldn't use on stage because it's too small of a group can kind of be tied into it in a way. So yes, you're going to get some really brutally generic, like write a caption for my dog. And they're like, okay, I just learned that it can write. I just learned that it knows the subjects and it can just think that i think that's what i guess you're right but

i mean i think it can be that and there still should be some someone making more creative examples because both of these events and we'll get into the google one later were just felt like there was no wow factor it felt so simple yeah and it made all the ai stuff feel really boring and tedious okay i got an example for you because we were talking earlier and i was telling you that i feel like after the last two events i'm actually more optimistic about the state of ai than i was before these events because everything was so mundane so because there was because you're more optimistic about humanity living with ai and yes yes exactly i'm more optimistic that it's not going to take us over

and like destroy everything because i feel like the trend that they're all moving towards is having the like broad example and then everyone can kind of have their own little slice of ai for their own personal use case so for example this got me thinking i read books like i'm sure many people i know i'm flexing um but i like doing book clubs but i don't know anyone that's reading my same books at the same time can i go online and find people that are doing this sure but then david and i have tried to have book clubs like three times when it's like when we were reading the ai book the the Google book, like two years ago. So it's like sometimes.

Exactly. That is my problem. Sorry. Also, no one else in my life is really like super into fantasy books like that. Can I find one person and make sure they're reading the same chapter that I am every week? Sure. That's annoying. I just want to talk to someone and be like, that would be cool, right?

And having an AI know exactly where I am in the book and being able to have a conversation about characters and things i was doing it the other day with gemini just to test it out because this thought came to me and i'm reading a book right now so i was like oh without spoiling anything tell me about blah blah and this character and what you think and they came back with like legitimate answers which was like pretty interesting are you not worried that it's really cool it's gonna mess up the spoil part and just be like oh great uh character a was so good and then like but it's really sad when they die 25 pages later i have to say that now but at some point and i think in the near future it can easily know where i am in the

book and know not to but you really want to have a book club with a computer i don't want to have a book club period i just sometimes want to like talk about characters and stuff with the computer with anyone i don't know man i hate to break it to you there's this thing called reddit and any discussion you want to have about a book is already on i'd rather they're all spoilers or spoiler free but like it's not synced up exactly with the page that i'm up to you know there's also and i have a i have a tangent example that's like when you're in an extremely specific case like when you have sometimes it's tech support or just like a product that you use and you're like digging through forums like i need to find someone with my exact issue yeah

with these exact symptoms and these exact bugs and like you can go to a forum or Reddit and like type it out and like wait for people. Or you can go, hey computer, like look at all of the people who have ever had any issues related to this and then let me have a conversation with you about those issues and maybe I can figure this one out. Because then you're sort of like bringing in hopefully the most relevant information

and instead of having to click, click, click through every single comment, you can sort of like talk to the person who knows all the comments. And then when new issues get populated throughout the universe and they don't get populated onto the Internet, the AIs will no longer be able to solve your problems. I agree. Like, I think Marquez is really right. Like, that is like kind of part of the dream of the AI, the AI future. What's so concerning is increasingly.

And I do also agree with you, Marquez. Like, they have to give these sort of weird broad examples to satisfy all these different camps but it does feel like increasingly there's this message being subtly projected at us at the events that's like you know what's so exhausting sympathy sympathy love uh being an emotional being that let's have the computer do it i think that's just again poor creative examples like they could think of so many better examples where this would actually be like useful and you know how the ipad mini specifically targets pilots yes and like i'm listening you don't really know about that except i'm sure that the pilot community is always

like super we love ipad minis yeah but if the whole event was about pilots you'd tune out i don't know look i i feel like i'm interested in how can a specific group of people use this in a specific way you know because like i can at least sympathize i can at least empathize well i guess empathize is not the right word but but I can understand like, oh yeah, that is being helpful to part of the human race. If you're a company like Apple, you need everyone to picture themselves as that part. I was just going to use Apple as an example

for not doing Apple Watch Ultra. Their examples were like scuba diver, extreme like hiker or trail runner. And like, yeah, and that's still sold to hundreds of thousands of people who will never do anything. Because it's aspirational marketing. Yeah, that's the pickup truck effect.

That's like, this thing is built up. Nike is that good too. The pickup truck effect, 100%. People are like, but what if I need it at some point? What if I want to help my friends move in four years? It's like driving over rocks. Like, build tough.

Who does that? Nobody's ever done that. You live in Brooklyn. Hey, man, I have a gravel driveway. There are some leaves on my street sometimes. It snowed once. I need the clearance. Yeah, that is very much the challenge.

The prepper mentality of America. We have to take a break. I think we do. Okay, I just need to finish the event real quick. Okay. Irony, super ironic. There's a Mac app coming for ChatGPT. Only a Mac app, which is hilarious

because Microsoft basically kind of owns OpenAI. Not really, but sort of. And they also, sorry, I'm going to butt in just because of that sentence. They opened the whole thing with like, we are trying, our goal is to bring chat GPT anywhere you want. Except unless you have a windows computer, I guess,

which I think is because Microsoft, like their whole next big move is copilot everywhere. Like there's literally a copilot key on all the new windows computers. Like is there move already? I think that basically like whatever open ai does with chat gpt is just going to be co-pilot with us it's it's a skin that's it's called co-pilot but it's basically just chat gpt yeah so but it is awkward that they have all this marketing that's like chat gpt is everywhere except for our biggest funder um yeah they said it's coming to Windows later this year,

which is going to be super awkward. It's basically going to be like the Google Assistant and Gemini thing, because there's going to be Copilot, and then there's going to be a ChatGPT app on Windows as well. They're the same product, basically. It's strange. A branding issue.

That was basically it. That's all I wanted to say. I thought it was funny. Well, we got we got another event to talk about but we'll we'll get to that after the break basically the same event with a different name and shocker eight times longer yeah yeah but uh since we are taking a break we should also take some time for trivia the lights work again are they different colors they go by our voice dude i right the the lights have a mind of their own honestly are they ai run they are ai run uh and in this case uh ai stands for marquez can you please stop sorry people in their cars are really mad sorry about that and the people watching the dish is shut anyway so after the break we're

gonna get into all of our good old google io discussion but i was reading the google blog as i do from time to time called the keyword keyword very silly um and i learned something even sillier than the name the keyword and that's we all like know what google i like the io stands for like input output yeah google on the keyword actually lists two alternate explanations for why they chose io and i will give you a point for each of those explanations that you can give me is each explanation just what the i and what the o is or is there like more to it it's there's more they're not acronyms there's more to it but not like that much they're they're but they're acronyms right

uh wait like io is the acronym like does it stand for something uh one of one of them yes i think i know one of them and then the other one is more gray than that you guys shouldn't have asked you shouldn't have asked it well i did that's your you're the trivia master you got to decide what questions it stands for i'm out which is exactly how i felt when i started yesterday well that's a good segue we'll take a quick ad break we'll be right back. I'm out. Support for this show comes from Klaviyo. You're building a business. Klaviyo helps you grow it.

Klaviyo's AI-powered marketing platform puts all your customer data, plus email, SMS, and analytics in one place. With Klaviyo, tinned fish phenom Fishwife delivers real-time, personalized experiences that keeps their customers hooked. They've grown 70 times revenue in just four years with Klaviyo. Now that's scale. Visit klaviyo.com to learn how brands like Fishwife build smarter digital relationships with Klaviyo. Exactly like nothing else. Hand-selected for their inherent craft, each hotel tells its own unique story through distinctive design and immersive experiences, from medieval falconry to volcanic wine tasting. Autograph Collection is part of the Marriott Bonvoy portfolio of over 30 hotel brands around the world. Find the unforgettable at AutographCollection.com. Breaking news coming in from Bet365, where every nail-biting overtime win, breakaway, pick six, three-point shot, underdog win, buzzer beater, shootout, walk-off,

and absolutely every play in between is amazing. From football to basketball and hockey to baseball, whatever the moment, it's never ordinary at Bet365. Must be 19 or older. Ontario only. Please play responsibly. If you or someone you know has concerns about gambling, visit connectsontario.ca. Welcome back, everybody. As you may have noticed, we just kind of alluded to the fact that Google I.O. was Tuesday. i o was tuesday this was uh arguably in my most humble opinion because i'm the humblest person on this podcast in this current moment nice correct no i'm anyway uh one of the most soulless

google i o's i have ever watched whoa um i was thinking about this this morning i was like remember when google i o was sergey brin jumping out of a helicopter with Google Glass on and landing in San Francisco? Live. I remember that. Live. And then we got Project Aura, which was the modular smartphone with all the bits and pieces. I mean, we got I.O. announcement. We got like Project Loon, which was like bringing internet to like random countries.

Are all these things that you said so far dead? Yeah. Yeah. It's still fun at IO though. Like I remember being like. What was last year's Starlet? Starline. Starline.

Project Starline. We at least got that. Yeah. I just remember being like in high school and being like, wow, Google is such a cool company. I wanted to work there so bad. I was like, everything they do is just a moonshot idea. It might not work out, but it doesn't matter because it's all funded by search, which is cool. It's like they have an infinite money machine that keeps churning, and because of that, they can just do whatever the heck they want.

And this year, it kind of felt like they were pivoting to be a b2b company i'm not gonna lie they talked about ai models constantly they talked about the price that are charging developers to access their ai models which is something microsoft would do which is something open ai would do but that is not something that google used to do at google io yeah io ios changed i was definitely this year felt like this year felt it felt like do you know the like part in every io where they talk about all their servers and tpus and all that and there's like that exact same graphic every single year first of all that graphic got used probably 45 times whoever made that is not getting paid enough for how much they use the likeness of it but it felt like that like end 25 minutes of every io where you're like all right all the

cool stuff happens like what are we getting out of there that felt like the entire event it was like the most low energy io i've ever seen i mean i've only been here covering it for like seven years but just all the things nothing they announced had this like real wow factor there was very few times where my twitter timeline was all like this is so cool they just announced this like nobody really had that one thing where it was real and we've had like last year visualize roots on maps was this really cool visual example we've had like the chain link fence that i reference all the time like yes there are things that did not come out that was io yeah that was io but it was cool and it had that like wow moment the crowd seemed out of it this year almost all of the announcers i felt just also felt low energy except for samir

he did a great job and felt as high energy as normal but like yeah i don't know and the whole event felt like it was dragging from the first terrible taylor swift joke they made in like the first one minute and then they proceeded to make a bunch of other bad taylor swift jokes that really felt like gemini wrote it but yeah um this might be a silly question because i didn't watch it when you said samir do you mean like from youtube no no um i'm forgetting is like yeah if the guy from android and what was a bummer was he was basically like new things coming to android and then just had to say the exact same gemini things just in like basically a mobile version samir samad he's the president of android ecosystem and he had his like same energy which just made all of the presenters around him feel low energy everyone

felt really low i don't know what that what was going on but it felt low energy i think a perfect way to wrap up kind of that we weren't the same people feeling this is ben at nine to five google posted a title an article this morning that said so google made a 10 minute recap and his article says google's 10 minute io recap is somehow just as tedious as the full event and io usually isn't tedious until the last like 20 to 30 minutes like It's usually like, cool, cool, cool. Wow, I didn't even think of the fact that you could do that with machine learning. Wow, you can get rid of a chain link fence. You can't.

But still, all of this stuff that genuinely blew my mind that I feel like we also used to see when every pixel would drop. There would always be one or two cool AI features that were like, wow, I'm so excited for this year. always be one or two cool ai features that were like wow this is i'm so excited for this year and there was almost not there was like a couple things that were like i'm glad gemini's doing that now i can name like three which we'll get into which we'll get into but everything else felt really corporate it felt b2b which was really weird it was surprising to me because they also made the distinct choice to separate the pixel 8a announcement out of io right so we had a separate pixel 8a happen before like a week two weeks before io yeah and then to me that was like oh ios stack this year we don't even have room for the pixel 8a not and so that's

why it's surprising and i a lot of people say on twitter that like oh it's just because there's not a lot of hardware stuff that's tomorrow but like they have done cool visually interesting things software wise before and with ai before that's not the reason it for so it was just not a good io this year but i want to do two cool things about it while we then will continue to um be mean about it for the rest of this episode um i'm sad we didn't go because google io swag is always fantastic and the swag they had this year looked great um i posted some photos of it so like uh i think they did a great design this year the tote looks awesome the crew neck looks awesome the water bottle sick i'm really sad we missed that i'm sad if that's the most exciting part

well no the most exciting part was mark rebbeillet opening as the DJ. He did a fantastic job. If anything, is that, again, as an American, units of measurement really confuse me. And so when they measured Gemini's contextual ability in... Tokens. No, in number of Cheesecake Factory menus worth of words. Oh, right. I was like, Gemini can hold 95 Cheesecake Factory menus worth of context at one time. Wait, that's it? That seems like a lot. Have you been to the Cheesecake Factory? Oh my god.

Is the menu really big? It's a book. Adam could have a book club that no one would join him with. I need to go to a restaurant that has three options. I don't want to read the Cheesecake menu for three weeks now. I'm on the 70th page. Yeah, okay. Besides mark's immaculate

entrance in which he had a for viewers that and listeners that don't know who mark is he's like he makes music on the spot sort of person and they had him use the music lm thing to try to like generate music and then play with it he's like an improv genius yeah he's an improv it was very tough uh because the crowd was tough the music lm didn't work that well and the beats that it was giving him were not that funky a lot of problems he wears a lot of robes they had custom google io marco v.a robes and he shot them out of a cannon um and then and then sundar came on stage and the energy went boom into the ground. Okay, so there were actually some interesting things and they actually did start off with a couple pretty interesting things. They started off with what I think was one of the most interesting things, which was an update to Google Photos, where now you can just contextually ask Google Photos to show you certain pictures and also ask questions about your photos.

So you can say, and I actually used Google Photos to find my license plate a million times last year. I've never done that. Now I memorize my license plate. But in Google Photos, you can now say, what's my license plate number again? And it'll bring up pictures of your license plate yeah which is cool you can say show me lucia's how lucia's swimming has progressed and it'll bring up a bunch of photos and videos you took of your daughter swimming so it can understand the context of okay this is what you're asking and then map that to what it tagged the photos

as of being and i think that's actually really sick i do think the license plate example was perfect because they were like normally you would just search license plate and now every license plate you get every car you've ever taken a photo of shows up and then you say my license plate number it's like oh this is a car that i see pretty often so it must be your license plate let me find one picture that has it and only show you that one picture so you don't have to scroll through a bunch yeah that's cool yeah i like this and i think because the results are specifically pulling from your photos it can avoid hallucinating because it's going to give you the exact source it also because yeah it just has to show you an

image it's not generating something because before like source. It also, because, yeah, it just has to show you an image. It's not generating something. Because before, like, so if I wanted to find my license plate, like, I had this actually, for example, I was in the back of a car and I was going to go to the airport. And they're like, I just need your passport number and your signature. And I was like, okay, here's my signature. What is it? I don't have my, it's in the trunk. But I just pulled up Google Photos and I just searched passport and I got the latest photo that I took of my passport and I just got it from there. Yeah.

Instead, theoretically, I would just ask Google Photos, what's my passport number? Yeah. And it would give me my passport and I just got it from there. Instead, theoretically I would just ask Google Photos what's my passport number? And it would give me my passport number. And as long as I also can see that it's referencing an image of my passport and not some other random photo I have of a passport I think I'm good. It doesn't give you a text output

it just shows you the photo. The one picture. So I think that actually solves the issue. I think you could probably ask Gemini at some point what's my passport number and it would pull it up and then it would probably potentially reference the photo but right now this google photos update it's just an easier way to like ask google photos to show you specific pictures it kind of feels like which i like more than generative ai no I agree it feels like the magic of what Google search has been where everyone makes the joke of like what's that song that goes ba ba ba ba ba it's like you can ask it some pretty crazy questions and now in Google photos you can

just be way more specific and it can find things easier for you yeah and Google photo search has always been one of my favorite features ever just like being able to search through different things and it can tell kind of what they are and create albums or look at places that you've been to and now being able to go a little more into that where maybe i'm like what was that trail i hiked in glacier national park in 2019 and like if i took a picture of the trail sign it'll probably come up as like cracker lake trail and then that's awesome and i don't have to just search through every single thing that was in montana yeah i often will think oh yeah i went on that trip i think it was in 2019 so i'll search like mountains

2019 and i have to sort through all the mountains that i took a picture of in 2019 until i find it maybe it's not that year so now i can ask it like, show me the mountains when I was in Italy the last time. And, yeah. It can still do sort of those things right now, but it's just a lot more contextual, which is beneficial. It's helping because a lot of people didn't even know. You could search Google Photos for things, and it would find them very specifically. So it's just making this prompt engineering less of a skill. You just type what you're thinking, and it can find it.

Which is why I think they specifically used the the prompt what's my license plate number again because it sounds more human like the way that i would probably prompt that is my license plate and it would bring it up you know whereas a normal person says what's my license plate number again because i think the ultimate goal is like be able to have natural speech computing with these computers with every prompt yeah i think our generation grew up learning how to speak google which is keyword it's a total language you have to know how to google and you're basically just picking out keywords we've talked about this like prompt engineering is the skill that we all learned yeah and now it's it wants the old people and the young people to be able to do it where

young people just go like write me some code and it just does it you just don't have to know how to do it right there's a lot of other stuff so we're just going to go through it a little bit randomly i kind of went linearly through the keynote so if it jumps around a little bit that's google's fault no one remembers exactly how the keynote went so yeah everyone probably hallucinate that uh okay google search generative experience which was a google labs feature that you could opt into for the last year since google io 2023 which i have been opted into for a while and has annoyed the heck out of me i've been using it as well yeah i had an example recently where the ai thing gave me one answer and then the top suggested Google thing told me a different answer.

Yeah. Nice. That happened to be great. So it's gone great. Yeah. So that was an opt-in feature that I forgot was opt-in because I would have turned it off a long time ago if I remembered that. It is now rolling out for everybody.

And it also now generates a bunch of extra information and tiles and pages for you and in my opinion it was a very bad look because basically the entire visible screen was all generative ai and i didn't see any links yeah anywhere yeah which is bad yeah it was basically like creating a bunch of almost like windows 8 tiles of like all potentially different things you might want whether it's the sale link for something with prices or the reviews of a restaurant just like all in these different tiles and you didn't see a single link on the page yeah i also just want to like paint mark as the picture of when they announced it the way they announced it is you know the huge screen on screen on the I.O. stage. Yeah.

It was all white and had the Google search bar, but it didn't say the word Google above it. And all of us kept looking at each other like, are they going to rename search? Are they going to change? Or is this going to say like, I don't think any of us thought it would actually. But we were all pretty sure it was going to say Google powered by Gemini. Yeah. And I think part of that hype is why none of this stuff felt that cool because we're all like they're gonna do something big they're gonna do something big they're gonna rename it they're gonna change how this looks and then they just didn't yeah um i

think they're trying to completely reframe in your mind what it means to search something now you they don't want you to have to search something and then find the information yourself and actually what they kept saying over and over again was let google do the googling for you which was super weird it's like so like never leave google it's yeah it was basic because i think that what they were afraid of is a lot of people use chat gpt as a search engine even though it's not a search engine i've heard crazier things yeah you know how many people use tiktok as a search engine even though it's not a search engine i've heard crazier things yeah you know how many people use tiktok as a search engine a lot of people more than you would i mean it's like using i use youtube as a search engine because it's the second largest search

engine in the world yeah i like watching videos of people answering my questions and doing things specifically i like people dancing to my questions this This is how you change a tiger. Forget Quorum. People dances. Yeah. Yeah. That feels like a big part of it is Google has historically been kicking people into links and then they leave Google and they go find their answer somewhere else.

Yeah. But if Google can just like siphon all the content out and service the thing that you were looking for in the first place, then Google helped you, not the site. Yeah. And then that site never gets any credit, even though the site is the one that had the information which i just want to point to the fact that like not that long ago like two to three years ago google got in serious trouble for like scraping some information and putting it on the sidebar of google just saying like these sites say this this this and everybody freaked out and then google got in trouble for it

and now they're just completely Siphoning away sites altogether to the point where there is now a button called web So you know how when you google something it'll be like images videos news There is now going to be a button called web Where you press the web button and it actually shows you the links Wait, it wasn't a scroll away before uh i you probably can keep scrolling but there's so many things there but like do you know how they're yeah they're shopping maps photos now there's one that's web so it's only web links and i hate to say i'm excited for this because google search has become such a pain in the ass

that i like just want to look at different links it basically gets rid of everything that google has added in the last five years it's just it's like old google it's it's the opt-out of the reddit redesign for google yeah yeah uh that's hilarious that i find that hilarious somebody tweeted uh that that's the funniest april fool's joke that google has done um yeah it yeah it's just a weird thing um neilai from the verge refers to this whole redesign as google day zero which is basically like what happens when a search is just a generated like everything is generated and there is just zero incentive to go to websites like we've talked about this forever because we always were like this is

eventually going to become the thing and currently right now because they haven't like fully rolled out this thing it'll give the little generative ai preview but you still have links but all most of the examples that they showed at io were like the full page was generated and once you hit a point where the full page is generated that's when you start to hit the existential threat of like how our website's gonna make money anymore yeah this is again why uh i think this this lm stuff needs fact checking built in uh i just googled it real quick. Google's mission statement is to organize the world's information and make it universally accessible and useful.

And if you think about it, if you're Google, for years and years and years, you've been collecting and organizing all of these links. And then people go to you and they say, I want to know where this thing is on the internet and then you show them the list and then they leave. And you've been doing that for so, so long that you have all this information. If you're Google, probably what seems like a natural graduation is, all right, the internet, it's here, it is what it is, it's this giant thing. Now what we can do is learn all of it, summarize it all for ourselves, and now when you ask me a question, I can just reference it

and hand you exactly the information you need, and you never even have to go through that mess of links ever again. People will just, this is a crazy statement, kids will not really learn the skill of navigating through search results. Yeah, not at all.

That's another thing that they won't really that's also scary because that makes me think that kids are just going to believe the first thing they see which is made by a generative ai and that's why you need it to be fact checkable it needs to be verifying itself because yeah the skill still of like seeing the google it button and going yeah i'm gonna verify this one and like looking through the results that's still a skill that we have. But if you never have that skill and you just go to Google the thing

and it just surfaces the thing, you just believe it, that could be pretty dangerous. Here's a question for everyone, just on topic of this. When's the last time you went to the second page of a Google search? All the time.

I do it all the time. It's pretty, you know, you're in for one. I do. Yeah, it's like 50% of the time. Yeah. 50% of the time? 50 of the time i wouldn't say i wouldn't say for me but when i'm doing research i go through like five pages i got a long tail i got a lot of stuff on the first page but then every once in a while i get something on

the like 36th page that's it's wild that there's probably people who don't realize that there's like the 10 and then like it keeps going after that yeah it keeps going wait quick question marquez with what you were saying isn't that the best custom well i guess user experience customer ui in theory it is if it's correct yeah and who's gonna write content this is if you have a magic box that tells you all the answers nobody can make the content that is scraping from this is the we've been talking about this for months i know that's the problem that's the scraping from. We've been talking about this for months. I know. That's the problem. That's the fun part. Has nobody thought about this?

Does the incentive to create content disappear completely? When you don't get paid. If people stop going to websites and they don't make any ad revenue, then yeah. Then things go under. Is the fastest way also the best way for consumers? Like, yeah, you'll get the information faster because you don't have to click through websites. Sometimes. I think it's also a what about the journey i think there's also a tale to this where like at first the fastest way is best but then when it gets to a more in-depth

like think of someone who's about to buy something yeah exactly like if you're i'm about to make a big purchase i'm gonna buy a laptop real quick if i just google like what are some good laptops for a thousand dollars yeah then i just set my frame for the first page but then on the second or third weekend where i'm like i'm about to spend the money now i'm gonna really dive in no there it's like something i google almost every week chicken drumsticks oven temp time don't need don't need more than 15 seconds for that one you know what i mean but yeah like if i if i want if i wanted to go on a little little internet journey i think there's times though when also you don't have time for the journey if that makes sense like chicken drumsticks i'm hungry i think about the time i was out and claire calls me it's like a pipe burst in the

downstairs and we need something what do we do with it and i just ran to lowe's and i was like what's the best thing and instead i'm sitting there where like i wish it could just be like is it a small leak is it this and i could just it'll give me that temporary fix right there and i don't have to go through 20 pages inside of lowe's while my basement's flooding yeah yeah i think there's some fast things where that is the ideal scenario but yes the journey sometimes is fun because you learn about things on the journey that turns into actual usable information in your own brain yeah that we get to use right imagine that imagine that okay imagine experiences we gotta move on because there's so much there's so much so much um okay

gemini 1.5 pro which you and i have been beta testing for what feels like months yeah at this point they doubled the context window to 2 million tokens uh and now they're just spouting millions of cheesecake factory menus yeah they're just flexing on every single other company that they have the most tokens which we yeah well i still don't understand tokens tokens at all they're vbucks a word is like a token it's like tokenization of a word so you can map it to other and they just cost money transformers transformers transformers because people make fun of me for saying that a lot um power which costs money they're called tokens because it's like it's the smallest you and this is like the dumbest possible way of explaining it but it's like it's the smallest you can break down a piece of information

in a data set to then have it be evaluated across the parameters. And mapped to every other token. Exactly. So like in a sentence, like you would break down each word into a token and then analyze those words as independent variables. Yeah, it's tokenization because you're like splitting it up. In an image, like a pixel could be a token or a cluster of pixels could be a token. Okay, so then quick question. When they say 2 million tokens, do they mean that i can like do a two million word word yes

okay got it oh so it's per individual query it can take up to two million tokens yes okay that's the context so the context window is basically like how much information can i throw at this because theoretically in these models the more information you give it the more the more accurate they can be okay okay remember the dolly prompts that were like give me a man in an astronaut suit with a red handkerchief around his neck it was like be more blah blah blah you can just keep going okay right yeah cool okay um now they are also embedding gemini into a lot of google workspace stuff so you can have Gemini as like an additional person in your meeting that's like taking notes for you

and you can interact with during Google Meet meetings. They should call it GoPilot. Why? Because it's Google. It's like CoPilot, but Google. Oh, come on. Was it that bad? No, it's not as good as...

That joke was that bad? I'm picturing I'm killed by Google website in like three months sorry that just reminded me what did Mark call music goops Google loops gloops he called it gloops at one point

which they should use they introduced a new model called Gemini 1.5 Flash which is a lighter weight faster cheaper model that handles lighter weight, faster, cheaper model that handles lighter weight queries. Hooray! Microsoft is so scared.

We got Project... Okay, so Project Astra is what I think is basically their answer to the human and rabbit thing, except it's better because we always knew it would be. The demo they showed was basically on a thing except it's better because we always knew it would be the demo they showed the demo they

showed was basically on a pixel it has a live feed of your surroundings so on humane or rabbit you have to take a photo and then it analyzes the photo and talks about it on this one it was basically a real-time image intake where it was taking in a video feed with this person walking around with their pixel and they could just be And they were just kind of like, what does this code do? And it'd be like, oh, it does this, this, this. Okay, cool. Where can I buy this teddy bear? Oh, the teddy bear can be bought on Amazon for $14.99.

Cool, cool, cool. Over here. And then they did this casual thing where they switched to these smart glasses that had cameras on them. Which is also strange because they were like where did i leave my glasses and it remembered where but they never showed it in that initial feed so are they remembering it previously so here was the weird thing yeah they said like where did i leave my glasses i was like it's on the side table it only knew that because the person was walking around with their pixel camera open like for five minutes and it happened

to see it in the corner while they were walking around obviously in the camera open like for five minutes and it happened to see it in the corner while they were walking around obviously in the real world i think this was basically the same thing where it's like in the far off future if you had a humane ai pin that was constantly taking in all of your video feed information all the time it would always know where you left all your stuff because it was constantly watching which nobody wants um so that's the convenience though think of the convenience yes storage yeah remembering everything all of that google just put it on their servers yeah so it on youtube so i think this is just a demo to show that yeah they they can do what human and rabbit are doing but way faster and way better and it's a video feed instead of a phone

and because it's the live video feed they also did this thing where it's like you could draw in the video feed and be like what is this thing and like an arrow to it so if like for some reason you can't just describe the thing in the viewfinder as well you can it's basically circle to search through live multimodal yeah like which is something the open ai was basically demoing too on the day um because you could point your you could point your phone at things and it would help you through math problems as you were solving it yeah so it was cool uh they didn't talk about you know if it's ever going to come to anything ever they just demoed it right they said it was a project kind

of like those like translation glasses that just never became a thing and i think they were trying to make like a nod to those by saying like yeah we've got we're working on stuff but they're probably never going to release anything it kind of also made me feel like because this was just like another blip on the radar during all of io made me kind of feel like the humane pin and the rabbit pin needed to make the hardware versions because everyone just like kind of moved right past this even though it's doing the exact same things better than both of those but since they're not a shiny product uh-huh everyone's like cool it's just something in i.o they were basically like yeah you can do this on a phone moving on better yeah i don't think about you

child's play yeah yeah yeah all right so that was already a lot um there's a lot more google i.o stuff uh either you're welcome or i'm sorry or you're not reading all that or not listening to all that. But we got to do trivia because that'll help your brain break up this episode. Hey, what is this? Is it some purpose? It is. Yeah.

I tried to use Google's music effects to recreate our normal trivia music um it got none of the things i put in my prompt in here like not a single one oh i guess i asked for drums terrible sounds like is that drums or snapping i asked for six seconds of interstitial music intended to transition from the main segment of a podcast to the trivia segment that happens before an ad break. I kind of like it anyway. The track should feature a Hammond organ and drums and be bright, punchy, and fun. That's not wrong. I would say bright, punchy, and fun.

Yeah, but where's the organ? Where's the six seconds part? Well, let's chill out a little bit. Give it a chance. This is Google. Small startup company. I was wondering. He told me earlier, like, let me handle the trivia music.

And I got something cooked up. And I was like, what is he going to do? And now we know. All right. Second trivia question. So a bunch of people were talking about how the open AI voice sounded like Scarlett Johansson in the movie Her, like David mentioned. What was the name of the AI that she voiced in that movie? I don't watch movies. you guys are gonna get some points

yeah it's been a while since I've seen this film so much for that I've never even heard of that movie what? you're serious right now? go home we can't tell Andrew

anything about this movie. We have to get through the trivia answers and not spoil a single thing, and then, Andrew, you need to go home and you need to watch her. No. Support for the show today comes from NetSuite. Anxious about where the economy is headed? You're not alone. If you ask nine experts, you're likely to get 10 different answers.

So unless you're a fortune teller and it's perfectly okay that you're not, nobody can say for certain. So that makes it tricky to future-proof your business in times like these. That's why over 38,000 businesses are already setting their future plans with NetSuite by Oracle. This top-rated cloud ERP brings accounting, financial management, inventory, HR, and more onto one unified platform, letting you streamline operations and cut down on costs. With NetSuite's real-time insights and forecasting tools, you're not just managing your business, you're anticipating its next move. You can close the books in days, not weeks, and keep your focus forward on what's coming next. Plus, NetSuite has compiled insights

about how AI and machine learning may affect your business and how to best seize this new opportunity. So you can download the CFO's Guide to AI and Machine Learning at netsuite.com slash waveform. The guide is free to you at netsuite.com slash waveform. netsuite.com slash waveform. You've always wanted to be part of something bigger than yourself. You live for experience and lead by example. You want the most out of life and realize what you're looking for is already in you. And realize what you're looking for is already in you.

This is for you. The Canadian Armed Forces. A message from the Government of Canada. I am so dreading groceries this week. Why? You can skip it. Oh, what? Just like that? Just like that. How about dinner with my third cousin? Skip it. Prince Fluffy's favorite treats? Skippable. Oh, what? Just like that? Just like that. How about dinner with my third cousin? Skip it.

Prince Fluffy's favorite treats? Skippable. Midnight snacks? Skip. My neighbor's nightly saxophone practices? Uh, nope. You're on your own there. Could have skipped it.

Should have skipped it. Skip to the good part and get groceries, meals, and more delivered right to your door on Skip. groceries, meals, and more delivered right to your door on Skip. Okay, Notebook LM, which used to be called Project Tailwind, which Adam was particularly excited about, it's now more focused towards education, but effectively it can take in information from your Google Drive, like photos and documents and videos, and basically create a model of these AIs that can help you

with specific things related to your specific information. And the way that they showed it was basically a live podcast where they had these two virtual tutors. They were kind of like both talking separately being like, okay, Jimmy, so we're going to be talking about gravity today. So you know how something's 1.9.8 meters per second? And then the other AI would be like,

not only is it that, but if you dropped an apple, it would drop at the same speed as a rock. And then you can call in almost, and then you become the new member of this conversation, ask it questions, and they'll respond to you. Interesting. It was some virtual tutors, and they were very realistic, very similarly to OpenAI's 4 tutors and they were very realistic, very similarly

to OpenAI's 4.0 model that felt very realistic. I felt like this week is the week where all of the companies were like, we are no longer robotic. We can have humanoid-like voices that talk in humanoid-like ways. Great. So that was interesting. Okay. I would actually love to play with that to see if it's any good. It was kind of cool that there was like two virtual AIs that were sort of like talking to you at the same time, but also interacting with each other. Didn't it also just like pause and not finish the question? I think that he interrupted it. Oh, he interrupted it. Because he was basically saying, I'm helping Jimmy with his math homework.

Hey, Gemini 1, like, what do you think about this? And be like, like wow great question and then it just like paused and didn't finish answering the question yes okay probably just a bad uh demo on on stage yeah um okay they also introduced i thought it was called imogen but it's imagine imagine three which i thought it was imogen because image generation, but it's Imagine, which is more like Imagine. It's like a triple entendre. It probably still is that, but just a funny, better way of saying it. Yeah.

Basically, the third generation Dolly-esque model with better photo creation. Yay, cool. Music AI Soundbox, which they had Marc Rebillet at the beginning used the music ai soundbox to create these beats and they also had like a childish gambino ad where he was talking about it i think uh he was talking about video stuff later wasn't he because he was talking about veil yeah he's talking about veil because he was put up there as a uh why can't I think of the word? Film. Like film.

Like he was doing movies, not music, which I thought was funny. Got it, got it. Okay, yeah. So the music generation was, you know, better. They basically were just like updating all these small AI things that they were like, I can do this too. And better than last time. Yes, which is where we go with vo or yeah i think it's vo um

vo vo i don't know i don't know it probably is vo i'm just saying that because it's already a word in spanish so oh what does it mean in spanish like i see basically i see like from 7-11 like to see like i see like yo veo oh okay well that would make sense yeah yeah it can create 1080p video from text image and video prompts uh you can further edit those videos with additional prompts they had testing extending the scenes which was that was the coolest part of it i think which i think was like a direct shot at uh run runway runaway because that can only do like 30 second video clips right now, I think, possibly even one minute. Well, I thought it was cool because it wasn't just that it extends that is I think it was you could put a clip in and say, make this longer. Right. And then it would

make the clip longer by just like it basically is like content aware fill where you need a photo that's bigger. Yeah. But it does it on a video i think that's awesome there are so many times where you're like found footage this doesn't really roll the beat enough of the b-roll that i need here like if this could be five more seconds yeah i do remember i remember right after dolly came out people were basically making these videos of like how to make your your a-roll setup cooler with ai and it was basically like they just had a very contained version and then they generated a filled out image that's awesome of their office to make it look like they were in a bigger space that's sick which is kind of cool but now vo can just do it for you uh the nice thing too is it maintains consistency over time so if you have like a character the

character will look the same and it doesn't do that random stable diffusion thing that uh ai generated video used to do or it would like flash in and out and like change shape slightly and keep moving around it feels like we're only a year or two away from like being able to remove a fence from the foreground i don't know dude only god can do that i don't think that's that's nobody knows how to do that. Wait, sorry, real quick. Adam and I were just like fact-checking some stuff and we found something that's like too funny not to share with you guys like immediately. This is when Google announced last year Lyra,

which is the actual model that does a lot of their generative audio. Look at this image that they use. I just sent it to you guys in Slack. Wait, that's just the waveform logo. That's just like our thing. It's literally our waveform clips background. Yeah.

Like the colors are the same. And like the gradient's the same. And like, it's kind of the same. That's exactly the same. Yo, Google, send us some IO swag from this year to make up for this. Cause it's saying steal this from us. You stole this from us. It's like right on the top of the page.

I'm tagging Tim in this and seeing. It's literally the purple into scarlet color. We'll get back to that. Yeah. Okay. All right. With the VO thing, they said it's going to be available to select creators over the coming weeks. Which is weird that you have to be a creator to apparently

use it and i think what they're trying to do is control the positive narrative by giving limited access to artists who will people will carefully that will say positive things about it which is uh super fun okay the pain in your voice as you said that this actually is kind of cool we found there's some nuggets in this needle stack you know what i'm saying so nugget wait you mean there's some needles in this haystack no needles in this nugget i don't mean that there's some nuggets in this hay in this needle stack if you like hey that's why i said nuggets like chicken nuggets so they have multi-step reasoning in google search now which is cool this is actually something that andrew was specifically calling out a couple weeks ago this is so close to my google maps idea i feel like but it's not quite yeah okay do you want

to describe it yeah it's pretty, it's kind of like using Google search and maps and reviews and putting it all together. So you can have a more specific question or suggestion that you want. So there's was like, I want a yoga studio that's within walking distance of Beacon Hill is what they said within a half an hour walk of Beacon Hill. And it's going to bring up different yoga studios that are within those parameters that you set and then based on the ratings the higher ratings in there so it's an easier thing for you to choose rather than just being i guess it's not that much different than just searching on maps where but in maps you would have to see like oh that's 20 blocks away from me this is actually like gathering that into what it deems as walkable in half an hour.

So yeah, the specific example they used was find the best yoga or Pilates studio in Boston and show the details on their intro offers and walking distance from Beacon Hill. And so it generated, again, this is like the generated page. So it's like kind of scary in a lot of ways, but it pulls up like four that are within the 30 minute walking distance from beacon hill and it shows the distance and it shows their intro offers and

it shows their ratings and that's that's really cool i can't deny that i will we'll say like a real world example of this is last saturday saturday because friday night there was an insane eclipse all over north amer. Not eclipse. There was a Rora Borealis because apparently it was so crazy. Everyone in New York and New Jersey would have had to get anywhere else in the United States.

I got into bed after getting dinner with Ellis on Friday night and it was 1230 in the morning and I opened Instagram and I was like, what the heck is happening? And what do we do the next day? We drove to Montauk. But the whole morning I was like Googling like, uh, what's the weather here? Okay.

And then I had to go on Google Maps and I'd be like, how far is that? Okay. And I had to look at the radar of it again. And so I was jumping back and forth between these tabs like crazy. I was going back to Google Maps and like these weather radar sites and like how clear is the sky during that part of the year? So if I was able to just, if you could trust these generative answers, ask the question like what is the closest dark sky park that won't rain today that I'm most likely to see the Aurora Borealis from? And it would just tell me the number one answer and I could trust it, which is the biggest thing,

that would have been a lot easier than me jumping in forth between all these apps because this legit took me like three hours to figure out where I wanted to go. This poses an interesting question on if it would be able to use real-time information and real-time information that Google provides, but they didn't show examples like that

because what you're saying would need to know whether at a certain time that it's knowing and updating where everything they showed is stuff that could have been posted months ago. Like yoga studio walk intro offers, I guess is more closer to real-time, but like if you wanted to do that or if I wanted to say,

and this is something that's available in Google, what's the closest four plus star restaurant in a 20 minute walk that is not busy right now that like i wouldn't have to wait for yeah could it pull the busy time thing that they have in google and reviews and would it pull that i would hope but i don't know yeah they should have showed it if it can pull that live in there but I don't know if they would be able to do this but yeah this is so this is so close to me being able I just want this to be able to use in Android Auto on maps with my voice and say stuff back to me through the car speakers and be able to do a little more within a route um I feel like this feels like a

step close to that yeah i'm excited for it this is very close to the thing that you mentioned the other yeah yeah we're almost there i'm almost able to find the closest taco bell serving breakfast on my route next next to a third wave coffee that doesn't add 30 minutes yeah that doesn't which is pretty much exactly what you asked i know it is i know it's so close yeah we'll We'll see if that, if that works. Um, Gmail. Now you might think that the updates to Gmail would be things that I would care about. No, no.

So all I want in Gmail is better contextual search because right now you can search keywords and that's basically it. And you have to sort through all of the different things and you have to sometimes just doesn't show the email, even though it has the exact words that you're searching for. Is it safe to say Gmail search function is Google's worst search function out of anything that they do? I would say so. I think it is impossible.

Do you know how many times I try and find my mileage plus number in my Gmail, but I just get 400 united promotional emails that have never been opened it's like if i'm searching for something i probably want it to be something that's been opened and read before this needs the exact same update as google photos just got yes yeah which is well tell me my mileage plus number and then it goes here's what it is and here's the email that showed it okay so, so that's a perfect analogy because the Google Photos thing, you ask it a question and then it brings you the photo that you want to see.

Right. The update to Gmail that they just added is you ask it a question about your emails and it generates you a response. It doesn't bring you to the email that you need to see. It just tells you about your emails, which I don't trust because I just want to see the email. you need to see it just tells you about your emails which i don't trust because i just want to see the email yeah yeah so it can summarize an email chain which is like i guess sure maybe how long your email chains i know corporate email chains

are probably really long and possibly annoying still don't trust a generated answer of things you can ask gemini questions about your email um okay uh it has suggested replies that are generated by gemini which is not that different from the suggested replies it already has right now except that now it's adjusting like full replies instead of just like hey martha as like the beginning of the reply or whatever um one of the examples that they got on something you can do in gmail with jim and i is they asked it to organize and track their receipts so it extracted receipts from emails that they got from a specific person put all the receipts in a folder and drive and then created a google sheets document that

in a bunch of cells like organized the receipts by category. That was awesome. That was cool. It was cool, but it was like so specific and neat that it can do all that. And it still probably can't find my mileage plus number. Yes. I bet the rabbit could do that

if I took a picture of every receipt I've ever had. I think this is cooler though, because it can show... Sorry, I missed the sarcasm completely there yeah i love that that was the number one thing that people were amazed by with the rabbit they were like it can make tables spreadsheets of simple spreadsheets and the humane pin can do that soon sometime sometime in the future yeah i think the biggest problem with with gmail's contextual search right now is the signal to noise ratio is terrible like you were saying like there's one email that says your mileage plus number all of the other ones are

promotional yeah signal is one noise is 999 000 yeah maybe they just did a bad job at explaining it i'm hoping gemini for gmail search could be really awesome because i need it so bad i'm so sick of digging through my email and just not being able to find things that are literally things i or a friend typed me i've legitimately just thought about just nuking my whole email and just starting a new one start over because i'm like i can't find anything that's not a bad idea i think it's a lot but it's and i wish you could do that with phone numbers but you're just getting someone else's like phone number that they're reusing so it's kind of pointless you're gonna get spamming yeah yeah i remember like sorry this is random moving moving a couple years ago and getting solicitation mail and being like how is this possible that i'm getting mail this is a new

address it doesn't make any sense anyway yeah yeah okay okay i would like that to happen so they're they're creating this new workspace side panel that's going to be going into a lot of the google workspace apps like gmail and like google meet and like google chat was a which is part of me but no one i don't think ever has used i just take a minute and be like what's google chat i forgot that i forgot it existed yeah um because it's a chat client by google and you just don't use them because they oh hangouts oh right oh no no sorry wait was it yeah i'll do it wait was it uh google messenger that's what it was oh message messages by google oh messages or is it android messages i thought it was inbox

all right uh yeah so so that side panel is how you're going to interact with Gemini in order to interact with all of your Google stuff. What I found kind of frustrating about this is that it's only in Google Workspace stuff. Oh. And to me, this is their version of Apple lock-in of, like, we're giving you all these Gemini features, but you can only really use it in Google products. But also, like, if you have, like, a regular account that's not a Workspace account, does that still work? No, you can still use it. Like, that's part of Google Workspace, I believe.

Oh, so, like, me as a person with Google Calendar and Gmail and three or four other Google services, I can... You can use Gemini through that. Okay. Yeah. Yeah. Yeah. three or four other google services i can you can use gemini through that okay yeah yeah um they introduced this thing called a gemini powered teammate uh if you use google chat

i forgot about this so this yeah chip so why did they name it stop stop name i actually named it chad before they named it chip and then they were like chip and i was like chad was to be fair that it wasn't that's not what its name is you can name it and they named it chip as just a joke in the thing got it yeah so the way this worked is imagine a slack channel where you're all talking about stuff you're all like sending files and documents and stuff do you have an additional teammate that you can name whatever you want we should name ours chad if we ever do it but you can say like hey chad what was the pdf that i sent to tim last month i can't really find it and chad goes i'll go find that for you and then it comes back and it gives it to you and can ask information

about it so it's basically an employee it's a personal assistant i'm on board yeah yeah so but that's like an only in google chat which literally no one has ever used in history this is gonna be dead in a week i did not know that this was a product yeah like i did most of them yeah on gmail you can see the little chat like because you can rotate between gmail and google chat like whenever you want which is a thing that you can like get rid of oh yeah out of that i always i've no i know oh look you can invite everybody in our work yeah i'm pretty sure only google uses this so i'm not sure how helpful this is going to be okay but it's an interesting idea marquez the last time we talked and it was 2020 i didn't know how you use this Yeah. This isn't Hangouts. It probably was Hangouts when we were talking.

Chats. Yeah. Oh. Huh. Did we say anything fun? Well, this one says I haven't used it since 2013. Oh, my God. That means it was just Gchat, right?

I don't know anymore. I don't know. Who knows? I can't believe we live in this hellscape. It's basically like a live Slack bot doing things for you. Yeah, it's a Slack bot that is contextual and can grab different files that you've used, which is cool. I think that's pretty cool, yeah. It's really sad that it's only in Google Chat, but you know.

But also makes sense, I guess. They're not going to build it for Slack first to show off at I.O. Yeah. Okay. They have a new thing called GEMS, which is basically GPTs. You know how ChatGPT has a custom GPT you can build that's like, this is the MathTutor GPT, and it only talks about math. Which, like, there's an existential question here of like would you rather do

have like one AI agent that knows literally everything and is omniscient or do you have individualized agents? I'd rather just have an omniscient agent. That's just me. But you can have GEMs which are specialized Google Gemini, Gemini's

with ultra deep knowledge of specific topics. And yes. Well sort of. No it's the same knowledge as regular Gemini. It's with ultra deep knowledge of specific topics and yes well sort of no it's the same knowledge as regular gemini it's not deeper than regular gemini knowledge no i don't understand i still don't understand this that much yeah maybe it's less prone to hallucinating if it doesn't have a bunch of extra input maybe maybe there's more well it has the same input though because it's the same model but it doesn. But it doesn't have all the other knowledge of other topics to hallucinate about. I don't know. It's the same as the custom GPTs where you use chat GPT to create a custom GPT.

So it's like that input at the beginning has to be correct. I don't know. The funny thing about this was when the lady on stage announced it she it was really awkward because she was like now you don't use gemini in the same way that i use gemini and we understand that and that's why we're introducing gems and there was like a four second silence and she like laughed there had to have clearly been on the teleprompter like hold for applause and nobody it was just too deep into like you said this soulless event it was not her fault it was just like it was not a great naming scheme everyone was out of it at this

point there were multiple camera cuts of people yawning in the audience yeah and i just think she was waiting for an applause and it did not come and she had like an awkward chuckle i felt really bad about that yeah yeah uh okay some more stuff trip planning which is the stupidest i i just uh ah they're building out the trip planning stuff for some reason um and now you can like talk to gemini and tell you about the trip that you want and it'll make you with a trip and then you say oh but i'm allergic to peanuts so don't bring me to a restaurant that likes peanuts it goes oh okay and it swaps it out for you for a chocolate factory so it's like modular um i don't want to give up control to an

ai but maybe people do i can only assume they're so into this because of like the way they're making some sort of affiliate money on all of these different travel things like i don't yeah i don't want to just like say plan me a trip here and then i have i know nothing about the trip i'm going to yeah and that's what it wants to do there are some psychos who would do that i guess there are you've seen those tiktoks of like i spun around and threw a dart at a globe and when it landed in madagascar i booked my flight like that but then you want to go and have a good time off the cusp you don't want it to plan your entire itinerary right somebody would do it it's funny because when all of the video

actually that honestly would be an awesome that would be an awesome i liked the old chat gpt version of just like help me find like i had to give you the place and specifics of it and like you would help throw a few suggestions like a hike at rocky mountain national park tell me the top five that are over five miles yeah that felt good now we're at the point where it's like can you walk me down this hiking trail and tell me what I'm going to see? I don't like this part. I like the old GPT, not this new bold GPT. Okay. I think we're losing Marques.

So we got to wrap this up. We're giving him the full experience. Six more features. I might die. Can we use Gemini to summarize this event? If you're bored of this episode, use Gemini to summarize this episode but then leave it running in the background i can summarize this episode the rest of this for you

ready okay stuff that google already did is now powered by machine learning that's it well more different types of machine learning basically sure. Sure. Cool machine learning. Now it works 15% worse. And more neurally. Neurally with the brain stuff. Like what we're going to get to next, Gemini on Android. What are they using it for? Scam detection. Going to work 15% worse.

They're doing like a circle to search hover sort of thing. Which they already had. 15% worse. Yeah. I want to say, I'm pretty sure they showed how you can buy shoes with Gemini five different times during this event. They're very into live shopping. Yeah. Yeah.

I think Google's really trying to amp up the live shopping affiliate stuff because I think they see the writing on the wall for like search not making money anymore in the distant future. So they're like, how can we just like create new revenue streams they're killing off their their ad streams by summarizing all the websites that they sell ads on so now they're just going to make the affiliate marketing through the right the the e-commerce that just appears because google ad sense is like oh my god yep they're the backbone of the web and they're also throttling it and making sure it doesn't work anymore yeah all right gemini in android cool there's this gemini hover window now that can hover above apps so you can like ask it questions about the app and interact with the app and do that kind of stuff um seems helpful sometimes

annoying other times if it was one of those android things where it's like kind of pinned to the right corner and you can press it and then appear when you want it to be yep that'd be nice i'm not really sure how they're integrating it but you know what this is giving me the vibe of i feel like google doesn't know what to use ai for and it reminds me of like that too it reminds me of like when apple makes a new thing and they're like developers will figure out what to do with it yeah like google doesn't know what to show the ai doing so it's just doing a bunch of things it already does and they're just like treating it as a new language almost where they're like developers and you guys will eventually

figure it out and make it useful can i have a counterpoint to that go i think google knows exactly what their ai can do and just like with bing and like old chatbots that went off the rails google is way too scared of the power that they have and what people could do with it that they have to show the most minuscule safe examples of things and release it in a very safe way because they know how much power they have buying sneakersying sneakers. Buying sneakers. Yeah. Make us some money. Don't do anything that would hurt the entire,

our entire brand. Yeah. I think that's, that's my take on it. Yeah, but at this point, everyone already kind of sees them as losing to OpenAI. So like, why not just do it?

They know that they actually are winning and it's only a matter of time. And they know they're not behind. They're so much better positioned than open ai they're fine and this event really kind of showcased that because it showcased all the google workspace stuff that they could actually use gemini in um this isn't a joke though we still do have five more things to cover but i'm gonna go quick yes i promise okay so in the gemini through it. Okay. So in the Gemini and Android floating window thing, you could be watching a YouTube video. You can be asking questions about the YouTube video

while you're watching it, which is semi-cool. However, it hallucinates and gets stuff wrong. Someone gave an example. They talked about pickleball again because that's all the tech bros talk about. And the Google executive, who is in the pickleball because of course you are um asked was like is it illegal to put a spin on the pickleball and it was like yes you cannot do spin on the pickleball and some pickleball person on twitter was like what yes you can

um so that's fun uh it now has ai powered scam detection which is kind of creepy uh because it is consistent it's listening to your phone call when you pick it up and it's listening to what the other thing on the end of the line is saying because it might not be a person might be a robot by then it's too late and it will tell you it will so the example they gave was like someone who had answered a call and was like we just need you to like click this link and blah blah blah and then on the phone there's a little pop-up that says like this is probably a scam you should hang up now i actually think that's awesome great for old people except that it's listening to you all the time which is weird yeah which is weird and i imagine they're gonna probably do something about like it's only

on device and we don't do anything with the information we get rid of it immediately after i think it would be super useful for people who would fall for those scams because so many of them it takes 30 minutes to an hour some of them have to go out and buy gift cards and stuff like that and at that point it could be like this is is a scam. Hang up. That could save people literally life savings. Totally. I think that's awesome. Yeah. They're calling all of these new Gemini and Android features

Gemini Nano with multi-modality because we need more branding. And it's, yeah. Starting on Pixel later this year, I imagine it'll get moved out to more Android phones later. They're definitely trying to build a fortress against Apple and be like, look at all our AI stuff that only Android phones have. Google Chrome is also getting Gemini Nano. I'm not really sure what that means.

You can do basic stuff in it that you can do with Gemini Nano now if you use Chrome. That was sort of passive. They're not that scared of other browsers right now. They still have way too much market share. So last year, they also introduced this thing called SynthID because when they showed off the image generation, they basically talked about this open standard that they want to create

where it bakes in this hidden watermark within an AI-generated image because we all used to be worried about Dolly stuff, and now we're worried about other AI things now. But now they're expanding S id to text somehow um no idea how they're gonna do that but that's interesting and also video uh so i don't think there's gonna be an interesting standard but it's just their version of figuring out whether or not something's ai generated they made a joke at the end of io where they were like, we're going to save you the trouble with how many times we said AI. Okay.

We said it 120 times and they were like, ha ha ha funny. And then right as Sundar was getting off the stage, he said it one more time and that ticker went up to 121, but then they said it again. So it was one 22. I'm just saying. Also,

I think that sure they can like make fun of that and say all that stuff but they said gemini i think i think the final count on gemini was 160 times um and my theory is that they're just replacing ai with gemini because now they don't need to convince the stock market that they're an ai company anymore now they just want to bake gemini into your brain as hard as physically possible so they just said it over and over and over again um which was funny it was funny yeah yeah so i think um just to wrap this all up because i'm sure everyone listening or watching is begging for us to be done with this i'm sorry um it's um matt ansini from This Is and Austin Evans' channel, I think had a great tweet to kind of explain

why both of these events felt not great. He said, the problem with overhyping AI is that all the stuff we're seeing today doesn't feel new. This stuff is cool, it's impressive, but it's barely catching up to what we were told AI was potentially capable of two years ago. So it feels almost like no advancements have been made i thought that was a pretty good summary of everything like we've watched so much in the past

of like this is the stuff we're gonna do look how amazing it is and now we're getting the stuff it actually is doing and it hasn't made it to that i also think they just throw around technical jargon with like model sizes and token like parameterization. Yeah, they had a whole TPU. They talked about water cooling. They're like, we have state-of-the-art water cooling in our new TPU center. And it was like, who cares? Who are you trying to tell this to?

Big water. Big water. Yeah. So yeah, we can start to wrap it up. I'm so sorry, guys. So Marques, are are you gonna go back and watch the events i don't think i'm gonna watch it i do think i got the idea that like google is trying to make sure everyone knows that they've been an ai company for super long and then on the

other side of the spectrum is apple where they're like okay we don't need to say this over and over and over and over again but we do need people to understand that we have at least been doing something ai related and they're like these are opposite ends of the spectrum of like how in your face they want to be about that and both are fine both are fine i don't really have a horse in the race i'm just hoping they just keep making their stuff better well that is a great place to end this podcast which means it's a great place I don't really have a horse in the race. I'm just hoping they just keep making their stuff better. Well, that is a great place to end this podcast, which means it's a great place to do our trivia answers.

How do you have all of them? I don't know. Wait for trivia. But before we get started, quick update on the score. One thing at a time. quick update on the score one thing at a time in second place we have a tie with eight points between andrew who's currently carrying the one and what does that mean david uh they both have eight points wait marquez got something right last week yeah i didn't watch it oh right we didn't update you about that yeah marquez technically was closer than you last week on the uh how many times did apple say oh so he actually stole your point he actually got the exact number yeah so you because it

was you get it from quinn's tweet yeah he did i'm pretty sure right thanks quinn um yeah so uh yes unfortunately andrew your lead has been temporarily taken by marquez brownlee um uh who has nine points is this the closest trivia we've had this deep into a trivia i think so yeah yeah it's it's usually my fault that it's not close but but question one google io happened yesterday a few days from the time this is going to get published, but yesterday from recording. And we all know I.O. stands for input output, classic tech thing. We also know it sometimes stands for Indian Ocean, if you watched any of our domain name specials. But according to Google's blog, the keyword I.O. actually has two more alternate meetings to Google.

What are they? And I'm giving one point for correct answer. Two possible points. I have zero idea. Yeah. I feel like one of them you would never get. The other one you might get. I think both of them you'll both definitely get.

Not if I don't write it. Oh, interesting strategy. Interesting strategy. All right, Andrew, what'd you put? I put blank. Audio listeners, something happened. Who should go next? David, David.

All right, I wrote two answers. Information operation, clearly. And I also wrote, I'm out. But that is hilarious. Thanks. Marques, what do you got? I wrote IOM being like the one zero of binary oh it's so you're so close really so close yeah but his second answer is the best and google io

for in and out because every time we go we gotta go i almost want to give him a point for that because that's better than whatever they could have thought of no um so uh the first thing they listed is the acronym innovation in the open oh sure yeah right um yeah okay according to the keyword what first prompted them to use io prompt in the original google io was they wanted it to reference the number google which begins with a one and looks like an i and is followed by a zero it's technically followed by 99 more zeros after that zero yeah but that would make for a really long event title fair fair funny so now you guys know that. Okay. Cool. Okay. Next question.

What was the name of the AI in the movie Her, which we recently learned Andrew and Marques have not seen. So this is going to go lovely. VR 180 creator. Be me, 2013, midnight, Santa Cruz, California. You're a hippie. You're at a movie theater by yourself, sobbing at 1.30 in the morning when the movie gets

out. Actually, it was probably closer to 2.15. What is happening? You sit on a bus for 45 minutes on the way back to campus and your entire life has changed and now you're sitting in Kearney, New Jersey. And you could never see Joaquin Phoenix in the same light. Marques, what'd you write?

Siri. Nope. You wrote Alexa. I wrote Paula. Paula? Yeah. No. I wrote Samantha.

Correct. Let's go. Samantha. Who would have guessed the person who saw the movie knew the answer to that? I actually was thinking really hard. I would have thought

that you guys would have seen the movie. I don't know. I thought it was a pretty straightforward question. You thought I have seen the movie. I don't know I thought it was Sam Altman has seen this movie there was a funny article that was written about the open AI event that said I It was called like I am one I'm once again pleading that our AI overlords watch the rest of the movie. Yeah. And it was about how they were like referencing her at the OpenAI event. But the ending of her is supposed to be like.

Ah, ah, ah, ah, ah, ah. Shh. No spoilers. Anyway, it was fun. All right. It was ha ha. We're wrapping this up because as little do you know, we're recording another episode directly after this. I'm hot and sweaty.

So, okay. Yeah. Thanks for watching and subscribing, of course, and for sticking with us and for learning about Google I.O. alongside us. And we'll catch you guys next week. Did all the people who subscribed last week make you feel better? Yes. It worked.

No, I feel pretty good. It worked. Yeah. The I.O worked yeah yeah the io and google also stands for like and subscribe don't ask waveform was produced by adam molina and ellis rovan we're with vox media podcast network and intro outro music was by vane sill I don't think there's no, you might be the only person in the world who, who like doesn't even know the plot of this movie though. Like, you know what this movie is about,

right? Neither of you. Oh my God. Oh my God. You what this movie is about, right? Neither of you? Oh my God. Oh my God, you guys. This is like, you are the two perfect people to watch this movie with no context. I wanna stay as the perfect person to watch this movie with no context.

All of the AI things we've been talking about would make so much more sense. Can you picture Andrew coming in after sitting through yesterday's Google IO and then watching her for the first time and being like, I get it. Bro, her formed 90% of my personality.