AI transcript
0:00:02 (upbeat music)
0:00:06 – Hey, welcome to the Next Wave podcast.
0:00:07 I’m Matt Wolf.
0:00:09 I’m here with my co-host, Nathan Lanz.
0:00:12 And in this show, it is our goal to keep you looped in
0:00:15 on all the latest AI news, all the coolest AI tools,
0:00:18 and just generally what’s going on in the AI world.
0:00:20 And the last couple of weeks
0:00:23 have been absolutely insane in the AI world.
0:00:24 There have been so many announcements.
0:00:28 We had Google I/O, we had OpenAI’s announcements.
0:00:30 We had two different keynotes from Microsoft,
0:00:32 and all of these announcements
0:00:34 had so much going on around AI.
0:00:38 So we thought we’d do a fun little bonus episode,
0:00:39 maybe a little bit shorter than normal,
0:00:41 and just kind of riff and talk about
0:00:43 some of these announcements and, you know,
0:00:45 get you looped in on what’s going on
0:00:46 in the world of AI right now.
0:00:50 So the first event that came up last week
0:00:52 was the OpenAI event, which is kind of funny,
0:00:55 because we knew Google I/O was happening
0:00:56 last week on Tuesday.
0:00:58 So what does OpenAI do?
0:01:00 They go and make their announcement on Monday,
0:01:03 the day before the big Google I/O event.
0:01:06 That was sort of the kickoff to this announcement fest
0:01:09 of all of these AI announcements from these big companies.
0:01:14 And of course, the big OpenAI announcement was GPT-40.
0:01:15 And during this announcement,
0:01:18 they showed off some demos of GPT-40,
0:01:22 being able to sort of have conversations with you,
0:01:24 the inflections of the voices change,
0:01:28 the voices sound eerily similar to Scarlett Johansson,
0:01:31 which is the voice of the AI in her.
0:01:34 So there’s some drama going around about that.
0:01:36 But I figured let’s kick off this conversation
0:01:38 talking about GPT-40.
0:01:44 When all your marketing team does is put out fires,
0:01:45 they burn out.
0:01:46 But with HubSpot,
0:01:49 they can achieve their best results without the stress.
0:01:52 Tapping to HubSpot’s collection of AI tools,
0:01:55 freezes to pinpoint leads, capture attention,
0:01:58 and access all your data in one place.
0:01:59 Keep your marketers cool
0:02:01 and your campaign results hotter than ever.
0:02:05 Visit hubspot.com/marketers to learn more.
0:02:11 – So, I mean, Nathan, like what are your initial thoughts?
0:02:13 When you watch that keynote, like,
0:02:15 were you blown away by it?
0:02:17 Were you underwhelmed by it?
0:02:19 Like what were your thoughts when you first watched it?
0:02:20 – I was honestly blown away by it.
0:02:21 Like just like, you know,
0:02:22 in one of our previous videos,
0:02:25 we talked about how I think in a year from now,
0:02:27 voice will be the main way that you interact with AI.
0:02:29 And it’s like, okay, well, now here it is.
0:02:32 Like they’re showing you can actually just talk to the AI.
0:02:33 You can actually even interrupt it.
0:02:35 It can be talking and you can just like interrupt it
0:02:36 and start talking again.
0:02:37 It’s not like, it’s not like the old one
0:02:41 where you had to like record and then like processes
0:02:43 and then it says something and you had to wait till it’s done.
0:02:45 No, you can just like talk back and forth with it.
0:02:47 And it wasn’t perfect,
0:02:50 but damn it was pretty good for like a first version
0:02:52 of like this kind of voice interact with AI.
0:02:54 But yeah, I was pretty blown away by it personally.
0:02:55 – You know, one of the big announcements
0:02:57 that came out with the GPT-40
0:03:00 is that pretty much the latest state of the art models
0:03:02 that, you know, the general consumers
0:03:04 are going to get access to,
0:03:05 they’re going to give them access free, right?
0:03:09 So like the GPT-40, people are going to get,
0:03:10 they’re going to get access to that
0:03:12 just in the free chat GPT now before, you know,
0:03:15 if you had the free chat GPT, you were using, you know,
0:03:17 GPT 3.5 and older model.
0:03:20 If you were on the chat GPT plus subscription,
0:03:23 you got chat, you got GPT-4 turbo.
0:03:25 The naming conventions are all, you know, weird
0:03:30 ’cause you got GPT-4, GPT-4 turbo, GPT-40, GPT-3.5.
0:03:33 Like it’s all confusing, but you know,
0:03:35 Sam Altman actually made a comment
0:03:36 in one of the recent interviews
0:03:38 that he’s probably going to stop, you know,
0:03:41 naming them GPT-4, GPT-5, GPT-6, right?
0:03:43 They’re going to just sort of land on a name
0:03:46 and just keep on improving on that version
0:03:47 over and over again.
0:03:49 But I thought it was really cool
0:03:51 that they’re basically saying our most state-of-the-art
0:03:53 model, the best models that we’re putting out,
0:03:55 you’re going to be able to use just for free now
0:03:57 in the free version of chat GPT.
0:03:59 I thought, I thought that was really, really cool.
0:04:01 And then if you’re a plus subscriber,
0:04:02 then you just get more use, right?
0:04:04 You get, you’re going to get some of the features
0:04:06 a little bit earlier than everybody else,
0:04:08 but you’re going to get like five times more
0:04:10 input/output with the model.
0:04:11 – I think it’s a big deal
0:04:13 because you know, most people when they think of AI,
0:04:15 they’re thinking of chat GPT
0:04:17 and they’re typically thinking of the free model, right?
0:04:20 Which has been 3.5, which you know,
0:04:22 compared to 4 is really, really bad.
0:04:24 And so the fact now that someone gets,
0:04:26 you know, I think 4.0 in some ways
0:04:28 is not as good as 4 in some ways,
0:04:31 but it’s pretty darn close and it’s way, way faster.
0:04:34 So for most people, that’s going to be such a, you know,
0:04:35 step forward in terms of what they believe
0:04:36 is possible with AI.
0:04:37 Like, and so I think that’s going to lead
0:04:39 to like a lot more awakening around like,
0:04:42 oh, AI is actually here, it’s really powerful.
0:04:44 But, you know, I think one thing worth thinking about though
0:04:46 is like, apparently they were training this model
0:04:49 with a separate team ever since 2022.
0:04:50 So my understanding is like, yeah,
0:04:53 this is kind of their state-of-the-art model,
0:04:55 but I don’t think it’s their like flagship model.
0:04:57 I think it’s just like another model
0:04:58 that they were developing.
0:05:00 And so they put it out there for free.
0:05:02 My theory is it’s probably built on like a better architecture
0:05:03 where it’s more cost efficient.
0:05:05 That’s why they’re able to give away for free.
0:05:07 That’s why it’s faster.
0:05:09 And then five is probably going to be built
0:05:10 on top of that as well.
0:05:12 So I think we’re still going to see GPT-5 quite soon.
0:05:14 And it may not be multimodal.
0:05:15 That’s my big question.
0:05:17 Like, okay, so if it’s a different model,
0:05:19 are we going to get something that’s dramatically smarter
0:05:21 with GPT-5, but it’s not multimodal yet.
0:05:22 It’s just text.
0:05:25 – Yeah, but this new model, GPT-4O was, you know,
0:05:27 pretty multimodal, right?
0:05:29 They were, they were inputting video,
0:05:31 although I think the video they were showing,
0:05:33 it was just sort of looking at screenshots from the video.
0:05:35 It wasn’t actually watching the video
0:05:37 because there was one moment in their demo
0:05:39 where he picked up the phone and like smiled
0:05:40 into the camera and it was like,
0:05:42 oh, I’m looking at something wood.
0:05:43 And he was like, no, no, that was before.
0:05:45 Now look at my face, right?
0:05:47 So it obviously sort of like took a snapshot
0:05:49 of the table or whatever.
0:05:51 And then when he looked at his face,
0:05:53 it still was like thinking about the picture
0:05:54 of the table or something.
0:05:56 So I don’t think it’s actually using video,
0:05:59 but it does seem to be ingesting audio.
0:06:01 You can put images into it.
0:06:04 So I mean, it’s pretty multimodal already.
0:06:05 I think the real question is,
0:06:08 are we going to get like actual video input with GPT-5?
0:06:11 – Yeah, I guess we’ll see.
0:06:12 – So there were some other announcements
0:06:13 during the OpenAI event.
0:06:15 So I’ll just kind of rattle off some of them.
0:06:17 I’ve got my notes here in front of me.
0:06:19 They talked about a new desktop app,
0:06:22 which it’s pretty cool.
0:06:24 It’s only available on Mac right now.
0:06:27 They didn’t make it available on PC right away.
0:06:30 And I think I’m at the Microsoft event right now.
0:06:31 If you’re watching on video,
0:06:33 you can see I’ve got a different background than normal.
0:06:37 I think we found out why they didn’t release a Mac version
0:06:38 while I’m at the Microsoft event,
0:06:39 but we’ll get into that
0:06:42 when we start talking about the Microsoft event here.
0:06:44 But the desktop app, it’s pretty cool.
0:06:45 I tested it out on my Mac.
0:06:47 It doesn’t have the voice features in it yet.
0:06:49 It doesn’t have the feature
0:06:51 where you can view your desktop yet.
0:06:54 You can take a screenshot, drag the image in,
0:06:57 and chat with the screenshot essentially,
0:06:58 but it doesn’t just look at your desktop
0:07:00 like they showed in the demo yet.
0:07:02 It’s pretty much the same thing
0:07:04 you get inside of the chat GPT web app,
0:07:05 just on your desktop.
0:07:10 Now we also learned that that GPT-2 chatbot
0:07:12 that was all over Limsis on the chatbot arena,
0:07:15 you know, apparently that was GPT-4O.
0:07:17 That was them kind of testing that out
0:07:20 to see how people would work with it.
0:07:23 And you know, the other big thing was like
0:07:24 the voices in it, right?
0:07:28 Like the GPT-4O, it feels like a marginal improvement
0:07:31 with like the intelligence of it, in my opinion, right?
0:07:33 It doesn’t seem like huge leaps
0:07:34 above what we were getting before.
0:07:36 Other than in coding, it seems to work really well
0:07:39 with coding, but the like the voice stuff
0:07:41 was I think the thing that most people
0:07:43 were going nuts about from that keynote, right?
0:07:46 When you speak to it, you can tell it
0:07:49 to give you more emotion, get more excited about it,
0:07:52 talk like a robot, do all that sort of stuff.
0:07:53 And then when they were demoing it,
0:07:55 that’s also where we heard the voice,
0:07:58 which sounded very similar to Scarlett Johansson.
0:08:00 But I think that was the thing
0:08:03 that people really went nuts over was that voice thing.
0:08:05 And like you mentioned earlier in this episode,
0:08:07 we did a whole episode about how like
0:08:09 voice is that next thing, right?
0:08:12 Voice is what’s going to be the sort of input
0:08:15 of all of these large language models.
0:08:17 And when chat GPT showed off that voice
0:08:19 and there was really like no latency,
0:08:21 they would ask a question, it would respond
0:08:22 almost in the same amount of time,
0:08:23 a real human would respond.
0:08:26 You know, that was the thing that was probably
0:08:29 the biggest sort of differentiator about,
0:08:31 you know, that from what came before it.
0:08:32 – Yeah, I mean, I tweeted about the whole
0:08:34 Scarlett Johansson thing yesterday
0:08:35 and it got a lot of attention,
0:08:37 especially on LinkedIn for some reason.
0:08:38 People, you know, seem pretty upset about it.
0:08:40 You know, I have mixed feelings.
0:08:41 I mean, I guess to give people context,
0:08:44 like Scarlett Johansson put out a statement
0:08:45 or I’m not sure if it was officially from her,
0:08:48 but her publicist, you know, said that it really happened
0:08:50 where openly I contacted Scarlett Johansson,
0:08:52 asked like, can we use your voice?
0:08:53 She’s like, no, you can’t.
0:08:57 And apparently they kept trying and she just refused.
0:09:00 And they kind of picked a voice that seemed similar to hers.
0:09:01 And then she’s saying, not only that,
0:09:03 but then like they threw it in my face
0:09:04 ’cause like Sam Altman like tweeted to somebody
0:09:06 like the word her, you know,
0:09:09 and like referencing the movie where she’s like the,
0:09:11 you know, the main voice for the AI system.
0:09:12 I have mixed feelings about it.
0:09:14 ‘Cause like when I, when you actually,
0:09:16 somebody else shared a, you know,
0:09:18 side-by-side comparison of the voices, you know,
0:09:20 like here’s one, here’s the sky voice,
0:09:23 the open AI voice, and then here’s Scarlett Johansson’s voice.
0:09:24 You know, they’re quite different.
0:09:26 I mean, like they don’t sound like,
0:09:28 sure it sounds like an attractive woman talking
0:09:29 or something, but it doesn’t sound
0:09:31 really like Scarlett Johansson.
0:09:33 And then like, so can she really like,
0:09:34 what does she own?
0:09:37 Like AI, like feminine AI voices forever, right?
0:09:39 Like that’s kind of, you know,
0:09:40 that’s kind of like a wild statement to make
0:09:42 that you own that, but they, but who knows?
0:09:45 They’ll probably still end up having to settle with her
0:09:48 for the fact that like he said her, if I had to guess.
0:09:49 – Yeah, yeah, that’s true.
0:09:51 I mean, open AI did put out their own statement
0:09:52 around what was going on.
0:09:54 They said that they hired a different voice actor,
0:09:55 not Scarlett Johansson.
0:09:57 They trained on hours and hours
0:10:01 of this other voice actors work to get this in.
0:10:03 But from what I understand,
0:10:05 they actually are not planning on launching
0:10:07 with that sky voice now that, you know,
0:10:08 just to avoid all the controversy
0:10:10 and all the issues that might come up.
0:10:13 I think that, you know, right now they’re just saying,
0:10:14 we’re just not going to launch with that voice.
0:10:16 You know, we don’t want to deal with all that.
0:10:18 So we’ll see how that plays out.
0:10:20 So HubSpot just put out this really cool checklist.
0:10:22 It’s called the AI adoption checklist.
0:10:25 And if you’re trying to implement AI into your business
0:10:29 in any way whatsoever, this checklist has you covered
0:10:33 from security to privacy to planning to training.
0:10:34 This has you covered.
0:10:36 Click the link in the description below
0:10:38 and grab the checklist today.
0:10:40 – Yeah, so then the very next day
0:10:41 after this open AI keynote
0:10:43 where they made those announcements
0:10:45 was the Google IO event
0:10:48 where that’s Google’s developer conference
0:10:50 where they typically make all of their announcements.
0:10:53 And this year it was once again, all about AI.
0:10:55 In fact, at the end of the conference,
0:10:58 Sundar, the CEO of Google actually counted up
0:11:00 how many times they said AI.
0:11:02 And it was like 121 times
0:11:04 or something throughout the presentation.
0:11:06 One thing that I find interesting
0:11:09 between the two keynote presentations
0:11:12 was that I feel like ChatGPT and open AI,
0:11:16 they really sort of like honed in on like the one thing,
0:11:17 right?
0:11:19 They really sort of honed in on like the voice feature
0:11:22 and the chatting and the conversational element of it
0:11:25 and the tone and the intonations of the voice.
0:11:28 It was very, very focused on that thing.
0:11:31 The Google event in contrast to it was like,
0:11:33 here’s an announcement, here’s an announcement,
0:11:35 here’s an announcement, here’s an announcement.
0:11:36 It was just two hours of an announcement
0:11:39 after announcement, after announcement.
0:11:40 I was at the event
0:11:43 and I’m actually kind of struggling right now
0:11:46 to remember what were all the announcements Google made, right?
0:11:51 So like ChatGPT or open AI had like one big announcement
0:11:54 that everybody remembers, everybody talked about.
0:11:57 Google just like bombarded us with announcements
0:11:59 but they were all kind of so marginal
0:12:01 that I almost forget them.
0:12:03 Like I know I’ve got my notes in front of me,
0:12:06 but like there’s nothing that stands out in my mind
0:12:10 of like Google talked about this and that is game changing.
0:12:12 But I’m curious, Nathan, like from you,
0:12:14 was there anything from the Google keynote
0:12:18 that specifically stood out that was like,
0:12:19 okay, wow, that’s something that Google did
0:12:21 that I thought was impressive?
0:12:23 – No, I mean, I think it was like information overload.
0:12:25 Like they put out so many things
0:12:29 and if you look at the open AI event in comparison,
0:12:32 it was way more like Steve Jobs like, right?
0:12:34 Where there’s like, here’s this one thing that we have
0:12:37 and they just demonstrate in a beautiful way.
0:12:39 And I think open AI did a really good job too
0:12:40 where they even humanize it more
0:12:42 by having all these different team members
0:12:45 show the same thing in different ways.
0:12:49 Here’s the AI voice, our CTO co-founders showing it to you,
0:12:51 presenting it to you, and then here’s people from the team
0:12:53 and here’s how they’re using it.
0:12:55 And they showed like these six, seven different use cases
0:12:58 all with the same product, the same AI voice.
0:12:59 And I thought that was such a great way
0:13:02 to like reiterate the one thing over and over and over.
0:13:04 So by the end of it, it was like really ingrained
0:13:06 in your head like, oh, they’ve done an amazing job
0:13:08 with AI voice, whereas with, you know, with Google,
0:13:11 it’s, they release so many things and it’s like,
0:13:13 it’s not clear when a lot of it’s actually gonna come out.
0:13:15 That’s really hard to like pinpoint like,
0:13:16 oh yeah, what’s the one thing?
0:13:17 And I was like, it sounds like, oh yeah,
0:13:18 they made a lot of progress.
0:13:20 There’s a lot of cool things they did.
0:13:21 You know, like I put out Twitter threads
0:13:23 about both companies’ announcements.
0:13:26 And, you know, my Twitter thread about OpenAI
0:13:28 got like 5.9 million views.
0:13:30 And my tweet thread about Google,
0:13:33 I think it got maybe like 50,000 views just in comparisons.
0:13:36 It kind of showed you the general interest level
0:13:39 in OpenAI’s announcement versus Google’s announcement.
0:13:40 – Yeah, one thing that I found like really interesting
0:13:43 about the OpenAI event was that they had so much more
0:13:45 that they could have talked about as well, right?
0:13:48 Like they put out a blog post about all of the stuff
0:13:51 that they were rolling out with their new rollout.
0:13:54 This GPT-40 is its own image generator.
0:13:55 It’s not using Dali.
0:13:58 It’s like using its own built-in image generator.
0:14:00 It can do text to like 3D object.
0:14:03 It’s got like a ton of features on their blog post.
0:14:06 But in that keynote, they focused in on like the one thing
0:14:09 that I thought maybe they felt was gonna have
0:14:12 the most impact on people from watching that event.
0:14:13 So I thought that was really interesting
0:14:16 that they just, they could have themselves
0:14:18 done a two-hour keynote where they bombarded us
0:14:21 with OpenAI announcements, but they chose to focus
0:14:23 on just like one or two key things.
0:14:25 So I felt that was really, really interesting as well.
0:14:28 – Yeah, and I thought it was fascinating too
0:14:31 that, you know, Sam Altman didn’t appear.
0:14:33 And so for me, that’s like, okay, he didn’t appear
0:14:36 because he’s going to appear when GPT-5 comes out
0:14:37 in like three months.
0:14:39 – I’m at the Microsoft event now.
0:14:41 He was here. He was part of that keynote.
0:14:43 But we’ll get to Microsoft in a minute.
0:14:43 He’s like…
0:14:47 We’ll be right back.
0:14:49 But first, I want to tell you about another great podcast
0:14:50 you’re going to want to listen to.
0:14:54 It’s called Science of Scaling, hosted by Mark Roberge.
0:14:56 And it’s brought to you by the HubSpot Podcast Network,
0:15:00 the audio destination for business professionals.
0:15:02 Each week, host Mark Roberge,
0:15:04 founding chief revenue officer at HubSpot,
0:15:07 senior lecturer at Harvard Business School
0:15:09 and co-founder of Stage 2 Capital,
0:15:12 sits down with the most successful sales leaders in tech
0:15:15 to learn the secrets, strategies, and tactics
0:15:17 to scaling your company’s growth.
0:15:20 He recently did a great episode called How Do You Sol
0:15:22 for a Siloed Marketing and Sales?
0:15:24 And I personally learned a lot from it.
0:15:26 You’re going to want to check out the podcast,
0:15:27 listen to Science of Scaling
0:15:30 wherever you get your podcasts.
0:15:34 – I do want to talk about a few more
0:15:35 of the Google things that they announced
0:15:37 ’cause there was something I thought was cool
0:15:38 while I was there.
0:15:41 But they’re also like, if I’m being honest,
0:15:43 they’re kind of like, they’re sort of gimmicky
0:15:46 and they’re kind of like almost forgettable,
0:15:47 but they’re still kind of cool, right?
0:15:50 Like they showed off this feature called Ask Photos
0:15:54 where you can basically use like an AI search
0:15:55 to search through your photos.
0:15:56 And some of the examples they gave was like,
0:15:58 when did my daughter learn how to swim?
0:16:00 And it would look through all of your photos,
0:16:02 all of the timestamps, try to find photos
0:16:04 where your daughter is swimming
0:16:07 and then tell you your daughter learned to swim
0:16:09 based on all of your photos, right?
0:16:13 So it uses the photos as context to answer your questions.
0:16:15 That I thought was pretty cool.
0:16:16 Like I can see that being really useful
0:16:19 if you have a giant database of photos on your computer
0:16:21 with like thousands of photos
0:16:24 and you’re trying to find like one specific thing
0:16:26 or maybe you’re making a video online
0:16:29 and you need like some B-roll for your video
0:16:31 and you want the B-roll to just be like shots
0:16:33 of the ocean or something.
0:16:35 You could go in there and let the AI find
0:16:37 like just the shots that you’re looking for.
0:16:40 So I thought Ask Photos was a pretty cool feature.
0:16:42 – When I saw that demo, I was like, oh, that was amazing.
0:16:44 They probably should have made that better
0:16:46 and really focus on something like that
0:16:48 ’cause it’s like, they announced like 10 things.
0:16:50 And most of it seemed like they were just kind of like
0:16:52 tacking on AI onto existing products, right?
0:16:53 It’s like a whole joke about like–
0:16:54 – I think a lot of it was.
0:16:56 – If you see like the little three stars,
0:16:58 that means you’re just like adding AI on
0:17:00 for like the sake of it, right?
0:17:03 And it felt like a lot of their stuff was that.
0:17:04 And there was this engineer who made a statement
0:17:06 about that the other day.
0:17:08 Apparently he worked at Google for like 20 something years.
0:17:10 He was a pretty, you know, top engineer at Google.
0:17:12 And he said, they’re doing the same (beep)
0:17:14 they did with Google Plus when like Facebook came out
0:17:16 where like, they’re like panicking.
0:17:17 They had to create something else and they’re just like,
0:17:20 start, okay, we have to have a social network.
0:17:22 Well, let’s just put all Google’s existing stuff
0:17:24 into a social network and like it just doesn’t work.
0:17:28 And you know, they’re also changing search too,
0:17:29 which I mean, is a big deal.
0:17:31 You know, where they’re like using AI
0:17:33 to like analyze the results of a lot of search terms.
0:17:36 I guess it’s only like 1% or so of searches
0:17:37 are currently using that feature,
0:17:40 but it rolled out like two days ago and you know,
0:17:42 it’s apparently some people in SEO land
0:17:45 are already seeing like 20% drop in traffic from that.
0:17:46 Like in the last week.
0:17:50 And so it’s, you know, they’re in a tough situation
0:17:51 ’cause like all the stuff they released,
0:17:54 they’re gonna like cause issues
0:17:56 if they just start dramatically changing Google search,
0:17:57 you know.
0:17:58 – Well, I mean, their revenue is all from,
0:18:00 from ad revenue from search, right?
0:18:02 Like that’s the biggest portion of the revenue
0:18:03 for the business.
0:18:06 So if people stop clicking on those search ads,
0:18:08 well, what happens to Google’s business?
0:18:11 I’ve actually noticed that pop up a lot more often.
0:18:14 Like almost every search I’ve done in the last like a week,
0:18:15 I don’t search Google a lot anymore,
0:18:18 but like I would say almost all of the searches
0:18:19 I’ve done in the last week,
0:18:23 I’ve seen that AI sort of summary thing come up at the top.
0:18:24 I still end up clicking on the links though.
0:18:27 Like usually if I’m Googling something,
0:18:28 I’m trying to find a website.
0:18:30 I’m not trying to find an answer to a question
0:18:32 because if I’m trying to find an answer to a question,
0:18:35 I’m just going straight to chat GPT these days, you know,
0:18:36 or Claude or something like that.
0:18:38 – You know, people are making a good point too.
0:18:39 It’s like almost like Google is like violating
0:18:41 this social contract that always existed
0:18:43 with the open web, right?
0:18:46 Where it’s like, hey, we’re like searching all your stuff.
0:18:49 We are taking your data and we’re making money off of it,
0:18:51 but we’re going to also send people to you.
0:18:53 And so that’s always been this kind of like social
0:18:54 contract that existed.
0:18:57 So now if they are taking in people’s data,
0:18:59 but then they’re just giving the answer,
0:19:00 how are they, and then they’re like,
0:19:02 how could they possibly criticize chat GPT?
0:19:04 (laughing)
0:19:08 How could they criticize chat GPT if open AI actually did
0:19:10 crawl all of YouTube’s data, like it’s, you know,
0:19:12 the rumor that they did.
0:19:13 I think they have a hard argument there.
0:19:14 That’s probably why they haven’t like really
0:19:17 went after it yet, because like they’re doing similar.
0:19:20 – Google’s a web crawler scraping company, right?
0:19:23 Like it scrapes the entire web’s data also.
0:19:26 The only difference is like if when Google scrapes the web,
0:19:28 it’s actually beneficial to the creators
0:19:30 and when chat GPT scrapes the web,
0:19:32 it sort of disincentivizes creators, right?
0:19:35 There’s definitely a big differentiator there.
0:19:39 – For now, for now, but it’s changing almost every month
0:19:39 right now.
0:19:41 So it’s like every month it’s like, you know,
0:19:43 people are going less to websites,
0:19:44 they’re just getting the answer from Google.
0:19:48 So yeah, I think Google’s in a very tough situation there.
0:19:49 And I don’t know how they’re going to get out of it.
0:19:51 – Yeah, well, the other thing is, you know,
0:19:53 when it comes to like context windows,
0:19:55 Google is currently the king, right?
0:19:59 Like the Gemini now has a one million token context window
0:20:01 and they said that’s jumping to two million tokens
0:20:02 within the year.
0:20:04 And just for, you know, context,
0:20:09 one million tokens is about 750,000 words input and output.
0:20:11 So the amount of words that you can put into the prompt
0:20:13 and get out is 750,000.
0:20:15 So they go to two million.
0:20:18 That’s 1.5 million words in and out context.
0:20:22 That’s enough to put in the entire Harry Potter book series
0:20:23 and ask questions about it.
0:20:24 – Yeah, talk about Steve Jobs again.
0:20:25 Like, you know, it’s like Steve Jobs
0:20:26 wouldn’t talk about all the specs, you know,
0:20:28 which now Apple’s doing that crap too.
0:20:29 Like Steve Jobs would talk about
0:20:31 how this is an amazing new thing.
0:20:32 He wouldn’t talk much about the specs.
0:20:34 You know, Google’s doing that same thing where like,
0:20:36 here’s the specs, it’s all these, you know,
0:20:38 here’s the context window, how big it is.
0:20:41 And versus open AI, not even talking about that stuff.
0:20:43 And I think it’s a trap because, you know,
0:20:45 one of my friends was at the open AI event
0:20:48 and the rumor, you know, around town in Silicon Valley
0:20:52 is that GPT-5 don’t really worry about the context window.
0:20:53 There’s like some major breakthrough
0:20:55 where I’m not sure if the context window
0:20:57 is gonna be dramatically larger,
0:20:58 but apparently something has changed
0:21:01 and they’re like that’s not gonna be much of an issue in GPT-5.
0:21:04 And so I think with like Google trying to play that game
0:21:06 of like we have the largest context window.
0:21:08 I mean, maybe GPT-5 is gonna have like 10 million
0:21:10 or 100 million or something ridiculous.
0:21:11 Maybe they’ve made a breakthrough there.
0:21:13 And so like, the Google’s gonna look pretty silly.
0:21:15 Like we’ve been playing this like stats game
0:21:17 where we have the best, you know, whatever
0:21:20 and the best specs and that’s not even,
0:21:21 they’re gonna lose in that game as well, I think.
0:21:24 – Yeah, I honestly think we’re gonna hit a point pretty soon
0:21:26 where people just don’t even talk about context windows.
0:21:28 This is gonna be like a non-issue.
0:21:29 They’re gonna be so big that the context window
0:21:33 is essentially more than you need, like always.
0:21:35 The other thing that they talked about
0:21:38 that was like pretty big news was their project Astra,
0:21:39 which I don’t remember.
0:21:40 I don’t know if you remember the demo,
0:21:43 but they walked around with like an iPhone out
0:21:45 and the iPhone was seeing things
0:21:48 and they were having a conversation with the phone
0:21:49 with what they were seeing, right?
0:21:51 The example they showed in their demo,
0:21:54 was the video footage went to a speaker
0:21:57 and it was like, let me know when you see something
0:21:58 that makes sound, right?
0:22:00 And it was like, oh, I see a speaker on the table.
0:22:02 And then they drew a little arrow on the camera
0:22:03 and said, what’s this part of the speaker called?
0:22:05 And it was like, oh, that’s called the tweeter, right?
0:22:07 And then they walked around the room
0:22:10 a little bit more with the camera
0:22:12 and then the girl in the demo was like,
0:22:14 do you remember where my glasses were?
0:22:16 And it actually remembered from earlier video footage,
0:22:19 yes, your glasses were on that other desk over by the Apple.
0:22:22 So she went over to where her glasses were,
0:22:24 put on the glasses and it was like this little Easter egg
0:22:27 moment where she put on glasses,
0:22:29 which looked like a hint at maybe a new version
0:22:31 of Google Glass or something, right?
0:22:34 ‘Cause she put on these glasses, sent her phone down
0:22:38 and then continued to have a conversation with the AI bot,
0:22:40 but now it was seeing what the glasses saw
0:22:43 and it also looked like it had a little tiny heads up display.
0:22:47 So whatever she was asking and whatever it was responding with,
0:22:50 it was almost like captioning it in front of her face
0:22:52 on the heads up display.
0:22:54 And Project Astra was actually one of the things
0:22:56 that they had demos available.
0:22:59 So I was actually able to demo Astra
0:23:00 while I was at Google IO.
0:23:03 And I mean, it worked pretty much the way
0:23:04 they showed it did in the demo.
0:23:08 It wasn’t like previous Google demos
0:23:10 where they made it look like real time,
0:23:11 but it wasn’t really real time.
0:23:14 It worked just like the demo.
0:23:16 And I thought that was actually pretty impressive.
0:23:17 I can see that being useful,
0:23:19 especially when it gets into like glasses and stuff.
0:23:21 – So since you really used it,
0:23:23 so I’m curious for like, from my perception was
0:23:27 that the open AI voice was more responsive than the Google,
0:23:29 but maybe the Google seemed to be taking more video input
0:23:31 ’cause it seemed to actually understand the surroundings
0:23:34 a bit better than the open AI demo.
0:23:35 So is that kind of like–
0:23:36 – There’s definitely more of a delay, right?
0:23:38 So when you were using the Google one,
0:23:39 you would ask it a question
0:23:41 and then there would be like a, you know,
0:23:45 one, two, three, four, it would respond.
0:23:48 The new version of chat GPT, it was like almost instant.
0:23:51 But if you pay attention,
0:23:55 it uses like filler words to make it feel more instant, right?
0:23:57 So what it would kind of do is you would ask it a question
0:24:00 and it would almost like repeat the question back to you.
0:24:03 And while it’s repeating the question back to you,
0:24:04 that’s where that sort of delay,
0:24:06 that latency would normally be.
0:24:08 So you would say like, “Hey, what am I looking at?”
0:24:11 And it would say, “Okay, let me take a look.”
0:24:12 It looks like you’re looking at
0:24:14 and then it would go into, you know, its response.
0:24:15 – Which is brilliant.
0:24:16 – So it was using that like filler.
0:24:17 – Yeah, it’s brilliant.
0:24:19 – But it makes it feel more natural
0:24:20 ’cause we talk like that too.
0:24:22 – Yeah, and you know, engineers,
0:24:23 there have been a few engineers on Twitter
0:24:25 that have been speculating that the reason
0:24:28 that the OpenAI model is so much more responsive
0:24:31 is because it actually is a breakthrough
0:24:32 in terms of being multimodal,
0:24:34 whereas Google is basically kind of stitching
0:24:36 these different models together behind the scenes.
0:24:37 You know, they got like a video model.
0:24:40 They got a, you know, text one, all this, right?
0:24:40 In audio.
0:24:42 And then OpenAI actually announced it.
0:24:43 They’re like, maybe people don’t realize it
0:24:45 ’cause we don’t really talk about the technical side
0:24:47 much on the stage, but yeah, that’s the big thing here.
0:24:49 It’s like, this is a model where literally
0:24:52 it’s doing all three at one time.
0:24:53 It’s not just doing text and then do,
0:24:55 it’s not doing text or doing image
0:24:57 and then like kind of like sending it to another model.
0:24:59 No, this is all happening within the same model.
0:25:00 That’s why it’s so fast.
0:25:01 And so I think that was the thing
0:25:03 that maybe a lot of people didn’t realize.
0:25:04 – Yeah, yeah, I mean, you mentioned a video model.
0:25:05 That was another thing that like,
0:25:07 we’re not gonna cover everything
0:25:08 that they covered at the Google event.
0:25:10 There was just way too much, right?
0:25:11 It was like two hours of announcements,
0:25:14 but another one that was kind of interesting, I guess,
0:25:16 was Veo or VEO, right?
0:25:17 Like they were basically acting like
0:25:19 this was their Sora competitor,
0:25:21 but if you put them side by side,
0:25:23 Sora’s still quite a bit better,
0:25:24 but Veo, I think they said that
0:25:27 can generate over one minute long videos
0:25:29 in 1080p resolution.
0:25:31 So, you know, it’s got the length
0:25:33 and the kind of quality going for it,
0:25:36 but it’s not like as realistic as what we saw out of Sora.
0:25:39 It was something where, you know, people saw it, right?
0:25:42 But everybody’s like first thing
0:25:44 is they’re gonna look at it and compare it to Sora, right?
0:25:45 Sora’s the best we’ve seen.
0:25:48 Like that set the bar for AI video.
0:25:50 And now you look at something like Veo
0:25:51 and you’re like, oh, that’s pretty cool.
0:25:53 But because we’ve already seen Sora,
0:25:55 it’s hard to be impressed by it.
0:25:57 – Yeah, that’s, and I wasn’t impressed.
0:25:59 And, you know, and Sora set the bar so high
0:26:01 that like it’s even,
0:26:03 it’s actually heard all the other AI video companies, right?
0:26:04 Like Runway and all the other ones, Pika.
0:26:07 ‘Cause like I used to share all these AI video threads
0:26:09 and they would get millions of views.
0:26:10 And then now people are just not interested.
0:26:12 Like you share it and like people are like,
0:26:13 this is like garbage.
0:26:15 – So hard to impress people now.
0:26:16 – So hard to impress people.
0:26:17 – Yeah, yeah.
0:26:19 Well, one last thing that I will say about Google
0:26:21 and I did sort of like a recap video
0:26:23 of my experience of Google I/O.
0:26:24 And I said this in that video,
0:26:28 is that going to events like the Google I/O event,
0:26:31 you know, it’s, it is really easy to look at Google.
0:26:34 Is this like big, huge corporate, you know,
0:26:37 faceless company that just wants to harvest
0:26:39 everybody’s data and get as much money
0:26:40 out of everybody as possible, right?
0:26:43 Like I think that’s the perception of the Googles
0:26:46 and the Microsofts and the metas of the world, right?
0:26:48 But when I went to events like this,
0:26:51 I got to meet so many of the people at Google, right?
0:26:53 I got to meet like the various engineers
0:26:54 and I got to meet, you know,
0:26:57 project leaders on different projects
0:27:00 that were working on a lot of this software.
0:27:03 And the thing that kind of really struck me
0:27:04 at this event was like,
0:27:06 these people are really, really excited
0:27:08 and really, really passionate about the tools
0:27:09 that they’re building.
0:27:12 And so even though Google itself is like this big,
0:27:15 faceless, massive mega corporation,
0:27:17 the individuals working on these projects
0:27:18 are really, really excited
0:27:20 and really passionate about what they do.
0:27:25 And it really sort of added to the humanity of Google for me
0:27:27 like actually getting to meet these people
0:27:28 and talk with them.
0:27:31 So, you know, that’s kind of the last little wrap up thing
0:27:34 that I wanted to say about Google is the people at Google
0:27:36 do care about the products that they’re developing.
0:27:39 They do care about what people think of these projects.
0:27:42 They do want to make them the best they can make them.
0:27:45 And that was the overwhelming sense
0:27:46 that I got at that event.
0:27:48 Now, fast forward to this week,
0:27:52 I’m actually at Microsoft Build out in Seattle right now.
0:27:53 Same kind of thing.
0:27:57 I’ve got to talk to the CTO of Microsoft.
0:27:59 I’ve gotten to talk to a whole bunch of project leaders
0:28:02 on different projects at Microsoft
0:28:04 and actually getting to talk to the humans
0:28:05 underneath these projects
0:28:07 that are actually building these projects
0:28:11 has been really interesting to me.
0:28:15 It really, really does make me feel more connected
0:28:16 to the company as a whole
0:28:19 just because I see the excitement of the people
0:28:21 that are building these things.
0:28:24 And so while at this Google event,
0:28:26 there was really two keynotes that happened.
0:28:28 There was the Monday keynote
0:28:31 which was like a co-pilot slash surface keynote
0:28:33 like the consumer facing products.
0:28:35 And then on Tuesday was Microsoft Build.
0:28:38 And Microsoft Build was the developer conference
0:28:40 which was really geared towards software engineers
0:28:42 and people building the software.
0:28:45 Not a whole lot of huge, exciting, crazy announcements
0:28:47 that came out of Microsoft Build
0:28:49 but the event that happened yesterday
0:28:50 was really, really interesting.
0:28:53 They announced a new feature called Recall
0:28:56 which is really interesting.
0:28:59 Basically what Recall does is it watches everything
0:29:02 that’s happening on your computer and it saves it all.
0:29:06 So it’s like the history inside of your Chrome Bowser
0:29:07 but for your entire computer.
0:29:09 So anything you built in Photoshop,
0:29:13 anything that you made with DaVinci Resolve,
0:29:16 any tool that you used on your computer,
0:29:19 it actually saved a history of all of that
0:29:21 and you can go back and find all of that history.
0:29:24 All of your browsing, all of your internet browsing,
0:29:26 it saves all of that as well.
0:29:28 It saves it all and you can go back through your history,
0:29:31 scrub through it and find very, very specific points
0:29:33 throughout your day.
0:29:34 And I thought that feature was pretty cool
0:29:37 but it obviously brings up some like security,
0:29:41 some privacy issues like oh, if it’s recording everything
0:29:45 I do on my computer, is Microsoft getting that data?
0:29:46 Where does that go?
0:29:49 What if there’s websites that I don’t want to be saved
0:29:52 into my storage and remembered that I visited?
0:29:55 Well, the cool thing about the Microsoft event
0:29:57 and one of the other announcements that they made
0:29:59 was that they’re actually developing new computers
0:30:01 to be AI first.
0:30:04 So a computer typically has a CPU and a GPU in it.
0:30:08 These new Microsoft computers have a CPU, a GPU
0:30:11 and now what they call an MPU, a neural processing unit.
0:30:15 So it’s like a separate card just to process the AI.
0:30:18 So all of the AI that it’s using to record your computer
0:30:21 and help you find what you did throughout the day,
0:30:24 that is all happening locally on your computer.
0:30:26 It’s not actually going out to the cloud
0:30:29 and Microsoft isn’t getting any of this data from you.
0:30:31 You can be not connected to the internet
0:30:33 and still save all this history.
0:30:36 Plus, they have the ability to go back through
0:30:38 and delete history that you don’t want in there.
0:30:40 So let’s say you visited a website
0:30:41 that you don’t want other people to visit.
0:30:43 Maybe you’re gift shopping or something.
0:30:45 Let’s just use that as an example.
0:30:47 I’m sure there’s other examples we can go with
0:30:49 but let’s use the example of gift shopping.
0:30:50 You don’t want somebody to come back,
0:30:52 look at the history of your computer
0:30:55 and see what gifts you were looking at for them.
0:30:57 So you can go and delete that from your history.
0:31:01 So it’s pretty secure, pretty private.
0:31:02 They say they never are gonna train
0:31:04 on any of your data that it’s saving.
0:31:09 And supposedly it all stays local right on your computer.
0:31:12 So that was one of the big announcements that they made.
0:31:13 Well, two of the big announcements that they made.
0:31:17 One was that they’re making their PCs AI first.
0:31:19 And the second one is that you’re gonna have
0:31:23 this recall feature built directly into Windows.
0:31:24 So everything you do on the computer,
0:31:26 there’s like a history of it that you can go back
0:31:31 and find really, really easily using your AI search.
0:31:33 The other big announcement they made at this event
0:31:37 was that co-pilot is kind of getting an overhaul.
0:31:38 If you use Windows right now,
0:31:41 you might notice co-pilot is like a little pop-up thing
0:31:44 that shows up on the right sidebar of your desktop.
0:31:45 Well, now it’s an actual app
0:31:48 that can actually be snapped to Windows
0:31:52 or float in front of other Windows.
0:31:54 And they showed off this really cool demo
0:31:56 where you can actually chat with the bot
0:31:59 and it will actually know what’s going on
0:32:01 with your computer and actually speak back to you.
0:32:03 So going all the way back to what we were talking about
0:32:05 at the beginning of this podcast,
0:32:10 OpenAI released a Mac app for ChatGPT
0:32:12 that’s going to have the voice in it.
0:32:15 And I think the reason that they didn’t release
0:32:16 the Windows app along with it
0:32:19 is because they knew Microsoft was going to release
0:32:21 this new version of co-pilot,
0:32:23 which kind of does all the same stuff.
0:32:26 That was the reason I think that OpenAI
0:32:27 didn’t announce a Windows app
0:32:29 because co-pilot does that stuff.
0:32:31 It can actually see what’s going on on your computer.
0:32:33 You can talk to it with your voice.
0:32:35 It will respond with a voice.
0:32:37 It does that stuff as well.
0:32:40 One of the really cool examples they showed off
0:32:42 was they had co-pilot open
0:32:44 and then somebody was playing Minecraft.
0:32:45 And supposedly they were playing Minecraft
0:32:46 for the first time.
0:32:47 They didn’t know how to play.
0:32:50 And they were having a conversation with the bot
0:32:54 and the bot was explaining to them how to play Minecraft
0:32:55 based on what it saw on their screen.
0:32:58 So it was like, they opened up their inventory
0:33:01 and asked the bot, “What do I build with this inventory?”
0:33:03 And it’s like, “Oh, you need to build an axe.”
0:33:04 And he’s like, “Okay, I’ll build an axe.”
0:33:07 And then it closes an inventory and the zombie pops up.
0:33:08 And he’s like, “What’s that? What do I do?”
0:33:10 And the AI was like, “Run, run, it’s a zombie.
0:33:11 Get out of here.”
0:33:14 So the AI was watching in real time
0:33:15 as he was playing this game
0:33:17 and actually giving feedback
0:33:20 and telling him what to do next in the game.
0:33:22 And that was also a really cool feature
0:33:25 that they showed off at the Microsoft event.
0:33:30 But it has been a whirlwind of AI announcements
0:33:31 over the last couple of weeks.
0:33:33 We’ve just been absolutely bombarded
0:33:36 between open AI, Google, and Microsoft.
0:33:41 And then in June, we also have a Cisco event coming up
0:33:42 where they’re gonna be talking
0:33:44 about how they’re leveraging AI for cybersecurity.
0:33:46 We have a Qualcomm event coming up,
0:33:50 which is starting to make a lot of the AI chips
0:33:51 that a lot of these products are using.
0:33:53 In fact, at the Microsoft event,
0:33:54 a lot of the Surface Pros are using
0:33:57 Snapdragon trips from Qualcomm.
0:33:59 So Qualcomm has a bunch of AI announcements coming up.
0:34:03 And then also in June, we have WWDC from Apple,
0:34:06 which is the Worldwide Developer Conference.
0:34:07 This is their big event
0:34:10 where they announce all of their big new features
0:34:11 that are coming out.
0:34:13 That’s last year, we got the announcement
0:34:14 of the Apple Vision Pro.
0:34:17 This year, we’re expecting a whole bunch of AI announcements,
0:34:19 possibly a new AI series.
0:34:22 But it has just been a crazy few weeks
0:34:25 and we have a few more crazy weeks coming up
0:34:28 with these additional events that are happening.
0:34:31 And hopefully, Nathan and I are gonna be able
0:34:33 to keep you looped in and chatting
0:34:35 about all of these upcoming events
0:34:38 and filling you in on all of the big announcements
0:34:39 that are happening.
0:34:41 So if you’re not subscribed to this podcast,
0:34:43 make sure you’re subscribed.
0:34:45 We put out new episodes every Tuesday
0:34:46 and every once in a while,
0:34:48 when we have really, really exciting events
0:34:49 happening in the AI world,
0:34:51 we’re gonna drop these bonus episodes
0:34:53 in between our normal episodes
0:34:55 to make sure that you’re looped in
0:34:59 and keeping your finger on the pulse of the latest news.
0:35:02 So that’s the goal, a lot of exciting things happening
0:35:05 and hopefully you’re here for it ’cause we’re here for it.
0:35:08 And thank you so much for tuning in to this episode today.
0:35:11 (upbeat music)
0:35:14 (upbeat music)
0:35:16 (upbeat music)
0:00:06 – Hey, welcome to the Next Wave podcast.
0:00:07 I’m Matt Wolf.
0:00:09 I’m here with my co-host, Nathan Lanz.
0:00:12 And in this show, it is our goal to keep you looped in
0:00:15 on all the latest AI news, all the coolest AI tools,
0:00:18 and just generally what’s going on in the AI world.
0:00:20 And the last couple of weeks
0:00:23 have been absolutely insane in the AI world.
0:00:24 There have been so many announcements.
0:00:28 We had Google I/O, we had OpenAI’s announcements.
0:00:30 We had two different keynotes from Microsoft,
0:00:32 and all of these announcements
0:00:34 had so much going on around AI.
0:00:38 So we thought we’d do a fun little bonus episode,
0:00:39 maybe a little bit shorter than normal,
0:00:41 and just kind of riff and talk about
0:00:43 some of these announcements and, you know,
0:00:45 get you looped in on what’s going on
0:00:46 in the world of AI right now.
0:00:50 So the first event that came up last week
0:00:52 was the OpenAI event, which is kind of funny,
0:00:55 because we knew Google I/O was happening
0:00:56 last week on Tuesday.
0:00:58 So what does OpenAI do?
0:01:00 They go and make their announcement on Monday,
0:01:03 the day before the big Google I/O event.
0:01:06 That was sort of the kickoff to this announcement fest
0:01:09 of all of these AI announcements from these big companies.
0:01:14 And of course, the big OpenAI announcement was GPT-40.
0:01:15 And during this announcement,
0:01:18 they showed off some demos of GPT-40,
0:01:22 being able to sort of have conversations with you,
0:01:24 the inflections of the voices change,
0:01:28 the voices sound eerily similar to Scarlett Johansson,
0:01:31 which is the voice of the AI in her.
0:01:34 So there’s some drama going around about that.
0:01:36 But I figured let’s kick off this conversation
0:01:38 talking about GPT-40.
0:01:44 When all your marketing team does is put out fires,
0:01:45 they burn out.
0:01:46 But with HubSpot,
0:01:49 they can achieve their best results without the stress.
0:01:52 Tapping to HubSpot’s collection of AI tools,
0:01:55 freezes to pinpoint leads, capture attention,
0:01:58 and access all your data in one place.
0:01:59 Keep your marketers cool
0:02:01 and your campaign results hotter than ever.
0:02:05 Visit hubspot.com/marketers to learn more.
0:02:11 – So, I mean, Nathan, like what are your initial thoughts?
0:02:13 When you watch that keynote, like,
0:02:15 were you blown away by it?
0:02:17 Were you underwhelmed by it?
0:02:19 Like what were your thoughts when you first watched it?
0:02:20 – I was honestly blown away by it.
0:02:21 Like just like, you know,
0:02:22 in one of our previous videos,
0:02:25 we talked about how I think in a year from now,
0:02:27 voice will be the main way that you interact with AI.
0:02:29 And it’s like, okay, well, now here it is.
0:02:32 Like they’re showing you can actually just talk to the AI.
0:02:33 You can actually even interrupt it.
0:02:35 It can be talking and you can just like interrupt it
0:02:36 and start talking again.
0:02:37 It’s not like, it’s not like the old one
0:02:41 where you had to like record and then like processes
0:02:43 and then it says something and you had to wait till it’s done.
0:02:45 No, you can just like talk back and forth with it.
0:02:47 And it wasn’t perfect,
0:02:50 but damn it was pretty good for like a first version
0:02:52 of like this kind of voice interact with AI.
0:02:54 But yeah, I was pretty blown away by it personally.
0:02:55 – You know, one of the big announcements
0:02:57 that came out with the GPT-40
0:03:00 is that pretty much the latest state of the art models
0:03:02 that, you know, the general consumers
0:03:04 are going to get access to,
0:03:05 they’re going to give them access free, right?
0:03:09 So like the GPT-40, people are going to get,
0:03:10 they’re going to get access to that
0:03:12 just in the free chat GPT now before, you know,
0:03:15 if you had the free chat GPT, you were using, you know,
0:03:17 GPT 3.5 and older model.
0:03:20 If you were on the chat GPT plus subscription,
0:03:23 you got chat, you got GPT-4 turbo.
0:03:25 The naming conventions are all, you know, weird
0:03:30 ’cause you got GPT-4, GPT-4 turbo, GPT-40, GPT-3.5.
0:03:33 Like it’s all confusing, but you know,
0:03:35 Sam Altman actually made a comment
0:03:36 in one of the recent interviews
0:03:38 that he’s probably going to stop, you know,
0:03:41 naming them GPT-4, GPT-5, GPT-6, right?
0:03:43 They’re going to just sort of land on a name
0:03:46 and just keep on improving on that version
0:03:47 over and over again.
0:03:49 But I thought it was really cool
0:03:51 that they’re basically saying our most state-of-the-art
0:03:53 model, the best models that we’re putting out,
0:03:55 you’re going to be able to use just for free now
0:03:57 in the free version of chat GPT.
0:03:59 I thought, I thought that was really, really cool.
0:04:01 And then if you’re a plus subscriber,
0:04:02 then you just get more use, right?
0:04:04 You get, you’re going to get some of the features
0:04:06 a little bit earlier than everybody else,
0:04:08 but you’re going to get like five times more
0:04:10 input/output with the model.
0:04:11 – I think it’s a big deal
0:04:13 because you know, most people when they think of AI,
0:04:15 they’re thinking of chat GPT
0:04:17 and they’re typically thinking of the free model, right?
0:04:20 Which has been 3.5, which you know,
0:04:22 compared to 4 is really, really bad.
0:04:24 And so the fact now that someone gets,
0:04:26 you know, I think 4.0 in some ways
0:04:28 is not as good as 4 in some ways,
0:04:31 but it’s pretty darn close and it’s way, way faster.
0:04:34 So for most people, that’s going to be such a, you know,
0:04:35 step forward in terms of what they believe
0:04:36 is possible with AI.
0:04:37 Like, and so I think that’s going to lead
0:04:39 to like a lot more awakening around like,
0:04:42 oh, AI is actually here, it’s really powerful.
0:04:44 But, you know, I think one thing worth thinking about though
0:04:46 is like, apparently they were training this model
0:04:49 with a separate team ever since 2022.
0:04:50 So my understanding is like, yeah,
0:04:53 this is kind of their state-of-the-art model,
0:04:55 but I don’t think it’s their like flagship model.
0:04:57 I think it’s just like another model
0:04:58 that they were developing.
0:05:00 And so they put it out there for free.
0:05:02 My theory is it’s probably built on like a better architecture
0:05:03 where it’s more cost efficient.
0:05:05 That’s why they’re able to give away for free.
0:05:07 That’s why it’s faster.
0:05:09 And then five is probably going to be built
0:05:10 on top of that as well.
0:05:12 So I think we’re still going to see GPT-5 quite soon.
0:05:14 And it may not be multimodal.
0:05:15 That’s my big question.
0:05:17 Like, okay, so if it’s a different model,
0:05:19 are we going to get something that’s dramatically smarter
0:05:21 with GPT-5, but it’s not multimodal yet.
0:05:22 It’s just text.
0:05:25 – Yeah, but this new model, GPT-4O was, you know,
0:05:27 pretty multimodal, right?
0:05:29 They were, they were inputting video,
0:05:31 although I think the video they were showing,
0:05:33 it was just sort of looking at screenshots from the video.
0:05:35 It wasn’t actually watching the video
0:05:37 because there was one moment in their demo
0:05:39 where he picked up the phone and like smiled
0:05:40 into the camera and it was like,
0:05:42 oh, I’m looking at something wood.
0:05:43 And he was like, no, no, that was before.
0:05:45 Now look at my face, right?
0:05:47 So it obviously sort of like took a snapshot
0:05:49 of the table or whatever.
0:05:51 And then when he looked at his face,
0:05:53 it still was like thinking about the picture
0:05:54 of the table or something.
0:05:56 So I don’t think it’s actually using video,
0:05:59 but it does seem to be ingesting audio.
0:06:01 You can put images into it.
0:06:04 So I mean, it’s pretty multimodal already.
0:06:05 I think the real question is,
0:06:08 are we going to get like actual video input with GPT-5?
0:06:11 – Yeah, I guess we’ll see.
0:06:12 – So there were some other announcements
0:06:13 during the OpenAI event.
0:06:15 So I’ll just kind of rattle off some of them.
0:06:17 I’ve got my notes here in front of me.
0:06:19 They talked about a new desktop app,
0:06:22 which it’s pretty cool.
0:06:24 It’s only available on Mac right now.
0:06:27 They didn’t make it available on PC right away.
0:06:30 And I think I’m at the Microsoft event right now.
0:06:31 If you’re watching on video,
0:06:33 you can see I’ve got a different background than normal.
0:06:37 I think we found out why they didn’t release a Mac version
0:06:38 while I’m at the Microsoft event,
0:06:39 but we’ll get into that
0:06:42 when we start talking about the Microsoft event here.
0:06:44 But the desktop app, it’s pretty cool.
0:06:45 I tested it out on my Mac.
0:06:47 It doesn’t have the voice features in it yet.
0:06:49 It doesn’t have the feature
0:06:51 where you can view your desktop yet.
0:06:54 You can take a screenshot, drag the image in,
0:06:57 and chat with the screenshot essentially,
0:06:58 but it doesn’t just look at your desktop
0:07:00 like they showed in the demo yet.
0:07:02 It’s pretty much the same thing
0:07:04 you get inside of the chat GPT web app,
0:07:05 just on your desktop.
0:07:10 Now we also learned that that GPT-2 chatbot
0:07:12 that was all over Limsis on the chatbot arena,
0:07:15 you know, apparently that was GPT-4O.
0:07:17 That was them kind of testing that out
0:07:20 to see how people would work with it.
0:07:23 And you know, the other big thing was like
0:07:24 the voices in it, right?
0:07:28 Like the GPT-4O, it feels like a marginal improvement
0:07:31 with like the intelligence of it, in my opinion, right?
0:07:33 It doesn’t seem like huge leaps
0:07:34 above what we were getting before.
0:07:36 Other than in coding, it seems to work really well
0:07:39 with coding, but the like the voice stuff
0:07:41 was I think the thing that most people
0:07:43 were going nuts about from that keynote, right?
0:07:46 When you speak to it, you can tell it
0:07:49 to give you more emotion, get more excited about it,
0:07:52 talk like a robot, do all that sort of stuff.
0:07:53 And then when they were demoing it,
0:07:55 that’s also where we heard the voice,
0:07:58 which sounded very similar to Scarlett Johansson.
0:08:00 But I think that was the thing
0:08:03 that people really went nuts over was that voice thing.
0:08:05 And like you mentioned earlier in this episode,
0:08:07 we did a whole episode about how like
0:08:09 voice is that next thing, right?
0:08:12 Voice is what’s going to be the sort of input
0:08:15 of all of these large language models.
0:08:17 And when chat GPT showed off that voice
0:08:19 and there was really like no latency,
0:08:21 they would ask a question, it would respond
0:08:22 almost in the same amount of time,
0:08:23 a real human would respond.
0:08:26 You know, that was the thing that was probably
0:08:29 the biggest sort of differentiator about,
0:08:31 you know, that from what came before it.
0:08:32 – Yeah, I mean, I tweeted about the whole
0:08:34 Scarlett Johansson thing yesterday
0:08:35 and it got a lot of attention,
0:08:37 especially on LinkedIn for some reason.
0:08:38 People, you know, seem pretty upset about it.
0:08:40 You know, I have mixed feelings.
0:08:41 I mean, I guess to give people context,
0:08:44 like Scarlett Johansson put out a statement
0:08:45 or I’m not sure if it was officially from her,
0:08:48 but her publicist, you know, said that it really happened
0:08:50 where openly I contacted Scarlett Johansson,
0:08:52 asked like, can we use your voice?
0:08:53 She’s like, no, you can’t.
0:08:57 And apparently they kept trying and she just refused.
0:09:00 And they kind of picked a voice that seemed similar to hers.
0:09:01 And then she’s saying, not only that,
0:09:03 but then like they threw it in my face
0:09:04 ’cause like Sam Altman like tweeted to somebody
0:09:06 like the word her, you know,
0:09:09 and like referencing the movie where she’s like the,
0:09:11 you know, the main voice for the AI system.
0:09:12 I have mixed feelings about it.
0:09:14 ‘Cause like when I, when you actually,
0:09:16 somebody else shared a, you know,
0:09:18 side-by-side comparison of the voices, you know,
0:09:20 like here’s one, here’s the sky voice,
0:09:23 the open AI voice, and then here’s Scarlett Johansson’s voice.
0:09:24 You know, they’re quite different.
0:09:26 I mean, like they don’t sound like,
0:09:28 sure it sounds like an attractive woman talking
0:09:29 or something, but it doesn’t sound
0:09:31 really like Scarlett Johansson.
0:09:33 And then like, so can she really like,
0:09:34 what does she own?
0:09:37 Like AI, like feminine AI voices forever, right?
0:09:39 Like that’s kind of, you know,
0:09:40 that’s kind of like a wild statement to make
0:09:42 that you own that, but they, but who knows?
0:09:45 They’ll probably still end up having to settle with her
0:09:48 for the fact that like he said her, if I had to guess.
0:09:49 – Yeah, yeah, that’s true.
0:09:51 I mean, open AI did put out their own statement
0:09:52 around what was going on.
0:09:54 They said that they hired a different voice actor,
0:09:55 not Scarlett Johansson.
0:09:57 They trained on hours and hours
0:10:01 of this other voice actors work to get this in.
0:10:03 But from what I understand,
0:10:05 they actually are not planning on launching
0:10:07 with that sky voice now that, you know,
0:10:08 just to avoid all the controversy
0:10:10 and all the issues that might come up.
0:10:13 I think that, you know, right now they’re just saying,
0:10:14 we’re just not going to launch with that voice.
0:10:16 You know, we don’t want to deal with all that.
0:10:18 So we’ll see how that plays out.
0:10:20 So HubSpot just put out this really cool checklist.
0:10:22 It’s called the AI adoption checklist.
0:10:25 And if you’re trying to implement AI into your business
0:10:29 in any way whatsoever, this checklist has you covered
0:10:33 from security to privacy to planning to training.
0:10:34 This has you covered.
0:10:36 Click the link in the description below
0:10:38 and grab the checklist today.
0:10:40 – Yeah, so then the very next day
0:10:41 after this open AI keynote
0:10:43 where they made those announcements
0:10:45 was the Google IO event
0:10:48 where that’s Google’s developer conference
0:10:50 where they typically make all of their announcements.
0:10:53 And this year it was once again, all about AI.
0:10:55 In fact, at the end of the conference,
0:10:58 Sundar, the CEO of Google actually counted up
0:11:00 how many times they said AI.
0:11:02 And it was like 121 times
0:11:04 or something throughout the presentation.
0:11:06 One thing that I find interesting
0:11:09 between the two keynote presentations
0:11:12 was that I feel like ChatGPT and open AI,
0:11:16 they really sort of like honed in on like the one thing,
0:11:17 right?
0:11:19 They really sort of honed in on like the voice feature
0:11:22 and the chatting and the conversational element of it
0:11:25 and the tone and the intonations of the voice.
0:11:28 It was very, very focused on that thing.
0:11:31 The Google event in contrast to it was like,
0:11:33 here’s an announcement, here’s an announcement,
0:11:35 here’s an announcement, here’s an announcement.
0:11:36 It was just two hours of an announcement
0:11:39 after announcement, after announcement.
0:11:40 I was at the event
0:11:43 and I’m actually kind of struggling right now
0:11:46 to remember what were all the announcements Google made, right?
0:11:51 So like ChatGPT or open AI had like one big announcement
0:11:54 that everybody remembers, everybody talked about.
0:11:57 Google just like bombarded us with announcements
0:11:59 but they were all kind of so marginal
0:12:01 that I almost forget them.
0:12:03 Like I know I’ve got my notes in front of me,
0:12:06 but like there’s nothing that stands out in my mind
0:12:10 of like Google talked about this and that is game changing.
0:12:12 But I’m curious, Nathan, like from you,
0:12:14 was there anything from the Google keynote
0:12:18 that specifically stood out that was like,
0:12:19 okay, wow, that’s something that Google did
0:12:21 that I thought was impressive?
0:12:23 – No, I mean, I think it was like information overload.
0:12:25 Like they put out so many things
0:12:29 and if you look at the open AI event in comparison,
0:12:32 it was way more like Steve Jobs like, right?
0:12:34 Where there’s like, here’s this one thing that we have
0:12:37 and they just demonstrate in a beautiful way.
0:12:39 And I think open AI did a really good job too
0:12:40 where they even humanize it more
0:12:42 by having all these different team members
0:12:45 show the same thing in different ways.
0:12:49 Here’s the AI voice, our CTO co-founders showing it to you,
0:12:51 presenting it to you, and then here’s people from the team
0:12:53 and here’s how they’re using it.
0:12:55 And they showed like these six, seven different use cases
0:12:58 all with the same product, the same AI voice.
0:12:59 And I thought that was such a great way
0:13:02 to like reiterate the one thing over and over and over.
0:13:04 So by the end of it, it was like really ingrained
0:13:06 in your head like, oh, they’ve done an amazing job
0:13:08 with AI voice, whereas with, you know, with Google,
0:13:11 it’s, they release so many things and it’s like,
0:13:13 it’s not clear when a lot of it’s actually gonna come out.
0:13:15 That’s really hard to like pinpoint like,
0:13:16 oh yeah, what’s the one thing?
0:13:17 And I was like, it sounds like, oh yeah,
0:13:18 they made a lot of progress.
0:13:20 There’s a lot of cool things they did.
0:13:21 You know, like I put out Twitter threads
0:13:23 about both companies’ announcements.
0:13:26 And, you know, my Twitter thread about OpenAI
0:13:28 got like 5.9 million views.
0:13:30 And my tweet thread about Google,
0:13:33 I think it got maybe like 50,000 views just in comparisons.
0:13:36 It kind of showed you the general interest level
0:13:39 in OpenAI’s announcement versus Google’s announcement.
0:13:40 – Yeah, one thing that I found like really interesting
0:13:43 about the OpenAI event was that they had so much more
0:13:45 that they could have talked about as well, right?
0:13:48 Like they put out a blog post about all of the stuff
0:13:51 that they were rolling out with their new rollout.
0:13:54 This GPT-40 is its own image generator.
0:13:55 It’s not using Dali.
0:13:58 It’s like using its own built-in image generator.
0:14:00 It can do text to like 3D object.
0:14:03 It’s got like a ton of features on their blog post.
0:14:06 But in that keynote, they focused in on like the one thing
0:14:09 that I thought maybe they felt was gonna have
0:14:12 the most impact on people from watching that event.
0:14:13 So I thought that was really interesting
0:14:16 that they just, they could have themselves
0:14:18 done a two-hour keynote where they bombarded us
0:14:21 with OpenAI announcements, but they chose to focus
0:14:23 on just like one or two key things.
0:14:25 So I felt that was really, really interesting as well.
0:14:28 – Yeah, and I thought it was fascinating too
0:14:31 that, you know, Sam Altman didn’t appear.
0:14:33 And so for me, that’s like, okay, he didn’t appear
0:14:36 because he’s going to appear when GPT-5 comes out
0:14:37 in like three months.
0:14:39 – I’m at the Microsoft event now.
0:14:41 He was here. He was part of that keynote.
0:14:43 But we’ll get to Microsoft in a minute.
0:14:43 He’s like…
0:14:47 We’ll be right back.
0:14:49 But first, I want to tell you about another great podcast
0:14:50 you’re going to want to listen to.
0:14:54 It’s called Science of Scaling, hosted by Mark Roberge.
0:14:56 And it’s brought to you by the HubSpot Podcast Network,
0:15:00 the audio destination for business professionals.
0:15:02 Each week, host Mark Roberge,
0:15:04 founding chief revenue officer at HubSpot,
0:15:07 senior lecturer at Harvard Business School
0:15:09 and co-founder of Stage 2 Capital,
0:15:12 sits down with the most successful sales leaders in tech
0:15:15 to learn the secrets, strategies, and tactics
0:15:17 to scaling your company’s growth.
0:15:20 He recently did a great episode called How Do You Sol
0:15:22 for a Siloed Marketing and Sales?
0:15:24 And I personally learned a lot from it.
0:15:26 You’re going to want to check out the podcast,
0:15:27 listen to Science of Scaling
0:15:30 wherever you get your podcasts.
0:15:34 – I do want to talk about a few more
0:15:35 of the Google things that they announced
0:15:37 ’cause there was something I thought was cool
0:15:38 while I was there.
0:15:41 But they’re also like, if I’m being honest,
0:15:43 they’re kind of like, they’re sort of gimmicky
0:15:46 and they’re kind of like almost forgettable,
0:15:47 but they’re still kind of cool, right?
0:15:50 Like they showed off this feature called Ask Photos
0:15:54 where you can basically use like an AI search
0:15:55 to search through your photos.
0:15:56 And some of the examples they gave was like,
0:15:58 when did my daughter learn how to swim?
0:16:00 And it would look through all of your photos,
0:16:02 all of the timestamps, try to find photos
0:16:04 where your daughter is swimming
0:16:07 and then tell you your daughter learned to swim
0:16:09 based on all of your photos, right?
0:16:13 So it uses the photos as context to answer your questions.
0:16:15 That I thought was pretty cool.
0:16:16 Like I can see that being really useful
0:16:19 if you have a giant database of photos on your computer
0:16:21 with like thousands of photos
0:16:24 and you’re trying to find like one specific thing
0:16:26 or maybe you’re making a video online
0:16:29 and you need like some B-roll for your video
0:16:31 and you want the B-roll to just be like shots
0:16:33 of the ocean or something.
0:16:35 You could go in there and let the AI find
0:16:37 like just the shots that you’re looking for.
0:16:40 So I thought Ask Photos was a pretty cool feature.
0:16:42 – When I saw that demo, I was like, oh, that was amazing.
0:16:44 They probably should have made that better
0:16:46 and really focus on something like that
0:16:48 ’cause it’s like, they announced like 10 things.
0:16:50 And most of it seemed like they were just kind of like
0:16:52 tacking on AI onto existing products, right?
0:16:53 It’s like a whole joke about like–
0:16:54 – I think a lot of it was.
0:16:56 – If you see like the little three stars,
0:16:58 that means you’re just like adding AI on
0:17:00 for like the sake of it, right?
0:17:03 And it felt like a lot of their stuff was that.
0:17:04 And there was this engineer who made a statement
0:17:06 about that the other day.
0:17:08 Apparently he worked at Google for like 20 something years.
0:17:10 He was a pretty, you know, top engineer at Google.
0:17:12 And he said, they’re doing the same (beep)
0:17:14 they did with Google Plus when like Facebook came out
0:17:16 where like, they’re like panicking.
0:17:17 They had to create something else and they’re just like,
0:17:20 start, okay, we have to have a social network.
0:17:22 Well, let’s just put all Google’s existing stuff
0:17:24 into a social network and like it just doesn’t work.
0:17:28 And you know, they’re also changing search too,
0:17:29 which I mean, is a big deal.
0:17:31 You know, where they’re like using AI
0:17:33 to like analyze the results of a lot of search terms.
0:17:36 I guess it’s only like 1% or so of searches
0:17:37 are currently using that feature,
0:17:40 but it rolled out like two days ago and you know,
0:17:42 it’s apparently some people in SEO land
0:17:45 are already seeing like 20% drop in traffic from that.
0:17:46 Like in the last week.
0:17:50 And so it’s, you know, they’re in a tough situation
0:17:51 ’cause like all the stuff they released,
0:17:54 they’re gonna like cause issues
0:17:56 if they just start dramatically changing Google search,
0:17:57 you know.
0:17:58 – Well, I mean, their revenue is all from,
0:18:00 from ad revenue from search, right?
0:18:02 Like that’s the biggest portion of the revenue
0:18:03 for the business.
0:18:06 So if people stop clicking on those search ads,
0:18:08 well, what happens to Google’s business?
0:18:11 I’ve actually noticed that pop up a lot more often.
0:18:14 Like almost every search I’ve done in the last like a week,
0:18:15 I don’t search Google a lot anymore,
0:18:18 but like I would say almost all of the searches
0:18:19 I’ve done in the last week,
0:18:23 I’ve seen that AI sort of summary thing come up at the top.
0:18:24 I still end up clicking on the links though.
0:18:27 Like usually if I’m Googling something,
0:18:28 I’m trying to find a website.
0:18:30 I’m not trying to find an answer to a question
0:18:32 because if I’m trying to find an answer to a question,
0:18:35 I’m just going straight to chat GPT these days, you know,
0:18:36 or Claude or something like that.
0:18:38 – You know, people are making a good point too.
0:18:39 It’s like almost like Google is like violating
0:18:41 this social contract that always existed
0:18:43 with the open web, right?
0:18:46 Where it’s like, hey, we’re like searching all your stuff.
0:18:49 We are taking your data and we’re making money off of it,
0:18:51 but we’re going to also send people to you.
0:18:53 And so that’s always been this kind of like social
0:18:54 contract that existed.
0:18:57 So now if they are taking in people’s data,
0:18:59 but then they’re just giving the answer,
0:19:00 how are they, and then they’re like,
0:19:02 how could they possibly criticize chat GPT?
0:19:04 (laughing)
0:19:08 How could they criticize chat GPT if open AI actually did
0:19:10 crawl all of YouTube’s data, like it’s, you know,
0:19:12 the rumor that they did.
0:19:13 I think they have a hard argument there.
0:19:14 That’s probably why they haven’t like really
0:19:17 went after it yet, because like they’re doing similar.
0:19:20 – Google’s a web crawler scraping company, right?
0:19:23 Like it scrapes the entire web’s data also.
0:19:26 The only difference is like if when Google scrapes the web,
0:19:28 it’s actually beneficial to the creators
0:19:30 and when chat GPT scrapes the web,
0:19:32 it sort of disincentivizes creators, right?
0:19:35 There’s definitely a big differentiator there.
0:19:39 – For now, for now, but it’s changing almost every month
0:19:39 right now.
0:19:41 So it’s like every month it’s like, you know,
0:19:43 people are going less to websites,
0:19:44 they’re just getting the answer from Google.
0:19:48 So yeah, I think Google’s in a very tough situation there.
0:19:49 And I don’t know how they’re going to get out of it.
0:19:51 – Yeah, well, the other thing is, you know,
0:19:53 when it comes to like context windows,
0:19:55 Google is currently the king, right?
0:19:59 Like the Gemini now has a one million token context window
0:20:01 and they said that’s jumping to two million tokens
0:20:02 within the year.
0:20:04 And just for, you know, context,
0:20:09 one million tokens is about 750,000 words input and output.
0:20:11 So the amount of words that you can put into the prompt
0:20:13 and get out is 750,000.
0:20:15 So they go to two million.
0:20:18 That’s 1.5 million words in and out context.
0:20:22 That’s enough to put in the entire Harry Potter book series
0:20:23 and ask questions about it.
0:20:24 – Yeah, talk about Steve Jobs again.
0:20:25 Like, you know, it’s like Steve Jobs
0:20:26 wouldn’t talk about all the specs, you know,
0:20:28 which now Apple’s doing that crap too.
0:20:29 Like Steve Jobs would talk about
0:20:31 how this is an amazing new thing.
0:20:32 He wouldn’t talk much about the specs.
0:20:34 You know, Google’s doing that same thing where like,
0:20:36 here’s the specs, it’s all these, you know,
0:20:38 here’s the context window, how big it is.
0:20:41 And versus open AI, not even talking about that stuff.
0:20:43 And I think it’s a trap because, you know,
0:20:45 one of my friends was at the open AI event
0:20:48 and the rumor, you know, around town in Silicon Valley
0:20:52 is that GPT-5 don’t really worry about the context window.
0:20:53 There’s like some major breakthrough
0:20:55 where I’m not sure if the context window
0:20:57 is gonna be dramatically larger,
0:20:58 but apparently something has changed
0:21:01 and they’re like that’s not gonna be much of an issue in GPT-5.
0:21:04 And so I think with like Google trying to play that game
0:21:06 of like we have the largest context window.
0:21:08 I mean, maybe GPT-5 is gonna have like 10 million
0:21:10 or 100 million or something ridiculous.
0:21:11 Maybe they’ve made a breakthrough there.
0:21:13 And so like, the Google’s gonna look pretty silly.
0:21:15 Like we’ve been playing this like stats game
0:21:17 where we have the best, you know, whatever
0:21:20 and the best specs and that’s not even,
0:21:21 they’re gonna lose in that game as well, I think.
0:21:24 – Yeah, I honestly think we’re gonna hit a point pretty soon
0:21:26 where people just don’t even talk about context windows.
0:21:28 This is gonna be like a non-issue.
0:21:29 They’re gonna be so big that the context window
0:21:33 is essentially more than you need, like always.
0:21:35 The other thing that they talked about
0:21:38 that was like pretty big news was their project Astra,
0:21:39 which I don’t remember.
0:21:40 I don’t know if you remember the demo,
0:21:43 but they walked around with like an iPhone out
0:21:45 and the iPhone was seeing things
0:21:48 and they were having a conversation with the phone
0:21:49 with what they were seeing, right?
0:21:51 The example they showed in their demo,
0:21:54 was the video footage went to a speaker
0:21:57 and it was like, let me know when you see something
0:21:58 that makes sound, right?
0:22:00 And it was like, oh, I see a speaker on the table.
0:22:02 And then they drew a little arrow on the camera
0:22:03 and said, what’s this part of the speaker called?
0:22:05 And it was like, oh, that’s called the tweeter, right?
0:22:07 And then they walked around the room
0:22:10 a little bit more with the camera
0:22:12 and then the girl in the demo was like,
0:22:14 do you remember where my glasses were?
0:22:16 And it actually remembered from earlier video footage,
0:22:19 yes, your glasses were on that other desk over by the Apple.
0:22:22 So she went over to where her glasses were,
0:22:24 put on the glasses and it was like this little Easter egg
0:22:27 moment where she put on glasses,
0:22:29 which looked like a hint at maybe a new version
0:22:31 of Google Glass or something, right?
0:22:34 ‘Cause she put on these glasses, sent her phone down
0:22:38 and then continued to have a conversation with the AI bot,
0:22:40 but now it was seeing what the glasses saw
0:22:43 and it also looked like it had a little tiny heads up display.
0:22:47 So whatever she was asking and whatever it was responding with,
0:22:50 it was almost like captioning it in front of her face
0:22:52 on the heads up display.
0:22:54 And Project Astra was actually one of the things
0:22:56 that they had demos available.
0:22:59 So I was actually able to demo Astra
0:23:00 while I was at Google IO.
0:23:03 And I mean, it worked pretty much the way
0:23:04 they showed it did in the demo.
0:23:08 It wasn’t like previous Google demos
0:23:10 where they made it look like real time,
0:23:11 but it wasn’t really real time.
0:23:14 It worked just like the demo.
0:23:16 And I thought that was actually pretty impressive.
0:23:17 I can see that being useful,
0:23:19 especially when it gets into like glasses and stuff.
0:23:21 – So since you really used it,
0:23:23 so I’m curious for like, from my perception was
0:23:27 that the open AI voice was more responsive than the Google,
0:23:29 but maybe the Google seemed to be taking more video input
0:23:31 ’cause it seemed to actually understand the surroundings
0:23:34 a bit better than the open AI demo.
0:23:35 So is that kind of like–
0:23:36 – There’s definitely more of a delay, right?
0:23:38 So when you were using the Google one,
0:23:39 you would ask it a question
0:23:41 and then there would be like a, you know,
0:23:45 one, two, three, four, it would respond.
0:23:48 The new version of chat GPT, it was like almost instant.
0:23:51 But if you pay attention,
0:23:55 it uses like filler words to make it feel more instant, right?
0:23:57 So what it would kind of do is you would ask it a question
0:24:00 and it would almost like repeat the question back to you.
0:24:03 And while it’s repeating the question back to you,
0:24:04 that’s where that sort of delay,
0:24:06 that latency would normally be.
0:24:08 So you would say like, “Hey, what am I looking at?”
0:24:11 And it would say, “Okay, let me take a look.”
0:24:12 It looks like you’re looking at
0:24:14 and then it would go into, you know, its response.
0:24:15 – Which is brilliant.
0:24:16 – So it was using that like filler.
0:24:17 – Yeah, it’s brilliant.
0:24:19 – But it makes it feel more natural
0:24:20 ’cause we talk like that too.
0:24:22 – Yeah, and you know, engineers,
0:24:23 there have been a few engineers on Twitter
0:24:25 that have been speculating that the reason
0:24:28 that the OpenAI model is so much more responsive
0:24:31 is because it actually is a breakthrough
0:24:32 in terms of being multimodal,
0:24:34 whereas Google is basically kind of stitching
0:24:36 these different models together behind the scenes.
0:24:37 You know, they got like a video model.
0:24:40 They got a, you know, text one, all this, right?
0:24:40 In audio.
0:24:42 And then OpenAI actually announced it.
0:24:43 They’re like, maybe people don’t realize it
0:24:45 ’cause we don’t really talk about the technical side
0:24:47 much on the stage, but yeah, that’s the big thing here.
0:24:49 It’s like, this is a model where literally
0:24:52 it’s doing all three at one time.
0:24:53 It’s not just doing text and then do,
0:24:55 it’s not doing text or doing image
0:24:57 and then like kind of like sending it to another model.
0:24:59 No, this is all happening within the same model.
0:25:00 That’s why it’s so fast.
0:25:01 And so I think that was the thing
0:25:03 that maybe a lot of people didn’t realize.
0:25:04 – Yeah, yeah, I mean, you mentioned a video model.
0:25:05 That was another thing that like,
0:25:07 we’re not gonna cover everything
0:25:08 that they covered at the Google event.
0:25:10 There was just way too much, right?
0:25:11 It was like two hours of announcements,
0:25:14 but another one that was kind of interesting, I guess,
0:25:16 was Veo or VEO, right?
0:25:17 Like they were basically acting like
0:25:19 this was their Sora competitor,
0:25:21 but if you put them side by side,
0:25:23 Sora’s still quite a bit better,
0:25:24 but Veo, I think they said that
0:25:27 can generate over one minute long videos
0:25:29 in 1080p resolution.
0:25:31 So, you know, it’s got the length
0:25:33 and the kind of quality going for it,
0:25:36 but it’s not like as realistic as what we saw out of Sora.
0:25:39 It was something where, you know, people saw it, right?
0:25:42 But everybody’s like first thing
0:25:44 is they’re gonna look at it and compare it to Sora, right?
0:25:45 Sora’s the best we’ve seen.
0:25:48 Like that set the bar for AI video.
0:25:50 And now you look at something like Veo
0:25:51 and you’re like, oh, that’s pretty cool.
0:25:53 But because we’ve already seen Sora,
0:25:55 it’s hard to be impressed by it.
0:25:57 – Yeah, that’s, and I wasn’t impressed.
0:25:59 And, you know, and Sora set the bar so high
0:26:01 that like it’s even,
0:26:03 it’s actually heard all the other AI video companies, right?
0:26:04 Like Runway and all the other ones, Pika.
0:26:07 ‘Cause like I used to share all these AI video threads
0:26:09 and they would get millions of views.
0:26:10 And then now people are just not interested.
0:26:12 Like you share it and like people are like,
0:26:13 this is like garbage.
0:26:15 – So hard to impress people now.
0:26:16 – So hard to impress people.
0:26:17 – Yeah, yeah.
0:26:19 Well, one last thing that I will say about Google
0:26:21 and I did sort of like a recap video
0:26:23 of my experience of Google I/O.
0:26:24 And I said this in that video,
0:26:28 is that going to events like the Google I/O event,
0:26:31 you know, it’s, it is really easy to look at Google.
0:26:34 Is this like big, huge corporate, you know,
0:26:37 faceless company that just wants to harvest
0:26:39 everybody’s data and get as much money
0:26:40 out of everybody as possible, right?
0:26:43 Like I think that’s the perception of the Googles
0:26:46 and the Microsofts and the metas of the world, right?
0:26:48 But when I went to events like this,
0:26:51 I got to meet so many of the people at Google, right?
0:26:53 I got to meet like the various engineers
0:26:54 and I got to meet, you know,
0:26:57 project leaders on different projects
0:27:00 that were working on a lot of this software.
0:27:03 And the thing that kind of really struck me
0:27:04 at this event was like,
0:27:06 these people are really, really excited
0:27:08 and really, really passionate about the tools
0:27:09 that they’re building.
0:27:12 And so even though Google itself is like this big,
0:27:15 faceless, massive mega corporation,
0:27:17 the individuals working on these projects
0:27:18 are really, really excited
0:27:20 and really passionate about what they do.
0:27:25 And it really sort of added to the humanity of Google for me
0:27:27 like actually getting to meet these people
0:27:28 and talk with them.
0:27:31 So, you know, that’s kind of the last little wrap up thing
0:27:34 that I wanted to say about Google is the people at Google
0:27:36 do care about the products that they’re developing.
0:27:39 They do care about what people think of these projects.
0:27:42 They do want to make them the best they can make them.
0:27:45 And that was the overwhelming sense
0:27:46 that I got at that event.
0:27:48 Now, fast forward to this week,
0:27:52 I’m actually at Microsoft Build out in Seattle right now.
0:27:53 Same kind of thing.
0:27:57 I’ve got to talk to the CTO of Microsoft.
0:27:59 I’ve gotten to talk to a whole bunch of project leaders
0:28:02 on different projects at Microsoft
0:28:04 and actually getting to talk to the humans
0:28:05 underneath these projects
0:28:07 that are actually building these projects
0:28:11 has been really interesting to me.
0:28:15 It really, really does make me feel more connected
0:28:16 to the company as a whole
0:28:19 just because I see the excitement of the people
0:28:21 that are building these things.
0:28:24 And so while at this Google event,
0:28:26 there was really two keynotes that happened.
0:28:28 There was the Monday keynote
0:28:31 which was like a co-pilot slash surface keynote
0:28:33 like the consumer facing products.
0:28:35 And then on Tuesday was Microsoft Build.
0:28:38 And Microsoft Build was the developer conference
0:28:40 which was really geared towards software engineers
0:28:42 and people building the software.
0:28:45 Not a whole lot of huge, exciting, crazy announcements
0:28:47 that came out of Microsoft Build
0:28:49 but the event that happened yesterday
0:28:50 was really, really interesting.
0:28:53 They announced a new feature called Recall
0:28:56 which is really interesting.
0:28:59 Basically what Recall does is it watches everything
0:29:02 that’s happening on your computer and it saves it all.
0:29:06 So it’s like the history inside of your Chrome Bowser
0:29:07 but for your entire computer.
0:29:09 So anything you built in Photoshop,
0:29:13 anything that you made with DaVinci Resolve,
0:29:16 any tool that you used on your computer,
0:29:19 it actually saved a history of all of that
0:29:21 and you can go back and find all of that history.
0:29:24 All of your browsing, all of your internet browsing,
0:29:26 it saves all of that as well.
0:29:28 It saves it all and you can go back through your history,
0:29:31 scrub through it and find very, very specific points
0:29:33 throughout your day.
0:29:34 And I thought that feature was pretty cool
0:29:37 but it obviously brings up some like security,
0:29:41 some privacy issues like oh, if it’s recording everything
0:29:45 I do on my computer, is Microsoft getting that data?
0:29:46 Where does that go?
0:29:49 What if there’s websites that I don’t want to be saved
0:29:52 into my storage and remembered that I visited?
0:29:55 Well, the cool thing about the Microsoft event
0:29:57 and one of the other announcements that they made
0:29:59 was that they’re actually developing new computers
0:30:01 to be AI first.
0:30:04 So a computer typically has a CPU and a GPU in it.
0:30:08 These new Microsoft computers have a CPU, a GPU
0:30:11 and now what they call an MPU, a neural processing unit.
0:30:15 So it’s like a separate card just to process the AI.
0:30:18 So all of the AI that it’s using to record your computer
0:30:21 and help you find what you did throughout the day,
0:30:24 that is all happening locally on your computer.
0:30:26 It’s not actually going out to the cloud
0:30:29 and Microsoft isn’t getting any of this data from you.
0:30:31 You can be not connected to the internet
0:30:33 and still save all this history.
0:30:36 Plus, they have the ability to go back through
0:30:38 and delete history that you don’t want in there.
0:30:40 So let’s say you visited a website
0:30:41 that you don’t want other people to visit.
0:30:43 Maybe you’re gift shopping or something.
0:30:45 Let’s just use that as an example.
0:30:47 I’m sure there’s other examples we can go with
0:30:49 but let’s use the example of gift shopping.
0:30:50 You don’t want somebody to come back,
0:30:52 look at the history of your computer
0:30:55 and see what gifts you were looking at for them.
0:30:57 So you can go and delete that from your history.
0:31:01 So it’s pretty secure, pretty private.
0:31:02 They say they never are gonna train
0:31:04 on any of your data that it’s saving.
0:31:09 And supposedly it all stays local right on your computer.
0:31:12 So that was one of the big announcements that they made.
0:31:13 Well, two of the big announcements that they made.
0:31:17 One was that they’re making their PCs AI first.
0:31:19 And the second one is that you’re gonna have
0:31:23 this recall feature built directly into Windows.
0:31:24 So everything you do on the computer,
0:31:26 there’s like a history of it that you can go back
0:31:31 and find really, really easily using your AI search.
0:31:33 The other big announcement they made at this event
0:31:37 was that co-pilot is kind of getting an overhaul.
0:31:38 If you use Windows right now,
0:31:41 you might notice co-pilot is like a little pop-up thing
0:31:44 that shows up on the right sidebar of your desktop.
0:31:45 Well, now it’s an actual app
0:31:48 that can actually be snapped to Windows
0:31:52 or float in front of other Windows.
0:31:54 And they showed off this really cool demo
0:31:56 where you can actually chat with the bot
0:31:59 and it will actually know what’s going on
0:32:01 with your computer and actually speak back to you.
0:32:03 So going all the way back to what we were talking about
0:32:05 at the beginning of this podcast,
0:32:10 OpenAI released a Mac app for ChatGPT
0:32:12 that’s going to have the voice in it.
0:32:15 And I think the reason that they didn’t release
0:32:16 the Windows app along with it
0:32:19 is because they knew Microsoft was going to release
0:32:21 this new version of co-pilot,
0:32:23 which kind of does all the same stuff.
0:32:26 That was the reason I think that OpenAI
0:32:27 didn’t announce a Windows app
0:32:29 because co-pilot does that stuff.
0:32:31 It can actually see what’s going on on your computer.
0:32:33 You can talk to it with your voice.
0:32:35 It will respond with a voice.
0:32:37 It does that stuff as well.
0:32:40 One of the really cool examples they showed off
0:32:42 was they had co-pilot open
0:32:44 and then somebody was playing Minecraft.
0:32:45 And supposedly they were playing Minecraft
0:32:46 for the first time.
0:32:47 They didn’t know how to play.
0:32:50 And they were having a conversation with the bot
0:32:54 and the bot was explaining to them how to play Minecraft
0:32:55 based on what it saw on their screen.
0:32:58 So it was like, they opened up their inventory
0:33:01 and asked the bot, “What do I build with this inventory?”
0:33:03 And it’s like, “Oh, you need to build an axe.”
0:33:04 And he’s like, “Okay, I’ll build an axe.”
0:33:07 And then it closes an inventory and the zombie pops up.
0:33:08 And he’s like, “What’s that? What do I do?”
0:33:10 And the AI was like, “Run, run, it’s a zombie.
0:33:11 Get out of here.”
0:33:14 So the AI was watching in real time
0:33:15 as he was playing this game
0:33:17 and actually giving feedback
0:33:20 and telling him what to do next in the game.
0:33:22 And that was also a really cool feature
0:33:25 that they showed off at the Microsoft event.
0:33:30 But it has been a whirlwind of AI announcements
0:33:31 over the last couple of weeks.
0:33:33 We’ve just been absolutely bombarded
0:33:36 between open AI, Google, and Microsoft.
0:33:41 And then in June, we also have a Cisco event coming up
0:33:42 where they’re gonna be talking
0:33:44 about how they’re leveraging AI for cybersecurity.
0:33:46 We have a Qualcomm event coming up,
0:33:50 which is starting to make a lot of the AI chips
0:33:51 that a lot of these products are using.
0:33:53 In fact, at the Microsoft event,
0:33:54 a lot of the Surface Pros are using
0:33:57 Snapdragon trips from Qualcomm.
0:33:59 So Qualcomm has a bunch of AI announcements coming up.
0:34:03 And then also in June, we have WWDC from Apple,
0:34:06 which is the Worldwide Developer Conference.
0:34:07 This is their big event
0:34:10 where they announce all of their big new features
0:34:11 that are coming out.
0:34:13 That’s last year, we got the announcement
0:34:14 of the Apple Vision Pro.
0:34:17 This year, we’re expecting a whole bunch of AI announcements,
0:34:19 possibly a new AI series.
0:34:22 But it has just been a crazy few weeks
0:34:25 and we have a few more crazy weeks coming up
0:34:28 with these additional events that are happening.
0:34:31 And hopefully, Nathan and I are gonna be able
0:34:33 to keep you looped in and chatting
0:34:35 about all of these upcoming events
0:34:38 and filling you in on all of the big announcements
0:34:39 that are happening.
0:34:41 So if you’re not subscribed to this podcast,
0:34:43 make sure you’re subscribed.
0:34:45 We put out new episodes every Tuesday
0:34:46 and every once in a while,
0:34:48 when we have really, really exciting events
0:34:49 happening in the AI world,
0:34:51 we’re gonna drop these bonus episodes
0:34:53 in between our normal episodes
0:34:55 to make sure that you’re looped in
0:34:59 and keeping your finger on the pulse of the latest news.
0:35:02 So that’s the goal, a lot of exciting things happening
0:35:05 and hopefully you’re here for it ’cause we’re here for it.
0:35:08 And thank you so much for tuning in to this episode today.
0:35:11 (upbeat music)
0:35:14 (upbeat music)
0:35:16 (upbeat music)
Bonus Episode: How are the latest AI breakthroughs set to change our everyday lives? Matt Wolfe (https://x.com/mreflow) and Nathan Lands (https://x.com/NathanLands) are here to break down the monumental updates from recent AI events and explore their potential impact.
In this bonus episode, Matt and Nathan discuss the recent announcements from OpenAI’s GPT-4.0, particularly its groundbreaking voice interaction features, and compare it with Google’s latest AI advancements. From the intriguing controversy around the GPT-4 voice to the shift in search dynamics, this episode dives deep into the heart of the AI revolution. Don’t miss their insights on the importance of Google’s project Astra, Microsoft’s new AI hardware, and the potential of upcoming AI tools and events.
Check out The Next Wave YouTube Channel if you want to see Matt and Nathan on screen: https://lnk.to/thenextwavepd
—
Show Notes:
- 00:00 AI announcements discussed, GPT-4 demo highlighted.
- 03:41 AI advancement sparks awakening and debate.
- 08:02 Twitter buzz about disputed Scarlett Johansson voice.
- 10:21 AI dominates conference, Google focuses on announcements.
- 14:08 Microsoft event, Google announcements, AI photo search.
- 16:37 Google’s revenue relies on ad revenue. Ads matter.
- 20:14 Astra project uses iPhone to recognize objects.
- 24:57 Google employees are passionate and excited.
- 28:20 New Microsoft computers incorporate AI for local processing.
- 30:25 OpenAI and Microsoft apps for computer interaction.
- 32:41 Apple WWDC announces new features and products.
—
Mentions:
- GPT-4: https://openai.com/gpt-4
- Google I/O: https://io.google/2023/
- Microsoft Build: https://www.microsoft.com/build
- Project Astra: https://www.google.com/astra
- Askphotos: https://photos.google.com/
- Copilot: https://www.microsoft.com/copilot
—
Check Out Matt’s Stuff:
• Future Tools – https://futuretools.beehiiv.com/
• Blog – https://www.mattwolfe.com/
• YouTube- https://www.youtube.com/@mreflow
—
Check Out Nathan’s Stuff:
- Newsletter: https://news.lore.com/
- Blog – https://lore.com/
The Next Wave is a HubSpot Original Podcast // Brought to you by The HubSpot Podcast Network // Production by Darren Clarke // Editing by Ezra Bakker Trupiano