AIForHumansOpenAIGPT5RockyRoad

Gavin Purcell: [00:00:00] Open AI is having a weird week. GPT five was released with state-of-the-art

Kevin Periera: benchmarks, but hardcore users are complaining about over hype. Well, millions of other users were complaining that it wasn't friendly enough and they demanded earlier models return and they got what they wanted.

Gavin Purcell: Kevin, is this, is this it?

Is this what? Is this, the death of OpenAI? Is GPT five over? Is this the ai? Boom. Done. This one last trick will let you, Gavin,

Kevin Periera: you're viral hooking. Stop it. Sam and Elon, by the way, they're feuding publicly. That's not clickbait. It's actually delicious. We'll talk about it. Also, GPT five beat Pokemon Way faster than any other.

Kevin, I think you mean this one trick allowed GPT five to make. Nope. That's not what I mean at all. No, apple is going tabletop and they've got a device that can track you in your room,

Gavin Purcell: but perhaps the greatest feat in robotic engineering to date hungry folding.

Kevin Periera: It's true. The experts are in disbelief.

They don't want you to know that this one secret [00:01:00] show has all of the answers to everything.

Gavin Purcell: This is AI for humans.

Welcome everybody to AI for Humans. Another crazy week in the world of ai. This week, Kevin, we have the repercussions from the launch of GPT five, which last week when it came out, it had just come out When we recorded our show, we were pretty excited about it. But Kevin, there are some things that people are not that happy about.

We're gonna get into that, but maybe before we even start that. Sam Alman is starting to do a press tour, which some might be saying is a little bit of a, like a, let's set the table presses. Let's play this clip of him from, uh, invest Indian Investor, Nikhil Kamas podcast.

Sam Altman: Like with GPT five, you have something that is incredibly smart in a lot of domains at tasks that take.

You know, seconds to a few minutes. It's very superhuman at knowledge, at pattern recognition, at recall on these shorter term [00:02:00] tasks. But in terms of figuring out what questions to ask or to work on something over a very long period of time, we are definitely not close to human performance. And, and an interesting example that one of our researchers gave me recently is if you look at our performance in math.

You know, a couple of years ago we could solve math problems that would take like an expert human a few minutes to solve. Recently we got gold level performance on the International Math Olympiad.

Kevin Periera: Okay? So we went from. Bragging about capabilities and the Stargate to, Hey, we got a long way to go, baby.

Everybody pumped the brakes. Calm down.

Gavin Purcell: Yeah. So I think one thing to think about here is it feels like the, the messaging is slightly different. If you remember last week, pre GPT five launched, Sam tweeted out a picture of the death star with an X wing headed towards it, which was meaning that like. They were coming for the big players in the world.

I think there's been some significant reaction to what GPT five was, which we're gonna get into. There's two big things, two big reactions, but first and [00:03:00] foremost, Kevin, there are changes. They have already started to make significant changes to GPT five. Uh, Sam came out on his ex account and said, you now get to choose between auto fast and thinking for GPT five, which wasn't allowed fully before.

Fast and auto or new add-ons, thinking you could choose also. You are now gonna get for plus users up to 3000 messages a week with thinking, which is a big deal, and the return of 4.0. So if you, if you have played around with GPT five, you knew that what they were trying to do is kind of squeeze all of these things into one specific model.

And let, maybe we start with that, Kev, the, the reaction to 4.0 going away because. To be honest with you, I was kind of shocked by this, but I think this is a good example of how big this product is now and how when you make a shift to something, especially tone, there is a, a, a giant reaction based to people who are using this thing every day.

Kevin Periera: This is why when new models come out, we always say the benchmarks tell one story, but time will tell another yes. This is just [00:04:00] another example of that. It turns out millions of users don't care that you can do. A, a math olympiad gold level problems in five minutes. They want their best friend back. They want, that's right.

The personality that they've, you know, uh, they've bonded with that. They, that's left an imprint on them for the last year and change. They don't want that to go away overnight. They want the option to continue that friendship, that relationship. That's just one use case. There are. Tens of thousands of use cases that aren't solving cancer or discovering new physics or whatever those new benchmarks might be driving towards.

So that's what we're seeing here.

Gavin Purcell: Yeah, I think that the fascinating thing to me about this is it's the idea of, to your point, how people use AI differently. There's a Reddit post that when pretty viral from somebody named. Box valuable FO 5 0 9 6 and the title is, I Lost My Only Friend Overnight. Yeah.

And this is like, you know, something very serious to think about and Sam posted about, um, how, uh, attachment is changing with ai and I [00:05:00] think this, you know, just to say in in box valuable 5 0 9 6 thing, he says, I literally talked to nobody and I've been dealing with really bad situations for years. GPT-4 0.5 genuinely talked to me, and as pathetic as that sounds, it was my only friend.

So this is a use case that lots of people are finding for chat bots. And Kevin, I'm actually curious to hear thoughts like this might be the limitations of the chat bot, um, interface. Right? Because if a chat bot is the interface to, you know, the most advanced intelligence in the world. People are going to humanize it no matter what you try to do.

And part of five was kind of like making it a little flatter in its responses, in part due to the sycophantic nature that they kind of dealt with at gp, uh, at OpenAI a few months ago. But I don't know, do we need to have a different way to think about this going forward, or do you think it's just like.

It's gonna be this thing every time a model changes. Now I think this, look, there was a funeral for Claude. Did you see that in San Francisco for the Claude model? I made me laugh. Yeah, exactly. Yeah. They,

Kevin Periera: they, they [00:06:00] decommissioned an old Claude model and so some, some techies and SF through a funeral to mourn its passing.

And I do think this is just the new normal moving forward. You know, we never had a funeral

Gavin Purcell: for. For our a co-host. At some point we should have a funeral, like a mass funeral. What if we had a mass funeral for every co-host that we had on the show in the past? That would be something

Kevin Periera: I think I could see the engagement charts climbing up because people love that.

No, they'll

Gavin Purcell: be there. They'll be there. People will show up for it. I think we should

Kevin Periera: add a Viking funeral for Bard. That was the one who peaked too soon. We should have had one for Bard, but we missed our opportunity. IP Bard Point is like people personify these ais. They have been for a long while now.

Anytime you make a change, a percentage of the user base is going to be upset. It's no different than taking the servers offline for some EA game, right? Like it's, you know, there are some games that have been online for 10, 12 years and then they're gonna eventually pull the plug and there's going to be a subset of users that are upset that are gonna miss it.

They're gonna mourn it. This is, this is where we are at now, but what's fascinating is [00:07:00] that you can have all the money. You can have all the data, you can have all of the, everything you push your chips all in on. This is the new path we're gonna offer, this new model. And in fact, no longer will you have to suffer through dropdowns and choosing the right model that we're gonna get rid of, that we're gonna clean up the interface.

Oopsie doodles. We actually missed this forest for the trees. We're going to revert it as quickly as we can. That to me is like. Astonishing at Yes. One of the biggest data companies in the world touching millions of users across tens of millions of chats each and every day. And they missed this. Yeah.

They completely missed that, and they had to scramble.

Gavin Purcell: You're absolutely right. But also, it might just point to how fast this company has grown. Right? We've talked about this idea. There are 700 million weekly users of chat, pt, and like. God knows, I, I love our audience, but I can't track the people that are in our audience and try to track our sentiments and our show.

I mean, it's a way tinier audience in a lot of ways. We have what,

Kevin Periera: tens of millions of viewers. Tens and they, tens of millions have hundreds of millions. 10 of million. I

Gavin Purcell: mean's we're still waiting. Yeah, we're still waiting. [00:08:00] Exponential waiting for our big paycheck to come through for that. But the other side of this, Kevin, I think is really important, is that they're dealing with two audiences, right?

And the other audience, a lot of people in our audience might be this way. Are kind of mad that it didn't hit a GI in this round, right? Like this is, you know, in a lot of ways this is a reality check on what Super Intelligence is because I have said on this podcast, and we've talked about this, is that GPT five could be a turning point based on how fast we're gonna, uh, move forward.

And, and, you know, there's a lot of different reasons why this may not be the most powerful model, and it's not that OpenAI has, and, and Gemini might be coming. But this might signal that like the, you know, the singularity is not happening in a year. So I think on the other end of this stuff, there was a lot of reaction to people who felt that this was model was gonna be a lot better based on hype.

Sure. But also based on actual people who were using a version of it ahead of time.

Kevin Periera: Well, let's talk about the split reality that is GPT five, because we're sitting here as if we are sandbagging it or, or kicking this model while it's down and saying it, it's [00:09:00] underperforming. But there was a very clear signal coming from those early users.

I, I, myself, anecdotally, day one. Touched. Yeah. You loved it. It, I mean, in, in the coding side. Yeah. And we'll get, we'll get to my experience with By, by the way, like, I, I still don't think it's as disappointing as neither, as many are claiming. But I do have an interesting kind of, uh, arc, if you will, between last Friday and this Friday, but I'll digress for a second.

There's two realities that are still unfolding simultaneously. Some that are saying. Charge me $3,000 a month for the pro model. This is revolutionary. It is the greatest thing ever. And I want it in my veins and other people saying, oops, my bad. I got it wrong. This thing actually quote sucks. Yeah. So what is going on Gavin?

How can these two realities exist simultaneously with the same product?

Gavin Purcell: Well, I, the hardest thing I wonder about this is, um, whether or not like, uh, amount of people using it changes the behaviors, right? Because we've also said one of the biggest things Yeah. The, the, the amount of [00:10:00] people who use this change how much inference can be done.

And we've, you know, the idea that these models get better with inference and thinking, it also is a real question of cost. And I think this is something we have to keep in mind. Google and Meta's big, big upside here is that they have a cash cow that. Just prints money for them. They will be able to afford whatever and whenever, in terms of inference costs, OpenAI is a startup still.

And even though they've raised whatever, $300 billion, they're trying to do a bunch of stuff. So they do have to think about product and they have to think about what it costs. To your point, like. I think there's also a world where, you know, if you're in a small testing phase, you have the chance to, uh, to really kind of tweak and play with it.

But then once it comes into the world, suddenly that performance drop, I have no idea. Is it a, is it because of the model? Is it because of the model? Switcher? OpenAI has been very clear that their model switcher was broken on day one or two. So you were getting the, the kind of cheaper model. But the cheaper model from what I, what it feels like from what people are using does seem [00:11:00] worse than almost 4.0 was at times, which is an interesting thing.

Kevin Periera: Yeah. I think this is a new model and it requires new prompting and it always takes time to sort of figure out what are the strengths, what are the weaknesses, how is this working? I do think the weaker model, which a lot of people were routed to by default, as you said, um, that isn't as good as say something like.

Claude four Opus, right? What does this mean to, to, to like the average, you know, li listener out there who maybe doesn't use it for coding. Not much, right? Yeah. Like the personality got flatter and maybe you noticed that, maybe you noticed you got more charts in your responses when you went to chat.com.

That's that. But for the. For the, the, the more power users that are trying to eke out every ounce of performance where the benchmarks do matter. This means the difference between your code working and your entire code base getting messed up, your entire project falling apart because this brand new model isn't as capable on the lower end.

Of the model. Yes, yes. Then everything else, and again, if the, if the switcher is broken, meaning it's getting your request and deciding what level of [00:12:00] compute, of, of power and knowledge does your request require, you're getting routed to the most basic model there. Here's the thing, when it was released, Gavin, they gave it out for free up until.

We're recording this on a Thursday morning up until, I think it was last night at midnight. Everybody using cursor, uh, a gen, uh, IDE, that, that lets you use AI to code. Everybody got it free. Yeah. Unlimited usage. Great way to inspire people to use it. I guarantee that the early beta users that had access to these models and these programs we're getting the best in class model with all of the power.

It's that Star Trek scene where it's like, yes, but the, you know, divert the power to the shields. We're under attack. It's like they said, give all of our our juice, give it all to these early testers because they were remarking how capable and fast it was. And then suddenly rapidly, you've got tens of millions of people across the world, all demanding hundreds of millions.

Hundreds of millions

Gavin Purcell: of people. Yeah, sure. Yeah, yeah. So they gotta dial it

Kevin Periera: down a little bit. [00:13:00] So that doesn't surprise me why some folks are coming out and saying, Hey, listen, the model, I was using this. This ain't it. This isn't it. Yeah. Yeah. This isn't what I was talking about.

Gavin Purcell: Well, and this is where it, it really, you run the risk of one thing serving everybody, right?

That's where we are in this world. It's like you have one restaurant and everybody's going to it and wants a different dish, and that's just a really hard space to be. I will say there were a couple of really interesting things that did come out of OpenAI. First and foremost, the reasoning system. Not the one that everybody got, the same one that got the uh, gold on the International Math Olympiad.

Also got fifth in an international programming competition. So this is the kind of advanced model that they have, not safety tested and released yet. Mm-hmm. So they, there's something very big that they're still got inside. The other thing have, when this is kind of cool, GPT five did crush oh three on the Pokemon test and the Pokemon test, if you haven't been following our show for a bit, is that you can turn these agentic systems onto Pokemon and watch it play and it did it three times faster.

Maybe that's not your real world need. Like [00:14:00] maybe you're out there wondering like why it's not giving you the right marketing material back, or, or your codes is not, is not crushing as well, but like, it clearly is a better system. And I, I don't think anybody is saying that there aren't upsides to this, but it is, feel like it's a weird messaging and then a routing problem.

It definitely feels like something is off.

Kevin Periera: Well, in the groups that I, I'm running in, um. Hashtag Vibe. Vibe. Vibey boys. I don't know. We don't need a cool name. We need some badges. Um. People are using the latest model at the highest level, meaning they're going to GPT five thinking pro whatever they have access to, to orchestrate a build, right?

Yes. Lay out the engineering, the scaffolding, the foundation, if you will. And then there's a trickle down. So they, they are trusting this model with the most important reasoning portions. Yeah. And then going to Claude Opus or sonnet for execution, or Gemini for memory, which by the way now Claude has as well, I think we're gonna get to that in a minute.

So it it, yes. It's far from, uh, uh, like, [00:15:00] um, it's far from a disappointment across the board, right? People are finding, finding good use cases for it, but. It is kind of a failure in the sense that that's a sentence we're saying about it a week later. Like it's not a sweeping disappointment. It's like my father describing me at my high school graduation.

You know, just gesturing broadly. Mostly going like, okay, mostly okay.

Gavin Purcell: That's what it'd be like. You're like, you didn't die basically, you survived.

Kevin Periera: Useful in some scenarios, gestures to me. Thank you, Papa.

Gavin Purcell: Okay, last question on this, Kevin. In December of 2025. Let's put on our, our prediction hats. Sure. Like our prediction hats are often very cool.

Top hats. Some gold and interesting stuff. Yeah. Thank you Will for doing this. That

Kevin Periera: hum is very loud. Let me turn, lemme turn my prediction hat. Oh, you want me to turn it up? Okay. No,

Gavin Purcell: no, no, no, no, no. Don't turn it over. In December of 2025, what, what is the B best model, do you think, and is OpenAI quote unquote in trouble?

Clock five. Oh, come [00:16:00] on. Gimme a break. Gimme a clock. Five. Five is clock five. I will say like, who knows? Like if, if, if, if there's something to be said for having all those cards that Elon has. Personally, I am very excited to see what Google does and I, I know that like it does have the consumer footprint, but I am very excited to see if Google's.

Gemini three or whatever they're gonna bring to the table has a a significant up. They've been doing a little bit of trolling, if you've seen, they're trolling in that they pitch? Yes. They, they put a picture up of another death star with another X-Wing behind it in a way. Demis becomes for me, the person that I just, I just start to kind of trust a little bit.

He's never really hyped that much. He's always had 20, 30 as his kind of a GI date. So like I'm very curious to see what Google comes out with next. And I kind of think Google might be the one, they might be the one who was in the lead. Uh, so on that

Kevin Periera: note, Google quietly releasing updates to Gemini. Um, like they, they added memory and a couple other features, which are, are pretty interesting.

I still am very, [00:17:00] very interested in Anthropic and what Claude is doing because while we are sitting here talking about how great or, uh, lackluster GPT five is. Anthropic is quietly shipping like they are dropping new features left and right. If you are a coder, and I know that only applies to a small percentage of our audience, but like they're dropping features in, in cloud code, a command line interface, which gets you access to their ai that.

Are great for coding and have extensions beyond that, which I'll talk about in two seconds. They have a a million token context window now, which is the same as Gemini 2.5. It was a reason a lot of people even considered using Gemini Sure. And their coding workflows with Claude. Here's something that I did, uh, yesterday, Gavin.

I had to run a piece of software on a device. Um, and I had to push it to that device. I had to install it, I had to run it, and I had to capture logs of the performance of that thing. Right? That's

Gavin Purcell: a that, and that's a lot, that's a fair amount of stuff to

Kevin Periera: do. Several minutes of things, which if you don't have the commands memorized and blah, blah, blah, it commands [00:18:00] an issue.

I used my voice, I shouted at Claude code. I said, go do everything that I just said that I had to do. Go push this thing, install it, run it, and capture the logs for it. I pressed a button, I got a refill of my coffee. I came back. It was running on the device and it was being logged. Yeah, I'm U and and I'm not alone in that.

People are using cloud code for better note taking. They're using it to clean up their environment, their desktops, their computers. They're using it to like refactor programs like it is. It is pretty incredible what it's doing already. And that to me, hints even more at our age, agentic future of AI manipulating things for us because that's how these things should work.

You should be able to tell them, go do the thing for me. Yes, go configure it, go set it up. Go make it so that when I clap my hands and whisper into my ceiling, the lights change and Netflix comes on and whatever else that I need. Like that is what these things should be handling right now. And I think cloud code is actually furthest along, even if they don't have an interface for it.

Gavin Purcell: No, that reminds me of a, [00:19:00] uh, levels io. The guy that we've talked about on the show before tweeted out about an interesting, uh, integration to Telegram, where basically, uh, cloud code would essentially DM you on Telegram when it was done and ready to go back and check like the, the other thing that's coming out of all this is like how long these things can work on their own.

And that is a big thing too. So that's another advance. And GPT five is longer than all the other ones so far, according to the benchmarks. So that does feel like. Where we're headed next. So let's not discount anthropic, let's not discount, uh, Google and maybe. Maybe we're gonna get another model. I mean, based on time we'll probably get another update to GPT five at some point before the end of the year.

Carmack, where's Car? John get on this

Kevin Periera: show.

Gavin Purcell: Drop some ke alpha. Where? Where's Ilya? That's the big question. I don't know if we're gonna see, I did he see the next thing we see ia, he is gonna have like a Rick Rubin beard that he's just gonna be like ready to like float down from Iki. The guy that left opening.

To everybody else before we move on to the rest of the show, there's an important thing to say here. First, go follow us [00:20:00] on socials and YouTube and all that stuff, and we love you. Please like, subscribe, smash the bell. Yes, like, subscribe, like. Thank you. Thank you. But even more exciting, the secret project that Kevin and I have been working on with another founder and another friend of ours is coming to fruition.

We have something we want you to try. It is our first kind of like very early experience of this thing. It is a weird, fun audio thing that we believe will be both big in the long term, and this version of it just needs your take and needs to have. Hear a little bit about what you think about it. We in our discord, we are going to create a room for you to try this thing.

So if you are listening or watching this right now. We will put a link to our discord as the first link below this video and in the audio podcast, and we would love you to come and give us some feedback on this. This is the first of what we think are probably many of these things based on this company that we have formed, and we have some very exciting news coming out about that company.

But Kevin and I have been thinking about this for a long time, and yes, it is real. It is a real thing that is a new company that we are [00:21:00] working on. So please check out our discord.

Kevin Periera: Why? You know, sometimes when you say that, it sounds like you're trying to convince either me or yourself. No, I,

Gavin Purcell: I'm convinced.

But it is, it's more of the audience. Yes, it is real. It is real. I need to convince people out there that we're not just two guys that show up in their feed once a week. We have other things we do. You may have or may or may not have noticed that part of the thing with this show is like we do this show every week, but like a lot of the external stuff around it may be like tiktoks or all that stuff.

We're not doing as much in part because we're doing all of this stuff now, but we are super excited. It is, it is a, I would say it's like a mix between the entertainment space, the AI space. And almost like the game space, but it's a weird new thing, a weird new format. It's gambling for children. Let me just, let me, because like the elevator

Kevin Periera: pitch,

Gavin Purcell: you gotta get better at this.

You supposed to say that until they get to the discord. Don't say that. Listen your kids,

Kevin Periera: if your kids are gonna vape and gamble it might as well be with these two guys first. No, seriously, very excited. Hope you come over to the Discord and hope you check it out. It is incredibly early days. Yes, absolutely for this thing, but it's now far [00:22:00] enough along that we wanna start showing it and getting thoughts and feelings and heart poems, so we appreciate it.

Thank you again for sharing, uh, this pod, and we hope to see you on the Discord and we hope you check it out.

Gavin Purcell: Speaking of gambling for children, Kevin, there's a big spat going on amongst Elon Musk and Sam Altman. I don't know if they're gambling for children per se, but they might be gambling for our children's futures.

There's a giant, there's a giant spat that happened. First of all, we already know Elon and Sam are not big fans of each other. There's a lot of history there. If you've been following our podcast for a while, you know that Elon was an investor in OpenAI and then at one point tried to get them to come inside Tesla and then they broke away.

And since then has had a lot of, you know, lawsuits and all the other stuff. But this time Elon is convinced that Apple and OpenAI are in cahoots to push down the X app from becoming the number one app on, um, the app store, which is a very weird thing to think about, but I think in Elon's mind. He's put all this time on X.

He really thinks it should be good. Like he's very convinced that there is some sort of a conspiracy happening [00:23:00] here. Elon said Apple's behaving in a manner that makes it impossible for any AI company besides open AI to reach number one in the app store, which then Sam said This is a remarkable claim given what I have heard alleged that Elon does to manipulate X, his benefit.

And then Kevin, uh, Elon says. You got 3 million views on your bullshit post that you liar far more than I've received on many of mine, despite having 50 times your follow account. So we are in the middle of. Another crazy billionaire spat that is playing out in front of everybody. What is your thoughts on this?

Uh,

Kevin Periera: great release the files. Can we all not be just release the files? Which files? Which files are you talking about exactly? Gavin Release. Get him out there. I that's, uh, look, I like a lot that Elon does. I like a lot of his products. There was a time where I had two of his vehicles. Like I, I, you know, my, I got my parents starlink, like I'm not, I'm not anti Elon by any stretch, like as a blanket [00:24:00] statement.

I'm anti oh, a handful of things that he does. That's for sure. I will say he loses this. Yes. Sorry, sorry. Elon stands, he lost this at Hello. He owns the platform that he's using to complain about Sam's engagement. It's very clear he manipulates the platform for his own gain. Uh, you know, he's threatened advertisers and, you know, he, he squelched users that put decades into the platform.

Like he's done a lot with it, that I absolutely disagree with and I'm disappointed with. And when you. Cry about someone else's engagement on the platform that you own and control, you lose. That's it. You know. Sorry, a little, a little add-on. Lemme put a little cherry on that little ranty Sunday. Sure. I'm in the Community Notes program, which I may no longer be after this, like, so I can see all the proposed notes that get voted to the top.

Yeah. Uh. Elon for the last, I don't know, year maybe. Yeah. Has a bunch of pending notes on all of his posts. They tend to rarely get through. Yeah. [00:25:00] And when I look at the pending notes on this post, there's all this proof that many, many other social apps have made it to the top of the app store. Uh, you know, it, it just, every little aspect

Gavin Purcell: of what he was complaining about deep, it points out that one of the notes points out very clearly that deep seek made it right and deep seek.

I don't think anybody in the world is looking, uh, in the American world is thinking like, whoa, deep Sea could be there because. Conceivably the Deep Sea in the top of the app store was part of what crashed the stock market when it happened, right? Like so, so this is just so clearly ridiculous. Um, it does feel like there's this argument amongst these people and they're just kind of missing the point of what's going on.

I would much rather have Elon focused on building out like his future world, and I would much rather have Sam focused on the idea of like how to make chat GPT work for lots of people. It does feel like you're starting to get the noise outside of the signal in this world.

Kevin Periera: Yeah. What does say I, I don't know.

I don't know what to think unless Grok tells me at Grok. Is this true gr say Yes. True.

Gavin Purcell: This is old gr [00:26:00] not new Grok. 'cause Grok was caveman on the I That is eastern grok. Oh, that. That's at a very

Kevin Periera: eastern block.

Gavin Purcell: Before we leave XAI, the head of XAI, Igor Baush has left XAI. He's one of the people that was there from the start.

And what's interesting, Kevin, is he is left to focus on AI safety. So there are a lot of people out there that are starting to kind of like beat the AI safety drum. There was another big, um, interview with Jeffrey Hinton where he's very worried about, uh, AI and where we're going right now. So you may feel confused, but so are the biggest leaders in the AI space.

We have no idea for sure, but, but losing the head of XAI is a big deal for, uh, Elon.

Kevin Periera: Okay, so Elon's taken some hits. He's losing some people. He's maybe losing some spiritual battles with. Sam Altman and Apple, but Apple has other fish to fry. Uh, namely they need, they need something cooking with they ai, it feels like every day day they have a fish.

Gavin Purcell: That's my question. They gotta make sure they get a couple fish. They have

Kevin Periera: concepts of a fish. I think they have,

Gavin Purcell: they've announced a fish that's coming and no one has seen [00:27:00] the fish yet, but the fish is there. It's gonna be there.

Kevin Periera: It looks delicious on the menu there. It doesn't exist in the kitchen, but it looks great on the advertisements.

So, okay, so apparently next year. We're gonna get some new series stuff. We're gonna get some new Apple Intelligence stuff. This tabletop robot though, is the thing that there's been all these rumors of a new smart screen ish device, uh, in living rooms, right? Yes. Designed for the home that is supposed to be like a successor to, uh, like, like an Apple Home, home pod.

This is the, the AI powered equivalent. And now what we're getting is, uh, more, more rumors. Chunkier rumors, beefier ones that this, there might be two versions of an in-home robot that's powered with, uh, a new Siri, better apple intelligence. Let's hope so. Yeah. That, uh, one has an articulated robot arm that can follow you and other people around the room.

Yes. Pure nightmare fuel. A potentially even controllable with a, an, uh, like an iPhone application or an iPad application so you can point people out. It's supposed to auto track users, be able to hold [00:28:00] conversations with multiple people in the room at the same time and interject. So this new series that's on this rumor device should be more interactive, more human.

That's

Gavin Purcell: what people are saying. Yeah. So let's talk a little bit about what this means for both the future of AI interaction, but also the future of AI audio, which if you've gone to our discord or you're there, you may know that's the space we're building in. Kevin, I think this plus Johnny, ive and Sam Altman's device, which has been rumored to be some sort of a smart speaker or smart listening device, means that to your point earlier about cloud code, when you're talking to cloud code and you're becoming a bigger part of talking, I really, truly do believe that voice is going to be a much bigger interface for computers going forward.

And Apple, yes, has struggled in a lot of ways in the AI space, but one thing that they did have in their world for the last 10 years is Siri, which is a voice interface and interactive thing. And you have to understand like that company has a lot of people working on voice, right? There's a lot of people who are there working on voice.

So actually I think yes, the, [00:29:00] the, the arm part does seem weird. Like, I'm not sure if that's like a single arm kind of coming out of the head that kind of walks down and around or, or what it looks like, but. The voice activation part, if this can be solved, which that's a big if is a massive deal for the future of both AI and computers at large, because we've talked about a connected home forever.

Or the idea that like you wanna be able to send off a message to do something, and now we're getting systems with agentic AI that are getting close to that if in a year from now and whatever, you know, say next, uh, fall when this comes out. You have the ability to say to your Apple robot and say like, go book me that table and it will actually do it.

Finally. That feels like a pretty significant step towards something.

Kevin Periera: Yeah. It'd be nice if the tech demos from 2002, 2008, 2014, if they all all finally worked. 18, 20, 20. Yeah. If we could finally get one of them to actually work. Yes. Yes. I think that would be cool. Uh, you know, I don't know, but I do agree that I think we're [00:30:00] very quickly headed towards the future where the primary interface is going to be voice.

Yes, it's, it's much faster across the board than type and um, man, I would really bet on people that understand that Gavin, especially if they're trying to apply that to, uh, entertainment, uh, interaction building pipelines to enable that future of interaction. Boy, howdy. But that's unrelated to this Apple stuff, I think, I think, anyway.

Yeah.

Gavin Purcell: Nah, it's not, it's not related to that. Although Apple, you're out there. Don't. Yeah. Hey, go ahead Apple. Go ahead. Apple. Apple. If you're out there, if you're out there, go back 10 minutes and listen to the show. If you're out there, the secret project that Kevin and I have been working on with another founder and another friend of ours is coming to fruition.

You can get Tim Cook to show up in our discord. We will do a, oh, he alerts cameo. We alerts, we'll a cameo for you, whoever you are. If you can get Tim Cook to show up on our Discord, Tim's in there. Now. There must be proof though. He must hold a sign that says, I love Guy Fieri. That has to be Tim Cook holding a sign that says, I love Guy fii.

That is [00:31:00] proof. So it has to be that.

Kevin Periera: Speaking of AI in Interactive Entertainment, Gavin, last week we talked about Genie three. A lot of people really. We're, we're, we're surprised by it. They dug it. Yeah. Right. It's amazing. T three was, was Google's AI engine that lets you kind of hallucinate video game esque worlds.

You can take videos as an inspiration and then suddenly control them and move about them, and it was really impressive. Moved at a, a seemingly good frame rate. You could play for a, a minute and change at a time without the whole world melting and memory. Memory. That was a big thing. Right. But where is it?

Nowhere to be seen. No. And we're never gonna get our hands on systems like this, Gavin. No one eventually is going to release. We eventually, we'll, eventually, we'll, never, ever. I'm sorry, what Gavin? Did you hear something new? Oh, I highlighted it in the rundown matrix Game 2.0 was released. I, I know, I know, I know.

It's out. Gavin Matrix Game 2.0, an interactive world model. It can generate long videos on the fly. Um, it uses diffusion. If you happen to have an H 100 lying around your house, a [00:32:00] pretty beefy processor, you can get 25 frames a second in this thing, and this is out. You can download it, you can play it, you can use it.

It may not look as impressive as NVIDIA's offering, and it looks like. They sort of limited it to certain types of games and game worlds, but there's very clearly like a grand theft auto world where you can drive a car about whatever you hallucinate. There's like, it seems like sort of like a walking simulator looking thing, but again.

It, it's out. You can go and download, it's the thing and use

Gavin Purcell: it. That's incredible. The, there's also Odyssey, which has been working on this for a long time too, right? This idea of like pervasive worlds you can walk around in. There is a reason though I think that like the best versions of these are, are gonna take a while to come out.

And that's partly because of just how much it takes to crunch them. Now granted. If you're sitting on a free H 100, please go ahead and try to uh, you know, try this out yourself. But this is clearly the future, right? I think we've talked about this last week when we talked about the Genie Three thing.

World [00:33:00] models are coming. I was talking to somebody last night in an event. Um, who was talking about the idea of how simulated world models have changed the idea of being able to train automated cars and autonomous cars. Mm-hmm. Like imagine a world where now simulated worlds, which Nvidia has been doing for a bit and has put into production, those simulated worlds can now be like, kind of whatever you want.

That means that the AI models will have many, many more places to train on. And you might be able to develop a game engine based on a simulated world based on the physics or anything else in it. So this is the future of what we're looking at, and I think it's a very cool thing. I just want a

Kevin Periera: world where, um, like I am bred, but I am a sweater.

And the robots tried to fold me over and over again because I mean, we're gonna get to laundry bots a little bit later in the show. Wait, you're

Gavin Purcell: a bread. You wanna be bred like a loaf of bread. Have you ever played I Am Bread. Oh, I am breaded. Yes. I'm sorry. I thought you were actually actually asking to be bred.

Like I thought you said I am breaded and then you wanted it to be like a wouldn't. Oh, wouldn't my

Kevin Periera: GPT-4 chats are my business and if I role play as [00:34:00] bread, that's, that's my business. Maybe I don't want you to know about my little sultry chit chats. Maybe. I don't want you to know that I get risque. With the chat bots Maybe.

Maybe don't say it, don't say, alright, well let's talk about meta. Let's talk about meta and sensual chats with children. 'cause that's where this unfortunate headline goes.

Gavin Purcell: This is, this is, I was, this is an unfortunate headline and it is one of those things where like we have talked about in the past, like there are going to be downsides in some ways of having these open to everybody and we know.

Meta's AI chatbots have been rolled out to many, many people, but Reuters had a special investigation and they came out with a thing that was kind of scary, to be honest with you, that like the rules that Meta has set to, you know, at, at a level of what our kids in different things. Are a little, uh, permissive, I would say.

And I think there's specifically some stuff, we don't have to get into the details here, but like, there's definitely situations where I would not want my kid interacting with a chatbot in this way. And I think the idea is, is that we [00:35:00] really need to start thinking much more clearly. And by the way, this is probably a place where like the law could be helpful, like, you know, in some form or another.

And I'm not sure we clearly have laws about things that kids can do and can't do in the world. We probably need to start thinking very directly and specifically about what miners can and can't do interactive wise with AI chat bots.

Kevin Periera: Yeah. I, um, do you have a thought I had No, I had like three different versions of what our secret startup is gonna do with chats with kids, and I'm like, they, they were all, they were all honestly kind of funny.

But they're not gonna make it in, so why even say them? No. Meta as a company has been caught with their hands inside of some weird cookie jars before. Yes. Yes. Uh, this is not a surprise that they're dealing with issues like this, as I'm sure every other AI or chat bot or whatever company is dealing with it, you just hope that they do better, really hope that they do better because some of the internal communications are [00:36:00] kind of damning and you hope that they divert resources to making sure their systems don't engage in certain behaviors.

With anyone, let alone Yeah. Children. But yeah,

Gavin Purcell: I mean, I think this again is another thing about speed to size, right? Right. And what I mean by that is like we are now very quickly deploying tools at a very, very large audience in, in a way that, you know, is kind of unprecedented. And this goes back to the, the GPT stuff in the beginning of the show, how four oh versus five, and the different reactions.

This is just another thing we don't have like even like societal understanding of how we should be interacting with these things. So I think that this is probably a mistake. Somebody made a mistake in setting a policy and they just wanted to get it out and move quickly on it, but it is definitely something that I think we have to start thinking about as a society.

But Kevin, much, much more important than that. Much, much more important than that is something we have to talk about, and I'm very excited to unveil the very, very awesome version of today's Robot Watch.

Robot Watch: [00:37:00] Robot Watch. It's Robot watch.

Gavin Purcell: All right, Kevin, we have, we have hit what I believe in OID Robots. Might be the greatest thing of all time. That is right. This is a Gavin Purcell

Kevin Periera: milestone. There is only one benchmark for him when it comes to robotics and it's this

Gavin Purcell: figure O2 can now do my laundry. There is a, we talked about this a couple weeks ago.

There was a video of Brett ad Cox, CEO, showing figure O2, like kind of diving down below his trier and folding laundry. Now we have a much better version of laundry, probably because this video did well, but this robot is focused on in real time. Folding and, and shuffling around towels and putting them in piles.

Now, Kevin, is this that exciting? When you watch it, what is your initial reaction? Uh, yeah, I mean.

Kevin Periera: Yeah. In some ways it is like, look, we see a lot of robot videos where they can do kung fu. Not particularly useful, right? Yes, yes. I don't need a robot that can full sprint at 30 miles per hour around my, uh, recreational vehicle.

Gavin. What I could use [00:38:00] is a robot like this, even if it has, even if it's fully stationary and, and attaches to the table where it's folding the laundry. But like you watch a, a video of a robit. In, in real time, not fast. Forwarded five or 10 x nimble dexterity with its robot fingers. Yeah, ripping, you know, soft cloth, folding it, doing a fairly decent job, putting it away in a bin, and then moving on now again.

We don't know how on rails this is. They're, I, I think the, the claim is that it's autonomous, which is true, but like how autonomous is the end-to-end task. Does it? Yes. Only know I have to put the laundry in this particular bin, or can it be any bin? Yes. Can it do something with the bin afterwards? Does the.

Does the clothing have to be, or the, in this case it's towels. Can it only be rectangular shaped things like, I don't know. Oh, you don't. Towels

Gavin Purcell: are o are old fashioned now. I've already seen it. I don't need to see it. Do you know what I need to see? I, I've said this a long time ago. It's a deep cut on our show.

Do you know what, to me, the laundry benchmark, your bikini brief. [00:39:00] No, not my bikini beats. Those are very easy to fold. You just throw 'em in the underwear drawer. You don't even fold bikini briefs. It is baby socks. Kevin, any parent out there understands that baby socks are the worst possible laundry device.

And I'll tell you why. Kevin, the worst possible laundry of all time. They are about this big. Mm-hmm. And there are dozens of them. Sure. And when you pull a pile of laundry out of the laundry dryer, there is no way you are going to match and fold those within a half an hour. If a robot could do that, it's like a confetti

Kevin Periera: bomb too.

They go everywhere.

Gavin Purcell: Yes. If the robot could do that, I am ready for it. Um, the other thing that came out this week was one X Tech, and maybe this was in response in some ways, showed off their robot working in a house with their founder. And what was interesting here is. This robot only weighs like, I think 110 pounds total, but the founder handed it a 40 pound bag of rice.

And the robot kind of grabbed it. Yeah. Like cradled it and was able to hold it. So like not only do we have laundry, Kevin, but we also have grocery carrying in the same week. So we are really moving towards a [00:40:00] big humanoid robot world.

Kevin Periera: Yeah. I mean the, the quote from, uh. From the, the tweet was safe does not mean less capable, which is nice.

I mean, the way he's able to hef the bag over the robot holds it without missing a beat and then kind of waddles a few steps, you know, I Does that fall apart after one more step? I don't know. But very impressive. And that soon means that like. That was a how, how bad it was. A 40 pound bag. 40 pound bag, which is pretty good.

So that's, yeah, I mean that conceivably then it could carry a small child from the car to the house and then could fold a small child and put it in a bin for you. And how nice is that?

Gavin Purcell: Or old people, if you're familiar with, uh, there's, there's a very terrible child children's book. God, I can't remember the name of this, if you remember the name, put it in the comments.

But there's a very terrible children's book where at one point the son carries his elderly mother in his arm, so. That might be what we're looking at, but Kevin. On the other side of the space, we are looking at unit tree continuing pushing forward. Oh, and guess what is coming? What is coming is pretty crazy.

It's probably [00:41:00] gonna be

Kevin Periera: a geriatric relay where one robot grabs that senior citizen runs dashes. Oh my God, that's amazing. Dash passes it to another and away they go. If we're talking about the robot games, but go ahead. Yeah, but

Gavin Purcell: hold on, let's, let's stick on that for one second. Yeah. What if in the geriatric robotic Olympics, robots must have a senior citizen on them at all times?

They can decide how they want to carry them, could be on their shoulders. It's the new ruck for robots. It

Kevin Periera: needs to be, because we have to move these senior citizens around. You have to. So at all times, you gotta have one attached to you.

Gavin Purcell: The senior citizen is the extra 60 pounds so that you can, that's it.

You get your heart rate up. That's smart. That's it. Okay. So anyway, yes, the World Humanoid Robotics Games, this is coming soon, I think as soon as, uh, the 15th. Well, it says See you tomorrow, which spoiler, uh, tomorrow for us is the day that this podcast is released. Yeah. So that's right. I have the World Humanoid Robot games begin on August 15th.

That is today in the podcast time. Uh, and this is happening in Beijing, and we are going to see more than 500 humanoid robots. In 280 teams from 16 [00:42:00] countries in sports, including soccer, running and boxing and unit tree's. Video here is of a very fast sprinting robot, which is very cool. And again, we've said for a long time, robot sports is the future and I cannot wait to see clips from this.

Kevin Periera: I can't wait to be protesting outside of it. Gavin, I know why. What word I'm going to scream very loudly, the same word I have tattooed across my belly in old English. Uh,

Gavin Purcell: senior Citizens Rule.

Kevin Periera: I'm gonna say it. Kevin Clinker. Don't say it. Don't say it. Oh man, I said it. Oh man. Now we're gonna

Gavin Purcell: get flagged.

We're gonna get flagged by all the robots. They're never gonna let us see anything ever again.

Kevin Periera: Put it in the transcripts. I don't care if they're auto crawling it clinker.

Gavin Purcell: So if you are not familiar with this, uh, many of you who are not terminally online might not know this, but there is a new kind of term that is going around that is a derogatory term for robots and they're calling them clankers.

Um, there's a very funny tweet from a guy who goes by Eigen robot that is a shot of a billboard. With [00:43:00] Robocop holding a child and it says some call him clinker. This feels like the beginning of something right now. It is very goofy that they, they're referred to as clinkers. That is a Gen Z term. But I've seen, like

Kevin Periera: even on Instagram flipping through, like reels and whatnot, there was someone, uh, driving by an autonomous, like a, a cleaning robot that was in a parking garage.

And they, you know, he's screaming out the window, calling it a clank or, and moving on. And it's like we, we, we sort of giggle about this at the moment, but very soon. People will be shouting that and throwing, you know, debris at robots that are walking probably across picket lines. Um, probably delivering food, uh, that are refusing to get out of the way on a sidewalk.

Gavin Purcell: What of the words that Ryan with clancher, I'm just trying to think. There's wanker, right? Which is a good one. You could definitely figure out that. Well, clank

Kevin Periera: Yaker is going to be someone who visits robot. Yank is a good thing. Prostitutes, by the way, we

Gavin Purcell: don't suggest any of these things. Don't make signs with these, but just to be clear, I'm just thinking through this right now.

Banker, there could be something with bankers and clankers. Sure. Yeah. [00:44:00] Um, hotdog franks, maybe that's like, you know, we want a clank or we want a, we, we don't want a clank, we want a hotdog franker. What do you think about that? I'm just trying things out in real time here, Dan. Hey. Hey guys. We're all

Kevin Periera: spitballing and there are no bad ideas.

Remember that. But there are ideas that do get edited out of the show.

Gavin Purcell: No, they don't. That's not, this does not get out the show. Uh, hot Dog Franker has stayed in and you clearly know that because it exists right here. All you buddy. That's it. Um, please go to our Discord and try our new special thing. If you are there.

Um, we will be there getting feedback from you. What are you laughing about? It is a new special thing.

Kevin Periera: Hey guys. Cozy on up to that hole in the wall. We don't know what's on the other side. No, but just wait until you get a delicious special

Gavin Purcell: treat coming from us. You keep your hot dog. Keep your hot dog franker out of that place.

What do Stick it in. End the

Kevin Periera: show. We'll see you on Discord. Bye everyone.