Loading summary
Transcript139 lines
- [00:00]
A
There is a new Google Gemini model out today. It is cheaper, it is faster, and it's also not Gemini 3.5 Pro.
- [00:07]
B
And even though the Gemini team is teasing Gemini 4, we're left wondering, is Google kind of in trouble here?
- [00:14]
A
Meanwhile, OpenAI's Sami Altman has traveled to Washington to brief our dear government on its newest model, GPT6 and some.
- [00:23]
B
Kevin. Kevin, I'm getting. Some breaking news is coming through right now. Some big breaking news.
- [00:27]
A
What is it? Is it GPT6?
- [00:29]
B
No, Kevin, it's much bigger. The robots have solved laundry.
- [00:36]
A
Yeah, okay, that's cool.
- [00:39]
B
Listen, Kevin, we're gonna tell you how Sunday Robotics new Breakthrough will do all of your chores. And more than that, you can watch three hours of robots folding little children clothes.
- [00:49]
A
Okay, that feels wrong. It feels problematic. Instead of watching it, it feels right, Kevin. It feels so right. Don't watch that. Instead, watch District 9 Director Neill Blomkamp. He has a new AI feature that is very interesting. A lot of people very, very unhappy. Gavin.
- [01:06]
B
But also. Kevin. Laundry folding.
- [01:09]
A
Yes, Laundry folding.
- [01:10]
B
This is AI for laundry. Welcome, everybody, to AI for Humans, your twice a week guide to the wonderful world of AI. And Kevin, today there's a big news coming out of Google that some people think is big news, some people think is medium news. And some people are left wondering what the hell is going on at Gemini. Kevin, these are the questions we have to ask. Let us first say what the news is. What is the news that's come out of Google? You tell me.
- [01:40]
A
Well, the big news is it's 3.6 flash, Gavin. And there's a couple different versions of it. There's a cybersecurity leaning one that I actually think is pretty interesting, but this is a new teeny, tiny little model which might have some people going, hey, why do I care about something that is number 12 on the front end? Code Arena, Maybe the forest for the tree you're missing is that this might be the model that runs on a future Android powered phone that you have. And so. Okay. Nope, you don't.
- [02:11]
B
All that's fine, Kevin. All that's fine. And I get that AI better, cheaper, faster, which this is faster and cheaper, which is a really big deal. But, Kevin, why have we gone to 3.6 flash when there was a big thing we were supposed to get around Gemini, which was 3.5 Pro. We got 3.5 Flash. Why did we jump past pro? Do you have an idea?
- [02:34]
A
Well, okay, we got Gemini 3.6flash. We got 3.5flash light.
- [02:40]
B
Yes, we did.
- [02:40]
A
And 3.5flash cyber. That's too many. If you have too many things on the plate, Gavin, you can't bring it all over at once. So I think. I think the waiter's hands were flat.
- [02:50]
B
Oh, that's interesting. So the kitchen's back there chopping some stuff up and the waiter's gotta go cooking.
- [02:56]
A
Yeah. Do you mind with the deep fryer? They're cooking back there. But what is your theory? Why didn't we.
- [03:01]
B
Well, three.
- [03:02]
A
Before we go, let's.
- [03:02]
B
We'll talk about some. Some theories, and there is some news about them cooking up another new model. But we should really quickly talk about Flash 5.6, because you asked a damn
- [03:12]
A
question and now you're like, no, we'll get to that. I gotta.
- [03:14]
B
You can't handle the truth right now, Kevin. You keep yourself. You keep yourself every.
- [03:19]
A
If you're watching this right now, unclick the bell, unsubscribe to this.
- [03:23]
B
No.
- [03:23]
A
And remove the patreon. This is. This is. I wouldn't.
- [03:29]
B
Now put it back on. Now put it back on. You do both. Maybe that gives us extra boost from the algorithm. Anyway, let's talk quickly. Very fast. 3.65 flash cheaper. Faster. It is faster. It is about, I like a third, two thirds of the price of 3.3.5 flash. If you are doing stuff in the API and you want a very smart, cheap model, this is pretty good. The one thing I will say about 3.5 flash, I saw a post from Arena AI and we have had our doubts sometimes about these leaderboards because sometimes there are funky things that end up at the top. But, Kevin, the thing that kind of will lead us into our next conversation here is that 3.6 Flash ended up on the fronting code arena at number 12 below Muse 1.1 Spark, which is Meta's very cheap, fast model, but also below opus 4.6. Thinking. So this is clearly not a move towards the. Towards the front end of frontier models, but I do want to talk quickly about, like, what Google has going on and kind of where it's at right now. I actually asked specifically. I tweeted today when I heard about this. I said, well, this must mean that 3.5 Pro is kind of going out to the woodshed. And Logan from Google AI Studio actually replied to me, said, no, I wanna make sure I get his quote right here. One second. He said, 3.5 pro is testing with partners and will hopefully land soon. And this is also on top of another tweet that he put out today. Exactly when 3.6 flash came out where he said, very specifically, we have started our most ambitious pre training run yet for Gemini 4 and are excited for the progress. So Kevin, it seems like Google is kind of saying like we're stepping back a little bit right now from the
- [05:10]
A
concepts of a frontier model.
- [05:12]
B
Yes, exactly.
- [05:13]
A
That's exactly concepts of a frontier model. Look, I want them to be running in this. I want them making new chips dedicated to serving things bigger, faster, better, stronger. The 3.5 flash cyber model, I don't know if you saw it's pretty low down on the blog post, but it is punching when they let a few run in tandem, it looks like it's punching around where the Mythos preview was or where 5.6 SOL is. And so that gets me. It's a little, I mean, maybe less exciting to a broader audience, but the notion of engineering teams being able to run swarms of tiny little agents that are constantly monitoring for, for threats or doing pen testing, anything like that, that is, I mean that's pretty interesting for this. And again, yeah, this is not maybe the model that you'll reach for to do crazy high end something, but this might be the model that resides on your phone which can quickly generate a user interface for you or whip up an instant app for some need that you have in the moment.
- [06:15]
B
Yeah, I mean all of this is smart and I think ultimately we've talked about as these AI models progress that maybe the frontier, you know, two or three cycles from now we're just going to have everything done for us that we need done. So we may not need to access that. So Google may be setting itself up for like how do we emphasize cheap and fast and sometimes local models as well. It does feel like a little bit that they are either one of two things happening. There's either one thing that they are struggling with their AI. We know that they lost Gnome Brown from Gnome Brown. Is that the right gnome? There's another One of the two gnomes left open. Left Google to go work for OpenAI. But the other thing is that they might actually be saying like, hey, let's not push the gas so hard on this right now. We don't need to be there. And if we spend all of our money and time and resources on this thing, it's going to be kind of a race to the bottom depending on who's the last person to get there. And maybe that's an issue.
- [07:08]
A
Slightly spicy take Gavin. And this is a shout out to audience member of. Well, just. It's an audience of one Logan, okay, what's the time it takes for you to reply to Gavin on X and say, no, no, no, no, we're testing it with partners. Is the same amount of time it takes for you to tell your Hermes bot to add us to the partner program. So if you haven't unsubscribed already, and if you haven't already unbacked us on Patreon, take a moment. Juiciago, drop a comment below and tell Logan to let us become the partners we deserve to be.
- [07:40]
B
That's right. And just to be clear, so everybody's out there either listening or watching Logan's Twitter handle. If you're still on Twitter, I know
- [07:46]
A
some of you aren't.
- [07:47]
B
Is at official Logan K. So let's see what happens.
- [07:51]
A
Can that get us banned? Can that get us banned for the audience to Brigade?
- [07:55]
B
I don't know. We don't know. Brigade, take your time. Do it like. Everybody coordinate. Everybody say, like, every 10 minutes, we're going to put one tweet out. So make it look organic if you're going to do it.
- [08:05]
A
And by the way, no, that. That smell is not desperation. I'm wearing Drakkar.
- [08:10]
B
Oh, that is desperation, though. I didn't realize that's where you got into. All right, we should. Very quickly. Also, Google has announced a new AI chip that's in the works, which is pretty interesting. It is going to be called Frozen. So a little bit of, like, you know, Disney Elsa lore going on here, maybe Olaf. It is going to be six to ten times more efficient. The current TPUs, that's tensor processing Units. That is Google's chip, they have already made for Gemini. So maybe all of this has kind of come together, is a little bit like when Bard was, you know, the kind of, like, secondhand stepchild that was kind of trailing along. Maybe Google will come roaring back in like, a year from now and it will have the best AI solution. So I don't want to count Google out. It just feels like a slightly strange moment right now. Yeah.
- [08:56]
A
And by the way, these chips are. It's kind of interesting. They actually freeze some of the. Yeah, it's the frozen. They freeze some of the model architecture directly into the chip itself. So it doesn't waste a lot of energy or time processing, like, redundant parts, like repetitive parts of the model, basically. What's up?
- [09:15]
B
I was just thinking about how, you know, you and I have been kind of at the. I would say at the forefront at least of being aware of what these tool changes are and how everything changes. And we'll continue to be that way and so will our audience. But one of the things that's really interesting me about the AI space right now is just how much of it is becoming about hardware or physical device stuff that is just like my, I don't know about you, but like that it's not where like my background lives. And it is an interesting thing to think about the opportunities shifting from, you know, software type stuff to real hardware solutions. And a lot of these startups are now going to be like, you know, I think this is kind of sound crazy to some people, but like robots building cities for themselves, like things like that are the things that are coming soon. It is interesting. It's maybe me really much more curious about hardware processing in general and how like hardware is built. And, you know, if you want to do a deep dive on that, there's all sorts of really interesting stories around, like, you know, what's coming out of Taiwan, but also like, what's the future of all this stuff? Maybe we'll talk about that more in detail in a future episode. But it is really fascinating to me right now.
- [10:14]
A
Yeah, I mean, like we have a cursory understanding and interest in all things hardware, but when you texted me this weekend that F to the parenthetical X y z equals parenthetical. Parenthetical 1 plus x y parenthetical.
- [10:27]
B
The Jacobian prefecture.
- [10:29]
A
Is that what you're talking about? Of course that's what I'm talking about is the Jacobian prefecture. When you texted me about that, I was like, what? I didn't even know they were in the finals.
- [10:37]
B
Oh, don't lie to me. You knew that you were, you were three steps ahead of me already. You've already disproven that yourself, I'm sure.
- [10:43]
A
I, I walked by a chalkboard at a university university once and it sat there mostly unsolved. But while I was polishing the floor, as I went ahead and filled out the rest of it, look, I read. Exactly. This is. If you, if you don't know what we're yapping about. This is a big math problem from 1939. That fable, the put on the hats and put on the hats.
- [11:04]
B
This is like smart people talk, smart people talk time.
- [11:07]
A
Which, yeah, put on the tinfoil hat because math isn't real. It's a government construct to get you to buy jeans and listen to Britney Spears.
- [11:16]
B
Oh, wow, Is that right? Anyway, sorry, I didn't mean to interrupt. You keep talking about this because this is really interesting. It's something that fable here's the end
- [11:21]
A
of the road for me. Yes, it's a math problem that I guess Fable disproved and that's a big deal. And that's as far as I got with it, Gavin.
- [11:30]
B
So this is a big deal. I also entered that same world. In fact, I had my AI agents Fig and Moss make a video of it which ostensibly they're supposed to help me understand. I didn't understand their video about it either. This is a big deal though, amongst mathematicians and this was basically what Happened is a 80 year old math problem that has been kind of stymying PhD students for a very, very long time. A very smart mathematician who now works at Anthropic, on the night of the World cup final gave this problem to Claude Fable and Fable disproved it. And that might sound crazy to you, but the thing that was most interesting was like seeing all these other really smart mathematicians chime in on Twitter or on X and say like, this is crazy. This is something that hasn't happened for a while. So again we're seeing these very edge case math problems be solved. But I did see somebody say, I'll see if I can find it and dig it up for edit here. Math is the first thing to be solved. And then all these people, these very smart people, smarter than Kevin or I for sure, are predicting that like math is the building blocks to solving all sorts of other stuff. So if we are on the edge of these models and we'll talk just a second about what's coming next from OpenAI, this is like that turning point where maybe there are significant scientific discoveries coming down the pipe. So that is very exciting.
- [12:47]
A
Look, in between the whole, hey, we're drinking up the ocean. Even though I think that's factory farming and golf, but whatever. In between the we're drinking up the ocean and we're putting up data centers which are making people ill with their low frequency wave. In between all of that, the promise was supposed to be yes, you're cancer and maybe we're gonna like get anti gravity, you know, or we'll unlock something in the quantum realm. Like and okay, where is that? So far we got Will Smith eating spaghetti and some cool vibe codes. But like where is that? And maybe it's on the heels of this, the math.
- [13:21]
B
Yeah, well, you know, following up on this, Sammy Altman, as we said, is on his way to Washington or is already there today according to Bloomberg, and he is walking through the Trump administration. What we expect is GPT6. So this is the next stage that is coming. This is kind of the story and everything we have heard about GPT6 from the people out in the world is that this is a very good major step model supposedly coming in as soon as two weeks, but sometime maybe in August, they're saying. Kevin, this also follows up on the little rumor that you. The scoop lid I called it that you had last week, which is that Opus 5 from anthropic is coming out by Thursday of this week, which might be as good or better than Fable. We are entering that space.
- [14:02]
A
If it's not, I shoot my source.
- [14:04]
B
Oh, is that right? Is that the deal?
- [14:06]
A
If it doesn't come out this week, I shoot the source. Isn't that what.
- [14:08]
B
Okay, Source. You better have delivered here, Source. We'll see what happens. I hope you. I hope you're ready. Source in the world. You know who we're talking to. Source. Anyway, so this is a big deal. So. So I think we're entering again to what I would like to call. We got to come up a name for whatever this fall is going to be because it's going to have to be like, freak out, Fall. Or it's going to have to be like, I've thought of something I didn't want to say there, but like, you know, a sense of like, we're going to have from like, I'd say September to December, a real moment of, okay, we will know now what these are capable of. And it will be interesting to see from both a market standpoint, from a capability standpoint, from a coding standpoint, from a science standpoint, what we actually get.
- [14:51]
A
Well, are we going to get better computer use with ChatGPT or.
- [14:55]
B
That is the plan. That is the plan. Although I will say this, Kevin, one of the things that I found really interesting, and I just wanted to kind of tag this onto this section because I have been using computer use in Codex and ChatGPT work. And I have to tell you, if you have not tried it in say the last month or two months, you should try it again because it is freaking amazing.
- [15:16]
A
Like, it is like, it's far better than Anthropic's.
- [15:19]
B
So much better. So much better. And. And what's interesting is they've solved a lot of dumb problems. Like now there's an in browser, a window in browser that pops up so you're not having to fight the cursor going everywhere else. But like, once you can do stuff. I just did a small thing today. Cause I was curious. I was like, this seems like it's gotten a Lot better. I was logged into a page on my X account and once you're logged in, the computer can go use it through the Chrome tab plugin or through computer use mostly. It's a combination of those two things. It was able to take all that data and give me a really good sense of like, hey, here's a plan for what you do. X, Y and Z. Here's some things you could do. But, like, it is so good at navigating stuff now that we are, I think it's just a really important thing for audience to like. We are very close to the computer being able to kind of do whatever you can do. Right. And I know the dream, it seems like, of all these AI companies is that I could make a dinner reservation for you, which I should try as a thing. But I think we're pretty close to like the big. The big challenge I'm going to give it this weekend is I've been trying to learn the meta ads platform to put a couple ads up for AI for humans just to see. And if you. Have you ever used the meta Ads platform before?
- [16:27]
A
Ages ago, like eons ago, with lookalike audiences and this, that, the other. Yeah.
- [16:32]
B
Have you seen the Odyssey yet? You saw the Odyssey, right?
- [16:34]
A
Oh, yeah, yeah, that's the one. My troll bait tweet is doing numbers. Yeah, that's right.
- [16:40]
B
That's right. I forgot. I think meta ads is like going through the Odyssey there. Every single time I try to use it, I am being stopped by another monster that is known as Field not found or this was the wrong choice to make. And so what I'm hoping is that computer use has gotten so good that it can do something as complicated as basically create and serve a meta ad. Because if it can do that, it can do things like your taxes, or it could do things like, you know, all sorts of complicated stuff. And then we have reached a insanely useful way that AI can help everybody. I will say the other thing that makes me think about this, Kevin, is there was a stat from Olivia Moore, who we're big fans of this week, out of a 16Z, who said that only 2% of AI users pay for AI right now. Did you see that? So if you're not paying, you're not getting access to this stuff that we were talking about. And I'm sure most of our audience is probably, I would assume is paying in some way for AI. But the vast majority of the world, including all those people who are saying that AI doesn't do anything and data centers are Destroying. Blah, blah, blah, blah, blah. Don't know what this computer use case is. So anyway, I was just thinking it's really important to kind of shout that aspect out.
- [17:49]
A
Well, Gavin, I understand not everybody paying for AI, but I still can't fathom somebody not paying for AI for humans. Use your social currency. Click the bell. Subscribe, leave a comment. A five star review on a podcast platform, or you can throw literal money at us at our Patreon. Just go to AI for Humans show or follow the links down below and help us help you. And by help you, I mean, oh
- [18:14]
B
man, help me get off my cough. Yeah, I do want to say really quickly, you don't have to pay with your.
- [18:19]
A
Gavin's allergic. Gavin is allergic to success. Yes, that's.
- [18:23]
B
My wife would tell you very much that's true.
- [18:26]
A
We have been doing numbers. So actually it would be an honor if you would kill my co host, make us so successful that he succumbs.
- [18:40]
B
Okay, we're moving on from the kill your co host comment. We're gonna go straight into what my favorite thing of the AI week has been, was seeing this new demo from Sunday Robotics and they have this very cute little robot named Mimo, which I might have a couple problems with. We'll get into that second, because first and foremost, what's exciting about this, Kevin, is they have released a new version of their Act 2 software. This is not Runway, this is a robotics company. And what this is, is a software that allows these robots to perform a bunch of services autonomously. And Kevin, the coolest thing about this is they have basically solved a few problems and situations in the house all autonomously, which means that the robot is doing its own work. It is not teleoperated. It is also doing this successfully. Up to, I think the lowest one of these benchmarks was 94%, but most of them are at 99%. And Kevin, this is so close to my heart. The thing that I have solved right now is laundry folding. And I know that sounds crazy that that's a big deal, but it actually, when you watch this video of what the memo is doing and how it's folding laundry and like they'll put a piece of clothes on the ground, it realizes that it picks up the clothes, it folds it, it knows how to fold children's clothes. And if you're, if you're a long term listening to this podcast, you know that my life as a parent was, was completely dominated by having to fold baby and toddler socks for a long time, which is a Nightmare. So I am just so excited to see robotics get to this stage. And it starts to make me think, like, how quickly are we going to expand out to more stuff like this?
- [20:09]
A
But you cannot ignore the white paper that is, you know, 94% zero shot, brand new environment, you know, unseen. That's successful. But 6% of the time it folded the actual toddler and that is.
- [20:24]
B
Oh, toddler folding. That's a good title for this today's episode. Welcome to. I just like your reaction.
- [20:33]
A
Oh, it's gain score again. No, I mean, like, it's slow, but that's fine. It's early days. And again, like, that was a lot of the shade people were throwing. They're like, oh, if you. They're playing it back at 20x. If you slow down, it would take hours to fold your clothing. Guess what? I don't care.
- [20:52]
B
Like, if that's what I'm looking at right now, I'm looking at a pile of clean laundry that has sat in my laundry room now for over a day. So it would be folder by now
- [21:03]
A
privileged enough to have your stuff in another area in another room. I don't. It doesn't matter how long it takes. That's fine, you know, now fold around me while I sleep.
- [21:12]
B
Fold my toddler while I sleep.
- [21:13]
A
Oh, no.
- [21:14]
B
Oh, no. I will say one thing about this particular robot is that giant base is a little funky. Like these have these big bases, which is fine for now, but like, in my house that I'm in this right now, in this Airbnb, it would just not be able to get around. So, like, there's some mobility stuff. But again, just a very, very cool look at the future of robotics and when people talk about, like downloadable skills and all the stuff that we, you know. Joanna Stern had a great video this week about a company that's trying to get humans to record cooking and cleaning, to kind of give to robot training. The idea that, like these training models are going to exist, it's going to just get better and better from here. So, like, we are really not far away from what would be originally a pretty expensive thing in your house, but not that far.
- [21:57]
A
And we still haven't really seen V1 hit home.
- [22:00]
B
Yes.
- [22:00]
A
Or offices and warehouses in a meaningful way. And you can ask anyone. V3 is when a product gets actually good, when it's. Especially when it's hardware. V3 is where it's at. So, like, look, there will be some early adopters that have some folded toddlers, then there will be some, you know, wave two Folks that have, like, slightly wrinkly toddlers, but not completely folded. That's good. V3. All of the children live. And that's a good. That's a good place.
- [22:26]
B
That's what you want. And Waymo's almost there, right? Waymo's almost there. They haven't killed a lot of children lately, so that's a good sign. Kevin, we have to move on to what I think is one of the more interesting and exciting moments in AI video in a while. And it's not a new model, although CDance 2.5 supposedly still coming soon. Neil Blomkamp, one of the filmmakers behind District 9, one of my favorite sci fi films of all time, has put his feet fully into the AI waters.
- [22:49]
A
Sorry, Gavin, you said it wrong. It's Neil Slop Camp.
- [22:53]
B
Welcome to the world of criticism. Before we get into the criticism, let us talk about what this is. It is a 14 minute, fully AI generated film by Neil. So he did this himself. He actually announced that he's going to be starting with this in the AI studio, as a lot of other filmmakers are. Darren Aronofsky and a bunch of people like that. Kevin, did you watch this? Because I watched the whole thing. I'm curious what you thought.
- [23:13]
A
I watched the whole thing. I got through 3/4 of it in my first sitting and I was like, I should really go back and watch the whole thing. And I went back and watched the whole thing. And so this is the first drop from Barley Studios. This is his, like, AI wing. It's labeled as such on the old YouTube. There's a little informational panel there. He proudly puts in the title Fully AI. This is someone who, you know, very early on his vfx. Yes, that's. That's what.
- [23:38]
B
Remember Chappie too, right?
- [23:40]
A
Yeah, sure. And then he did. Was it the. Was it Oats? Yeah, it was Oats Studio. Yeah.
- [23:44]
B
Yeah.
- [23:45]
A
Oat Studio was his VFX arm and they had a bunch of shorts that were really interesting. And so, I mean, he's definitely an early adopter. He embraces technology, I think that's fair to say. And there's so many things to be said about this. Like, so, first and foremost, what did you think of it on its own? Let's take the AI away. What did you think of the. How long? The 13 minute short on YouTube.
- [24:08]
B
Yeah, well, here's what I was going to say. I think it is obviously made by a director, meaning that, like, there are shots that look really cool, that you can see that there are choices being Made visually. It does not feel sloppy in that way. One thing I did find, and this is interesting, a lot of people have complained about, like, the AI performances. I actually think this is an interesting example of it probably needed just better writing. It's not. It's not bad. But, like, I actually think the human writing that went into making this, there were a couple moments where the dialogue lines. It's not the performance, the performance is not perfect, but it's more what the characters are saying. And it does make me think. It's like, sometimes directors and writers in movies are a good pairing. And this is like, that's a human criticism.
- [24:48]
A
Right.
- [24:48]
B
It is not a criticism of the content itself. I don't know. What did you think about it?
- [24:52]
A
I get tickled that this is your criticism.
- [24:55]
B
Yeah.
- [24:56]
A
And I. And I. No, really. And I. And I further get tickled by, like, the comment section on YouTube that are like, if you watch frame 15 and then look at frame 16, you will see that the blood splatter moves from the upper left quadrant of the actresses. Excuse me, the model's mouth. Yes. And then goes to the. And I'm like, whoa, have the goalposts shifted now?
- [25:16]
B
Yes.
- [25:17]
A
I think for someone who's so into vfx, I think there's some subtle things that. That Neil and his team could have done to make some of the AI footage look more raw, look more real. I thought it was a bold choice to put. Put like these AI avatars with their fake lavalier mics to kind of put them raw in these interview setups and not affect the footage in some way, because I think that would have covered some of it. There are inconsistencies from shot to shot. Yes. There are issues with the model. There's a dialogue moment where she says, I'm going to go left or whatever in a helicopter. And then it looks like they go right. There's a bunch of little things in there that you can point at. But we are so far already beyond Will Smith eating spaghetti or six floppy fingers or whatever else we are now critiquing. Line reads.
- [26:01]
B
Yeah.
- [26:01]
A
Which were probably text to speech. Right. And we're now arriving to the point where it's like our eyes have become so good at analyzing what is human and what is not that. That now we're starting to really pick apart each frame in the models. I will say, like, I have made way worse content in my life, right?
- [26:19]
B
Oh, sure. With human. With human creators. With.
- [26:21]
A
Exactly. And I have been asked to watch so many short films and everything else from Friends and acting reels. There are performances and I'm using air quotes. There are performances in this piece that blow away some acting and there's some bits that are better written and there's some stuff that's better directed or prompted, if you will. Then a lot of stuff that gets made and that like, you could be mad at that. That's fine. But I don't think you should discount that.
- [26:48]
B
Well, and it's funny what you made me just think of then as polar. As Polar Express. Right. And the Polar Express was a famously. We' talked about on the show before. But it is a famously terrible CGI kind of early CGI movie from Robert Zemeckis starring Tom Hanks about Santa Claus and A train to Santa Claus. The issue with that was the characters look so phony because it was this kind of new technology that everybody was talking about the uncanny Valley and how. Just how creepy it felt. We feel like we're getting past that uncanny valley relatively quickly here. Right. None of it is perfect, but the uncanny Valley seems so much smaller with AI video. And it feels like. I'm really curious to see like when we get to like CDS 2.5, but then to 3 or whatever the next Omni model is. Like maybe that eliminates completely and then it opens up a whole nother world. I think a lot of the haters are never going to change and that's fine. I am excited to see people like Neil attempting to do this and trying to see what longer stories can be told.
- [27:44]
A
And by the way, you know, like along the lines of, of let's say traditional artists, which it gets. Is anybody but traditional artists embracing these tools? You know, Neil's clearly doing it and he's, he's, he's taken a lot for it. Ben Affleck sold his AI studio to Netflix. He's getting a lot, a lot less heat because it was treated as like,
- [28:02]
B
oh, it's a tool for professionals, but
- [28:05]
A
it still generative AI in some way shape or form. And then Notch comes out, who hated
- [28:11]
B
AI, who is a serious, serious AI hater. Who is he?
- [28:15]
A
Gavin? Yeah, exactly.
- [28:16]
B
Like Notch is the guy that made the original Minecraft and then sold it to Microsoft for whatever it was a billion dollars. Founded Mojang Studios, has some problematic stuff in his past, but for right now, we're going to talk about the vibe coding aspect of him.
- [28:29]
A
Yeah, he came out and said I just might give vibe coding a shot, which is like, okay, it has a million views. It happened as of today on this recording. He Says not because I don't like to do programming, but because I've been having a lot of trouble hiring talented programmers. And I'll probably feel less bad about firing a chatbot, which is, you know,
- [28:48]
B
it feels like to me, yeah, I mean, he has been so actively anti AI for so long that this is like, for some certain aspect of the, of the AI coding world, people, this is like a big tree falling, right? Like, this is somebody who has a big platform who has made a big thing that a lot of people care about. Now saying that, oh, maybe vibe coding and AI coding is okay. We said from a long time like, that AI coding was the thing that people would get okay with first because coders are really using, you know, systems and they're. And they're copying other people's work all the time and it's allowed. So this just feels like the natural progression. I, I think another person, Linus Torvalds, recently came out and said, like, if you're not using an. I don't know what the exact quote is, but like, basically said, basically, Linux
- [29:30]
A
isn't anti AI and if you want to keep it anti AI, you can go fork it. But the whole point of Linux is to be the most optimal, you know, badass system it can be. And he thinks AI is going to play a huge role in that and creating features and patches quicker.
- [29:44]
B
So, by the way, the invention of fantastic word, whoever invented fork it for that. Because now I can say, kevin, go fork off. Like, that's the most exciting part. And I will, I will. Okay, before we move on, it is time for the return of AI. See what you did there?
- [30:00]
A
Sometimes you scroll and without a care, then suddenly you stop and shout.
- [30:18]
B
All right, we have some very funny. I see what you did there today, Kevin. First and foremost, I want to shout out a guy named Riley Goodside who does a lot of really awesome experiments with AI stuff. This is just one of those guys who kind of finds his way to do cool things. This was a. A tweet that got a ton of attention. It. He basically made a crazy crossword puzzle with Fable 5 in the shape of a whale using the words from Moby Dick. And this is like a good example of looking at what Fable 5 can do both analytically and visually. And I was just kind of blown away by it. He's also gone on to do a couple other really interesting big kind of visual crossword puzzles. The more that people can do creative new stuff with these tools, I'm always impressed because it's so simple to say make me a Minecraft clone or make me subway racers. Like I'm sick of that stuff. So please, more people go follow Riley.
- [31:07]
A
It's super cool to watch and I hear you, Gavin. That's valid criticism of my camo cursor hiding game. I get it. But I do think I tried to do some interesting stuff with it.
- [31:15]
B
But that's. You did.
- [31:16]
A
That's fair.
- [31:16]
B
You did.
- [31:17]
A
You did.
- [31:17]
B
There's an update coming, right? You want to talk about your update for the cursor game coming?
- [31:20]
A
I mean it's going right now. Yeah, we going to have some private rooms so you can start games with only your friends and share it. But in those private rooms you'll be able to chat, obviously invite bots to the game if you want to pad some of the player count. But then you can upload custom artwork
- [31:36]
B
which, oh, I could hide in your face. I could be like up in your nostril and like. Yeah, okay.
- [31:42]
A
As of right now, each player can upload five pieces of custom artwork, randomly pick from each player's stuff and put them on the level. So if you're like a streamer and you have your own brand or you want to play with meme graphics and your own friends, like I'm going to try not to moderate anything and let it go. I have no idea. I should probably figure out what the liability is there.
- [32:01]
B
But the point is some version of
- [32:02]
A
that, it will only be in your private rooms and when you leave, it all gets deleted.
- [32:06]
B
So cursor cursed is not part of the AI for humans universe. It lives within its own corporation that is covered by a shell company. Just want to make sure we're all.
- [32:14]
A
Any complaints can go directly to gpercelli for Humans Show.
- [32:20]
B
Another cool thing Kevin I saw and this is really awesome and you can go check this out. This is something that I think will impress a lot of people. Vincent Woo made a giant Gossian splat of the San Francisco Grace Cathedral. And what's so cool about this? So if you don't know what a Gossian splat is, it is a way to use AI to make a 360 image that really allows you to kind of get almost like feel like you're at a place or inside a living world. If you go to this site and you are able to kind of see how high res these images are, it really does make you feel like you're living in there. And again, you and I have said this for a bit, but Gaussian splats plus some Version of a working VR or glasses system feels like the future that we were promised a long time ago. Whether or not the world wants that, I don't know. But like it is inching ever closer, I think.
- [33:06]
A
Yeah, I see this with like one of those like real time world models and probably one or two more tools in there to have it. Try to understand what the materials that the splat is made out of so you can deform them. But imagine this and then you get the first person shooter mod and then you could go in and blast the whole thing down block by block, brick by brick. That'd be kind of cool.
- [33:27]
B
All right. And finally we have a new episode in one of my favorite AI video series. If you are a child of the 80s video boom, like I am, you will love this. This is a called Bambi the Destroyer. And basically what this is, this was created by Dr. Dreams, who's a great ex AI video creator. Also has some really interesting interviews with AI video creators himself. This is a 80s style action movie starring a kind of blonde woman in a metal bikini. But it has that vibe of like those old movies. It has the kind of super fun weird creatures. Like there's. There's five of these so far. And he. Dr. Dreams is just doing really interesting weird stuff. You can kind of tell what AG is. He's probably closer to our age than he is to my kids because of his references. But it is super fun to watch. I just really enjoy these and I want to shout him out.
- [34:15]
A
That's dope. That's a good. That's a good find. It's a good watch. And I'm going to take it down from the monitor before April comes into the room. No, sweetie, it was for research. No, I was for research. Oh, hey, last, last. I want to shout out Wizard Brains Vibe Coded a game supposedly in a day that looks like there was an old collection of mini games. Was. It was on the Xbox, I think it was called Fusion Frenzy.
- [34:37]
B
Oh, I love fusion French. It was like.
- [34:39]
A
Yeah, yeah, it was like Mario Party but for the Xbox.
- [34:41]
B
Right?
- [34:41]
A
And so there was all these little games. There was one game in there called Twisted System which was just like you could jump or you could crouch and you were like on a spinning thing and it would get faster and faster. So yeah, yeah, that was one of my favorite mini games. So Wizard Brains Vibe Coded a game supposedly in a day. It is 3D and there is a spinning totem pole, if you will, that has like a dragon head on top that breathes fire. Oh, cool and a little spiky dragon tail. And the idea is you can spin to knock the tail back towards your opponents or you can jump over it. But of course, if you're jumping, you might get caught by the dragon's breath. So the breath is on one side and the tail's on the other. It looks like such a simple little something that, you know, he, he's putting it on steam and hopes to make his token credit back. Like, I kind of wish it were just browser based, but yeah, nevertheless, it was a cool looking something where I'm like, oh, simple mechanic. This could be a lot of fun. Love that someone was able to just kind of whisper that into existence.
- [35:37]
B
I love that too. All right, everybody, we will see you on Thursday, hopefully for some Opus 5 stuff. All right, see you soon. Bye.
- [35:44]
A
Bye. Bye.