← AI For Humans: Weekly AI News, Tools & Trends
AI For Humans: Weekly AI News, Tools & Trends
Microsoft Is Now An AI Agents Company. Seriously.
00:29:46
Loading summary
Transcript93 lines
- [00:00]
A
As predicted, it is a hot agent summer. I hope there's air horns. The entire tech world is rallying around new ways to connect you with corn, pewters.
- [00:10]
B
Even stuffy Microsoft is going fully AI agent first. And we will fill you in on everything they announced at their big Build conference.
- [00:18]
A
From AI agents in your enterprise to AI agents on your lanyard. Even AI agents putting their foot in your child's ribcage. You just set it to play the clip, didn't you?
- [00:29]
B
I did.
- [00:30]
A
I did play the clip.
- [00:35]
B
Plus, new AI image models from Reeve and Ideogram. Show us new ways to make things with AI creations.
- [00:42]
A
And 11 Labs just partnered with Hasbro to give you the voice of Optimus prime and other famous characters.
- [00:49]
B
You know what Optimus prime says, right, Kevin?
- [00:51]
A
I do, but I have a feeling that's not what this is.
- [00:53]
B
AI for humans. I mean, roll out humans, right?
- [01:00]
A
I actually really hope we don't affect the voice at all. It's just you doing that.
- [01:05]
B
This is AI for humans. This is AI for humans, everybody. This is AI for humans. I mean, roll out humans. Welcome, everybody, to AI for Humans. This is your twice a week guide to the wonderful world of AI And Kevin, today, the big stuffy beast that is Seattle's tech giant Microsoft has opened the door. They've opened their kimono to a world full of AI agents. It is clear that we are now entering into the name AI no longer really refers to only chatbots. We are really in a world now where AI agents have kind of taken over, I would say, the entire AI discussion. I don't know. Would you agree with that?
- [01:49]
A
Yeah, look, we.
- [01:51]
B
That's really exciting, Kevin. You were very excited about that. You're like, oh, yeah.
- [01:55]
A
Oh, it's my favorite pastime, Gavin, patting ourselves on the back because we said this would be the year of agents and harnesses. And that is what we got on stage. It's what, you know, Google really championed quite a bit on their stage, and now Microsoft is doing the same. Agents are everywhere. They're mostly in your enterprise, which we'll get to. You can trust them, but they want agents on your desktops. They want agents dangling around your neck. Gavin, that seems like a dangerous place to put an agent.
- [02:24]
B
Well, we know what happened when that kid got kicked, for sure. You don't want to put it next to you in that way. Watch out for your Nether regions. But let's talk about some of the things they announced at this Build conference. Satya Nadella, CEO of Microsoft, got on stage as all CEOs tend to do these days and kind of laid out his plan. There's some really interesting stuff that came out of this. And I think I want to start with the thing they're calling Project Solara. And that is the idea that in every device you own, there will be small AI agents that are doing stuff on dev and being able to kind of operate within that device. I know we're getting closer and closer to the agent in a fridge, which is always the most dangerous place to be across all technology. But there's a cool idea here that like, essentially, if you live in a world of connected devices and all the things from your phones to your computer to like the little thing that's next to your bed, all of those things have agents that can actually interact and talk with each other. We start to feel like the world is getting to be in a place where maybe this stuff will work. Now, I think you and I both
- [03:24]
A
know you remember the promise of the Internet of things where your light switch could talk to your washing machine, which could have a chit chat with your lawnmower, and you could control it all from your phone and then lose access to the app because you forgot your password and now your lights are on indefinitely and the water's always cold. Well, imagine that doomsday present as a future scenario, Gavin, where your light switch and your dishwasher have an opinion and feelings. How great does that sound?
- [03:52]
B
You wake up at your dishwasher. I didn't do the dishes last night because I was tired. I am sorry.
- [03:59]
A
Would it kill you to pre rinse? I'm tired of your bullshit.
- [04:05]
B
By the way, pre rinsing is bullshit. I want everybody to know that. I've got arguments with my family about this. You do not need to pre rinse. Also, cascade, the high end cascade is the best dishwashing soap. Anyway, we are back to AI agents. Let's talk about the soap. Okay, Project Solara is taking technically, quote unquote, a chip to chip platform designed for a open multiple agent world. So again, this is the idea very much like Internet of things, where you could have your devices talk to each other and in a system that works. Kevin, I think it's great. I will say something here about Microsoft versus Apple. There's one thing that Apple still does pretty well. It's not perfect, which is device to device stuff. I have been using a PC for different variations for a few things over the last six months and I still hate Windows. Like, it still feels so clunky to me. So maybe this is a Good thing. Maybe this means that the Windows ecosystem gets better, but maybe more likely in my mind, this could make Windows worse in some ways. I don't know. What are your thoughts on that? Wow.
- [05:04]
A
Wow. Well, it's funny, I'm of two minds. Like I, my, my MacBook is my daily driver and I do enjoy, I tend to like the Mac OS experience more than Windows, but I also have a lot of windows open on my screen that are just command lines. I'm running terminals and clis left and right, so I'm definitely of two minds there. But just for general Windows, like actual windows on your screen and file manipulation, I tend to lean more Mac. That's why I'm so hip and with it today.
- [05:29]
B
Yeah, you got a backwards baseball cap is quite the look. I feel like we're living in just
- [05:33]
A
trying to appeal to a new demo. Gavin. Okay. I'm trying to help with CTR and retention, baby.
- [05:40]
B
Great.
- [05:40]
A
But here, here's what's interesting about this. Like, so, uh, it's not just a device. Even though people, people are confusing Project Solara with being like a smart screen that sits on your desktop, like a Elgato stream deck or something. Yeah. And lets you, you know, tap the screen and talk to it. It's designed to be like an ecosystem, a new operating system where the AI itself can mold its interface based off the, the screen that it's presenting on.
- [06:06]
B
Right.
- [06:06]
A
So for those who didn't catch that, like if you're using the, this Project Solara chip or technology, if you're in a smartwatch, it automatically adapts your agent and your information to that screen, whether you're in glasses or earbuds or this, you know, this little set top device that they're showing off, it's sort of like, oh, it will use AI to generate an interface for that. In, in, in theory.
- [06:29]
B
In theory. In theory.
- [06:31]
A
I think it's fascinating and I love it. And I think we're, you know, we're, we're destined for that in practice. I don't, I, I, I have to, I have to see it, man. I have to see. I mean, here's the Even, even Apple. For all the touting of Apple or even let's say like Android CarPlay to CarPlay or Android Auto to Android Auto. Slightly different screens and slightly different experiences. Granted, they're not predicated on using AI to draw them. They vary drastically and sometimes they're really bad. Yes.
- [06:58]
B
The promise here is that as the AI systems continue to get better and we are expecting an update to ChatGPT 5.6 probably soon, but also Mythos. And as they get better, including all the open source models get better, that the intelligence will be able to kind of figure that out, which actually I kind of buy now in a way that I never bought in the Internet of Things world. Because, like, the goal is you at home never having to install a driver ever again, or never having to go over to your mom's house and install our new printer, which I'm gonna have to do this weekend, which I'm already not looking forward to, but that is where we're at. Kevin, there are a couple other big things that came out of the keynote I do want to say. They brought out the openclaw godfather, Peter Steinberger. Steinbrenn Berger Steinberger. And he talked about Microsoft's new scout agent. So this is an agent that Microsoft has built on the openclaw framework. I think this is interesting. It's just another way of them kind of leaning in and saying, like, hey, guys, we're AI agent friendly too. But what's cool about this is that you can tell Microsoft is really kind of trying to open the door to all the developers in the world. Because obviously Microsoft's platform different from Apple's platform, or even the closed systems of like Anthropic and OpenAI. They are trying to be very developer friendly and trying to open the door to be able to do stuff because they kind of know that's where their bread is buttered, which is. Which is just a cool thing, I think.
- [08:16]
A
Yeah. Fully agree. The notion that he's on stage now saying, you can trust the claw with your data, you can trust it in your enterprise. Yeah, we thought it would go there. Not to like totally backtrack, but really quick. I think the coolest thing of that solara in that, in that announcement was the lanyard like device. Did you watch that portion of it
- [08:34]
B
where they're part of Qualcomm?
- [08:35]
A
Yes. So, yeah, for those that didn't see it, maybe we'll drop a clip here. Imagine you're wearing a lanyard with a screen and it's again, using AI to draw the interface or surface important data at real times. They showed it off for medical professionals. Right. Or for people working in like a warehouse. You could scan a label, get context about the package, perform your thing. You're at the hospital, you scan somebody's wristband. It tells you what their allergies are and whatever else, and hopefully you don't accidentally lop off the wrong foot. Where I get interested was, would this be something I would wear as a daily driver if the interface was truly dynamic and work. And the answer is unequivocally for me, yes, sir. Because I take my phone out, and I'm sure you do all the time, or you put it on the table at a restaurant, or you're walking down a new street for the first time or whatever. And when you unlock your phone, you. It's the exact same interface that was with you in the morning, that was with you in the afternoon. Unless you've set up extensive shortcuts, you're running some sort of hacked os, it's by and large pretty much the same. I want to have this little lightweight something around me that is the screen, the camera, the microphone, and have it proactively suggest, here's the menu, here's the best dishes, here's the price that you should pay for 45 minutes of that massage. If you go through that door, Whatever. The suggestion is, I want a thing
- [09:47]
B
that I can hang around my neck and has just a big middle finger so that when I'm really pissed off at somebody, I can just show them very obviously. So if my AI can pull that off, if they can turn my UX into that. Now we're talking.
- [09:57]
A
You want to. So you want a hand that does this on the screen so you can hold it under your chin as an Italian expression.
- [10:03]
B
Exactly. That's a pretty good app, by the way.
- [10:04]
A
It's a pretty good screen right there. Oh, that is, yeah.
- [10:07]
B
Hey, what is that called?
- [10:10]
A
Sorry, is what it's called. That's.
- [10:13]
B
So anyway, so, yeah, these are very cool.
- [10:15]
A
Claws. Yeah, yeah, claws everywhere. Whether Nvidia is building them or Windows is adopting them, whatever, they've even got new models to power these things. Gavin, Microsoft announced a whole new family of models because why not? From image generation to text to speech, we got some new models.
- [10:35]
B
Yeah. So these are, you know, really interesting. There were seven new AI models they announced. And what was fascinating about this is they are developing these from the ground up. They are not distilled from other companies. And this has got a Mustafa Suleiman, who was the head of PI, if you remember, from way back when, and has been at Microsoft for a while. This is their first kind of like stamp in the ground and saying, like, we are also going to try to build flagship models. Now, they are not benchmarking that great yet, but this is their first stab. And what it's also interesting here, Kevin, is it's like a pulling away a little bit from OpenAI now, Microsoft still owns a lot of OpenAI, which is very good for them. When the IPO happens, they will probably make a fair amount of money from OpenAI, but this is them kind of owning their own future in the AI space. So, yeah, there's a bunch of different AI models. There was a thinking model. There's a bunch of different other models you should check out. It's a. It's kind of dense and, like, none of them jumped out to me as, like, really immediately, I'm gonna go drive them. But, like, you do appreciate the fact that they're out there, kind of like opening the door, trying to do more on their own in this space too.
- [11:32]
A
Now, what is this? You have a future tech Majorana 2 quantum chip. Is this the. Was this the thing that they held up at the end with, like, hey, our cub more stable than our neighbors cubits. And now you can phase matter so
- [11:47]
B
major, it's spelled, it's pronounced like mayorana.
- [11:52]
A
It's mayor's mask.
- [11:55]
B
It's Mayor's mask. So what this is, is if you've been following the kind of edge case quantum computing world, which we have talked about here briefly, quantum computing is still this kind of like, magical world where, you know, I don't go and explain it, but it involves qubits and that there's a qubit that can either be a positive or negative. It can be there, it may not be there. It is a very crazy large idea. Many people believe this is the future of computing at large. It can break down very complicated questions in ways that normal computers can't, but it is very unreliable. Microsoft revealed the next chip in their quantum computer, which supposedly is a thousand times more reliable than before, but this is not really that reliable. Still, there's a lot of kind of like, there's a lot of people surfacing who are kind of saying, like, well, this may or may not be proven yet, but, like, quantum computing is just one of those things. If you want to deep dive on the weekend, go learn a little bit about it. And, you know, it's one of those things, like, in five years, this could be the thing that, like, suddenly shoots us to the stratosphere in terms of once the AIs figure out and get very smart, quantum computing might be the thing that just, like, turns everything that we do into a much more sophisticated system. But it is just for right now, it is kind of like at the early baby stages of what we're talking about. So that wasn't the best, but I think you should go out.
- [13:11]
A
No no, no, I think that's fair. Even on the heels of this announcement, which is, wow, this new chip is so great, they're still saying 2029, this will be maybe viable for commercial uses. And it's like, if you don't know what it is. Yes. Quantum computing is what the AI is going to use to steal all of your crypto and break into your one.
- [13:28]
B
That's right.
- [13:28]
A
But at this phase, you can literally say, oh, it's a thousand times more reliable. That that means it's like it. Oh, this person's a thousand times more reliable than your dad, Kevin. Well, they're still not going to show up to the Little League game. No, let's make that very clear.
- [13:42]
B
My dog just didn't like that. He didn't like you dissing your dad. All right, last thing on this, I do want to shout out the. In the AI agents world, I tried a new AI agent that came out. Everyone you know, you'll see these every. Every day on. On Extra in different places, a company raises 25, $50 million. Do whatever. There's a new company called Town, which you can find@town.com which basically, I think has taken an open claw harness and put a very nice front end on it and is actually very good. So if you need a personal assistant, you want to try when you can try it for two weeks. This is not an ad. I just tried this myself kind of randomly yesterday when I saw a pop up and actually it's quite good. I'm actually thinking about getting my wife on this because it is overall doing a very clean interface. I woke up this morning and you go to town.com and you have a login and it tells you, like, what your schedule is. All this stuff, it's all the stuff that people have been doing an open claw kind of on their own, but. But just packaged really nicely. So I do think we're going to start seeing agents like that kind of like trickle down to different people in different ways. Kevin, I was thinking before we move on from the agents conversation, the one thing I keep thinking about is better agents to make things from an AI generated standpoint. Like, I do think that, you know, Dreamina has a thing called the Octo agent for Sea Dance that just came out. It's getting kind of okay reviews. But I do think there's a world where creative agents are going to be really interesting, especially for, like, specific workflows like YouTube thumbnails or stuff like that. So I think it's really important that everybody starts paying Attention to as the stuff that we've talked about that is super nerdy so far starts to get into production, there are going to be ways that both you can use but maybe eventually build something that either has value for you or other people going forward.
- [15:20]
A
I had a ch and I don't want to spoil the exact topic because I think they might actually release it now, but was just chatting with someone the other day and they said I really want to make a this and I'll just say it was a like a copy of a. An inspired version of an older like classic video game. I just don't know where to begin. And I said, well why don't you ask the AI agent to go and do deep research on that game, every aspect of it from why it was good to how you could modernize it and add some juice or whatever. And he's like well, how do I do that? And I said copy and paste what I just told you to ask it. Just toast it to the agent. And he's like okay, I get it. Be right back. And maybe two hours later he sent me a browser based playable something where he had an idea for a twist on it that was just interesting enough that I was like, you did it? And he's like, oh, I didn't know this was actually possible now. And I was like, you literally just scratched the surface of what's possible right now and you're blowing away like you're blown away. Goodbye to your weekend, goodbye to your next few weeks. Everybody listening needs to really be playing with these things.
- [16:19]
B
So just as an update on last show's my bear jump game, I'm now.
- [16:23]
A
Yeah, yeah.
- [16:24]
B
Hours into my goal, I'll update it. It updates itself.
- [16:28]
A
What's the new goal that you gave it?
- [16:30]
B
No, the goal's it's the same goal. I don't know if I said this last.
- [16:32]
A
Are you giving it the same goal just like just make the best bear game?
- [16:35]
B
This the goal literally is. I said okay, I see it now so clearly. I've had an interaction but the goal is I've got a job for you. Using goal, I want you to set yourself a bar for making the best possible version of this this game. This could be improving scoring, improving mechanics, using image gen, the tool to assets to improve graphics. But I want you to make it so compelling that people could get stuck playing it for hours. So that's my goal. And this goal on in Codex has been working now for 66 hours. And I come back to it. I just let it run on My desktop and I come back to it every couple hours and it's like, done the next step. And I will tell you, I am finding it starting to do really well. So anyway, this is just one of those things where, like, agents are in production. You can use them now. We expect there to be more and more agents going forward. But more importantly, Kevin, the what you should be turning your AI agent on is a very important thing in this world right now, and that is helping us get more popular, because clearly we are not popular enough. We're going through that same sort of like YouTube is trying to figure out what we are. Use your AI agent to help us and use your own fingers to help us.
- [17:39]
A
No, no one just ripped off their AirPods Max and went, wait a minute, these two aren't blowing up YouTube right now. I thought I was listening to bricks and minifigs. What podcast is this?
- [17:49]
B
There's a couple of comments every week that are like, you guys should be millionaires. Yes, we should. But first of all, we would like to be million followed on YouTube. So go like and subscribe to us on YouTube. Do all the stuff. We have a Patreon you can drop a few bucks into. When Kevin and I do these tools, it helps us support ourselves. But more importantly, thank you so much as always for listening and watching. We really do love doing this show and it's super fun to bring it to you. So thank you.
- [18:11]
A
Where are my comment comrades? Dropping that hammer and sickle in the chat?
- [18:16]
B
Is that what we're doing now?
- [18:17]
A
We're going to go. We're going to go old Soviets.
- [18:19]
B
Kevin, we have to talk about a couple of other big announcements. Hasbro has cut a deal with 11 labs for the voices of some of their most famous characters, which I kind of found was interesting to be able to, you know, iterate with these characters. We've already talked about 11 labs doing stuff with, you know, Michael Caine and the voice of Burt Reynolds and Stan Lee. And now you will be able to generate with people like Optimus prime, which is kind of fun.
- [18:45]
A
Yeah. They're creating a new entity called Sixth Wall, which I actually love as a name. And it'll enable them to allow their characters to be used in AI specific applications. And they make this very clear in their big release. And shout out to the Hollywood Reporter, who has a really good story on it. They're not going to use the real actors, AI clones in animations, in cartoons, in games, et cetera. Not just yet. They may do that down the line, but at least at the Moment they're saying, strictly, this is limited to, like, AI specific expressions of the ip, which is great, but they're coming to the conclusion sooner rather than later. Quote, you have a few choices as an IP owner. You could decide to enforce on everything, whack a mole, send a bunch of cease and desists. You could decide to enable the UGC directly with consumers. But the question we posed was, what if we just offered the authorized end to end blue check version of the character that a company could license from us? Boom. Thank you.
- [19:42]
B
Great. That's. That's what we're talking about. You're paying the companies that own the, or the people that own their rights, and you provide them to people to use. There will be different versions of this business model coming forward. I'm excited to see more and more of this stuff happening. I also believe there will be less and less AI hate. I mean, one of the fascinating things about this Martin Scorsese thing we covered last time was that I have definitely not seen nearly as much hate for this. And it is a thing where every one of those things chips away. Right? Like Martin Scorsese saying, this is. Okay, I. I understand. Again, we've talked about this forever and ever. I understand why people are upset and we've talked about the original training stuff. But, like, this is how things are moving right now. And like, opening the door, doing more of this is even better, which I think is cool.
- [20:29]
A
Whatever, bro. Sell out tarnished his reputation and his history. He was really getting somewhere with his, like, his media projects. I think he was on the right track and he just tumped himself over.
- [20:39]
B
Is he me or Martin Scorsese? Marty scores.
- [20:43]
A
No, I thought Marty scores. He was, he was really making progress. And you know what? He's no longer invited on the show. Marty.
- [20:50]
B
That's right. We're no longer inviting him because now he's all in on AI. All right, a couple other big things. Two new AI image models launched. First, we have Reeve 2.0. Reef 2.0 is an image model that's been around. Reeve has been around a little bit. And if it's Rev, I'm sorry to both the Rev folks and the world out there at large, but the idea with this is that this is a 4k image generation. But what's cool, very cool about this is that you can actually edit the images when they're, when they are created. Each individual piece of the image is a movable and editable thing. So unlike other image gens, which just give you a straight out output you are able to kind of like regenerate and move around stuff from within the image and that feels like a very big level up. I will say my first experience with the actual image generation was okay. I think that they, they might be better served as being like a harness to kind of like help figure out how to kind of do these piece by piece. And again, I'm not like starting at it, but Kev, I did share an example of. I uploaded a bunch of our AI for humans branding and I said go make a. Go make a visual of AI for humans, specifically an ad. And it did not great with it. We'll show it here. And then I gave those same visuals to ChatGPT or Image Gen 2 and it actually did pretty good. Now these are both kind of bland ideas. I kind of specifically said use a human, make it interesting like a magazine or a brand ad. But these are pretty. The ChatGPT one is pretty good and the other one is pretty bad.
- [22:13]
A
I feel like I would agree with that. Especially like something as basic as get the colors right if you gave it the same brand kit and identity, like get the core colors right and get the logo right. I don't know what that half melted microwaved logo is on the, the red model. But again, if, if compostability, if layered editing and segmentation of the objects, if that's like their jam, great. If they can improve the core output and the design sensibilities of it. And you know, looking at the examples that they tweeted out, it's like a lot of landscapes, animals, close ups of fruit and food. So maybe saying like, hey, do an elegant print layout. It's not something that the thinking model behind it is maybe a more elegant prompt. But I mean again, There's a reason ChatGPT image 2 is the king, right? Yeah, that's a really solid, solid ad.
- [23:05]
B
And I will say some of the designy looks that Reeve was able to do are very cool. Now again, we don't know how much actual human labor went into that. It may be much better served as a. If you're a designer that you can manipulate in different ways. The other big image model that came out this week was Ideogram four and, and Ideogram four is an upgrade of Ideogram, which we've been covering here for a while. But Kevin, the biggest thing about this open source, open weights, local abilities, you can do all of it and it is actually a really good production image model. So you can use this to do text. Which image Idiogram was always Good at. You can do it, you can do images, you can do all sorts of stuff. I don't know if you've seen some of the examples here, but there's some really, really solid use cases of Ideogram for. And I kind of can't wait to download this if I can. We'll see how big it is and start screwing around with it. It.
- [23:49]
A
This is going to get me messing around with, you know, AWS or run pod, these GPU accelerated, you know, computers in the cloud that you can rent. Because I just gave, you know, 20 bucks to Gemini to unlock better image capabilities for Claude. I, you know, I pay for codec so that it can use ChatGPT's image generation. This is now good enough where I'm like, oh, I'd rather put that 20 bucks into a hyper accelerated version of, of their open source model and I can just generate unlimited bajillion images. Yes. Yeah, yeah. Crazy that they're just giving it away.
- [24:22]
B
Yeah, I use the image not that long ago, which was fine. It's not like perfect, but this is like an upgrade. Anytime you get a real open source upgrade, I feel like it's worth trying out. Yeah, I think there's just some interesting examples from Venture Twins and from Bitbore that you can kind of see. And if you're watching the visuals of this and see what it's capable of. Again, Ideogram is really good at doing text and designy type stuff. So if you want to do a series of text images or if you're trying to create an ad plat, you're trying to create a series of ads or digital ads or Facebook ads or anything like that. It's a very good use case. And to Kevin's point, you can use it to generate, you know, depending on how big of a machine you have and how long time you have like hundreds of possibilities and then you get to choose from those.
- [25:02]
A
I just got. I'm sorry, Gavin. I'm getting a. Sorry, one second. I'm getting a message here on my WhatsApp. One second.
- [25:08]
C
Oh, I'm honestly in love with him and I'm so tired of pretending that I'm not. You know, I just don't really get it why people don't approve of the age gap thing. You know, like if we want to date, just let us date.
- [25:18]
A
I know, that's right. I'm sorry. My AI agent was messaging me. That was my.
- [25:22]
B
Was that.
- [25:22]
A
That was Claudia. That was my open claw.
- [25:25]
B
Talk to April and tell April something.
- [25:28]
A
That was me. So one Gavin, while we're on the
- [25:31]
B
topic of open source, he's so what?
- [25:34]
A
MISO one, let me be clear. The most emotive voice model in the world per them an 8 billion parameter text to speech model for highly expressive speech generation. Here's another. I actually thought that was a good example. Here's their therapist voice. Gavin.
- [25:51]
C
I know, I know, you don't have to explain it, just stay on the line with me.
- [25:56]
A
Startups that accidentally Destroy this is YouTuber Voice Entire Industries. Back in 1999 Sean Parker founded Napster to help people share files like it's not startups I don't like to do.
- [26:07]
B
Yeah, it's actually pretty good. The big thing with MISO one is, is that it's fast, right? And you and I both know this having worked with AI agents in AI audio is that speed sometimes can kill when you're dealing with AI audio because anytime you wait for it to reply. If you've ever been on a phone call with somebody and you're just like God dang it, just get the freaking response back from my health insurance. You can. This actually replies faster than a human can. So that is a big deal. It is again another open source audio model. I do think audio is really interesting right now in that it's bubbling up and I think audio is like the baseline stuff for when we get these real time agents in video as well. So we're not there yet with video agents. I don't think we're not to the place. But all of this is the underlying parts of it which is very cool.
- [26:52]
A
Here's the great news. If you're hearing about all these models and you're going like I want to play that but another subscription fee or oh, Kevin's rented GPUs in the Cloud so that it could generate dumb images for his game. I have a solution for you friends.
- [27:03]
B
Yeah, what is that? Involves burritos or somehow. What is the deal with this?
- [27:06]
A
Yeah, this is on, this is on like the Barbacoa net. This isn't on like the actual Surface BB Barbacoa, but it's, it's called, it's called Chipotle. It's a fork of open code that instead of plugging into OpenAI or into Claude, it can plug into Chipotle's chat support bot, aka Pepper1. And it doesn't stop there. Gavin. I Home Depot, Sephora, Nordstrom, Lowe's, ikea, Expedia, these are some of the companies which have basically left their endpoints for their chat support. Agents, at least as the. Of the time of this recording, like, unsecured and open, so that you can send your requests through the chat support bots and use their tokens.
- [27:52]
B
Yeah. So basically this is a way to hijack chat support bots and. And use their money to pay for their tokens as well. What you're saying?
- [28:01]
A
Yeah. Now, Now. Okay, to be clear, Rob Desen. Rob Desendorf, shout out to Rob said the only issue is that Chipotle already patched it before I could ship. However, the. The INIT API still works. The websocket connects, but the messages go into a void. So listen, it was a silly thing he did, like make a CLI that can connect to all these other chatbots. It's just an example of like, how weird and wild and I think uncharted and unsecured potentially this future is going to be where, yes, you know, we, we had stories of like Toyota chatbots selling Teslas and airline tech support or chat support bots giving away flights for free. This was Chipotle leaving an Endpoint unsecured long enough that someone was like, I can productize this. And I guess kudos to them for patching it. But still, there will be an agent launch tomorrow that lets you generate your next game or get free image credits out of it. So just keep an eye on that stuff.
- [29:00]
B
Yeah, I think the interesting thing with that is just like, whenever you're doing any sort of vibe coding, and by the way, more and more people are doing what you would think of as vibe coding because these AI agents are getting so good. Always do a pass before you release anything out. That is a security pass. Right. Like, and you're not going to be. Even if you're not a engineer, you're not going to be the best at this. But that's just one. A couple good prompts to put in at the end, when you're ready to let your thing out, make sure that it is secure because Endpoint's patched up. It does seem like a company like Chipotle should have figured this out, but of course they're just like, hey, let's try something AI in all these different places. Anyway, just keep your security together. We will see you all next week. And we expect big things. It is hot AI Summer, hot AI Agent summer. See you all then.
- [29:43]
A
Bye.