Every Major AI Browser Update You Missed (Atlas, Edge & Claude Code)

Hey, welcome to the Next Wave podcast. I'm Matt Wolf and today we're doing another episode where we deep dive in the news. So last week we put out an episode with me and Maria just riffing on all of the latest AI news. And people really liked that episode. We got a lot of good feedback about it. A lot of people loved what we were talking about, loved getting that sort of refresher of the news for the week. So we figured, let's do it again. People like it. Let's come back to this well and make another episode and break down all of the news that's happened in the last week. So once again today, I've got Maria, the head writer of the Mindstream newsletter, joining me on the show today, and we're going to deep dive in the news once again.

Using only 20% of your business data is like dating someone who only texts emojis. First of all, that's annoying, and second, you're missing a lot of context. But that's how most businesses operate today, using only 20% of their data. Unless you have HubSpot, where all the emails, call logs, and chat messages turn into insights to grow your business. Because all that data makes all the difference. Learn more@HubSpot.com.

So thanks again, Riya, for hanging out with me.

Thanks for having me. I was so happy that people liked it so much. I mean, a lot had happened last week, so we have a lot to cover today.

Yeah, yeah, there's been a lot of news. And, you know, anybody who does pay attention to, like, Both my main YouTube channel and the Next Wave podcast, you might have noticed that I actually slowed down on making, like, the sort of AI news roundup videos over on my YouTube channel. And I feel like this style of podcast can kind of replace that format a little bit. So if you're missing out on getting a lot of that, like, AI news, deep dive in the sort of bigger picture of everything that happened in the AI news world. Episodes like this are designed for you 100%. So if you haven't listened to that episode, definitely check that one out. You'll get a little bit deeper of a dive into who Maria is and her role with the Next Wave and HubSpot. But for this episode, let's start chatting about some of the cool stuff that happened this week. I feel like the biggest story of the past week or so was OpenAI's atlas. Oh, yeah, it kind of came out of nowhere. Like, I didn't know they were planning a livestream for Atlas until the Day of the livestream. So that was my sort of entry point to it. Didn't even see it coming.

Yeah, I mean, when it comes to Sam Altman, it's always the element of surprise. We thought, like, Mark Zuckerberg does that. But no, it's actually Sam as well. So I was really surprised by the launch. I saw, because I do my research for the newsletter and I saw it, like, as a breaking news. I was like, what do you mean? No one told us anything.

There's been rumblings for a while that OpenAI is working on a browser eventually.

Like, not like three days ago. So that was weird.

Yeah, it definitely came quicker than I expected because it was only, like, I don't know, maybe two months ago that OpenAI was literally trying to buy Chrome from Google.

Exactly. Exactly my point. Yeah.

But I guess underneath the hood, it's still Chromium. It's still like the bones of the Chrome browser, just with, like, OpenAI features baked into it.

Yeah, I think with ChatGPT, I think it's no longer your chat box anymore. I think, like, OpenAI just decided to turn it entirely into a browser. So the idea is that instead of switching tabs and pasting links and pretending to multitask, because a lot of people pretend to multitask, not everyone is very productive, as all of us. ChatGPT kind of follows you around and follows you everywhere. So you're scrolling through a recipe, it can order the ingredients of that specific recipe. And if you're reading a job post and you don't want your manager to see, also, it can summarize it. So if you want to plan a trip, it'll start building the whole itinerary, which is pretty wild. Basically, Atlas wants to be that friend who finishes your sentences, but, like in a browser kind of turn, it's huge. It's wild. Atlas isn't just answering questions anymore, it's actually doing the things. Because we were used to chatgpt answering everything for us, now it just does everything. It got agent mode built in, which basically means that ChatGPT can click Research, plan, and even book stuff while you're still browsing. And before anyone asks, because people are gonna ask this question, it's built with optional browser memory, so it can remember what you were working on.

But it does have an incognito mode. So if you want to be browsing something that you don't want it to remember, you do have that capability as well.

If you wanna be shady about it, it feels like OpenAI finally kind of stepped out of the chatbot phase and into the workflow sort of phase. And everyone's been saying AI will live.

And this is exactly it. So OpenAI is more or less saying that we're done being just an app on your phone or like app on your laptop or whatever. We're becoming your own desktop. And the timing is very smart with Chrome and everything. And like with Claude and Gemini getting louder, I think Atlas gives them a serious sort of edge, especially for like enterprise users who want AI to actually do things, not just, you know, talk about them.

Yeah, the way I see it, where I think OpenAI is kind of going with this is I think they want to get it to the point where you don't even use your browser anymore. Like, your browser is just a tool that goes and does stuff for you. That's where I think they're trying to go with it. So right now you get into the Atlas browser, you can chat with it in the little right sidebar that pops up. You can turn on the agent mode if you're on one of their paid plans and say, hey, like you just mentioned, right, Read this article that has a recipe on it. Go find all the ingredients for me and order enough ingredients so I can make this recipe for 20 people, right? It'll do that math. Figure out like the amount of ingredients, go to the store, order them for you, and then stop short of making the payment, right? It'll say, hey, I put it all in the cart. Do you want to make the payment?

Do you want to Apple pay? Like I could just Apple pay us. Yeah, exactly.

And so right now you're typing in this little sidebar, right? But I think what they want to do is they want to get it to a point where once they roll out a mobile version of it and they get the, the audio features more dialed in, you just prompt everything with your voice. So I can be hanging out with a friend and I really like the shoes they're wearing. And I'm like, oh, those shoes are really cool. Let me order some. Take a picture of the shoes and go, hey, Atlas, order me a pair of these shoes. It looks at the image, figures out what the shoes are, does the research on where to find them, finds the best price on the shoes among multiple outlets, and then places the order for you. And all you did was give it, you know, a spoken prompt into your phone. Like, I think they want to remove actually using the browser. I think that's the long term picture for them is actually make it so we're not even thinking about using a browser anymore, just make the Internet prompt. I give it a prompt and it can go and take actions for me in the browser without me needing to open the browser and do the things anymore.

100%. I cannot wait to steal someone's look. I have one person in my life that I love. Everything that she wears, I cannot wait for that day. I don't have to ask her every single time. So I'm going to do this. So this is a win, win for everyone, in my opinion.

Yeah. And then, you know, there is a sort of more pessimistic take that I have on things. I mean, it's not overly pessimistic. I just know that OpenAI, they need to figure out how to drive more revenue right now, right?

Yeah.

I mean, they're losing billions of dollars a year. I think every prompt you enter into something like ChatGPT, they actually lose a little bit of money on every time you enter a prompt. So they actually really, really, really need to find new monetization methods. And by having, you know, an app like Sora, where you're scrolling a feed like TikTok or Instagram, right. That opens up a new platform where they could put ads into it. When you have a browser that more and more people are using, that opens up another platform that you can put ads into. Potentially even opens up, like, recommended products. Like, I just gave you an example of me just talking into my phone and it going and ordering stuff. I mean, is it too far to believe that at some point OpenAI might partner with certain companies and say, like, when they order toilet paper, make sure they go buy Charmin. Right. Like, would it be so out of left field to believe that they might make a deal with a company like.

That they're gonna monetize it? Honestly speaking, you did say that they are losing money. I think the monetization part just is like a walk in the park for them. So I think they're gonna monetize everything. If everything is going to be accessible to everyone on their phones, especially when it becomes an app, everything is going to be monetized. Every ad that you see will be monetized.

So, yeah, yeah, absolutely. Absolutely. And then, you know, you also mentioned, like, the timing of it was interesting, Right, because it's only been a couple weeks since we got Perplexity's Comet browser.

Which is not free. Atlas is free, but Comet isn't, you know, so, like, that's also a thing.

I thought they just recently made it free for everybody. Within the last couple weeks. I Think so. If I remember correctly, they did open it up publicly, like to sort of get ahead of OpenAI.

Why don't I know this? I may not know everything as a journalist, it seems.

Yeah, yeah. I mean there's a lot, a lot.

Of new flags are happening. A lot.

Definitely easy to like, be nice to me.

Yeah.

But I'm fairly confident that you can use Comet now for free. But there are some features that are only available for paid members inside of Comet. Yeah, there's also dia, which is the one from the browser company, which I do believe is paid. I don't know if that one actually has a free version.

Maybe freemium maybe.

Maybe it is free to download. But some of the features are paid. Honestly, I haven't actually tried that browser, so I can't.

So many browsers out there, you can't just try everything. Honestly.

And then also Microsoft just rolled out a whole bunch of AI features into Edge. Google just rolled out AI features into Chrome. So pretty much every browser is going in this direction of the AI browser right now. But you mentioned something that you wanted to show off, an interesting thing that happened with Atlas that I'm kind of curious if we can take a look at.

So I was playing around with Atlas yesterday and I'm going to London next week. So I was like thinking to myself, like, I don't want to go to London and just have like any brunch. Like I want to have a brunch that feels like I'm a London girlie, even though I am not. So I said to Afla, I was like, treat me like a Napo baby that lives in London and plan out a day for me to have in the city with top notch restaurants from 9am to 10am this is how it talks to me. I was so happy when it started talking to me. It basically like gives me like a whole itinerary of everything. So that's what people should know. It gives you ChatGPT, but in your browser. So yeah, it's pretty cool.

And if there was anything on that list that it made for you that might need like a reservation or anything, theoretically you should be able to turn on agent mode and be like, hey, go make any reservations for me that need to be made to achieve this day.

Yeah, 100%. I love it. I love it. So I can't wait to try it. So that's gonna be awesome.

And then somebody from OpenAI named Adam Fry. Okay, so they're the product lead on ChatGPT Atlas. They actually made this post basically saying These are all of the fixes that they're going to be making from all the feedback they got. This is the fixes they're going to be making, which is also a pretty solid list of like the issues that you would find with Atlas. Right now I'm going to read these fixes out loud just for the people that are listening on the podcast. So this is just like the rapid fire list of fixes OpenAI claimed they're going to be rolling out into Atlas soon, so fixed text entry for Japan and Korea captive portal for WI Fi needs to work Multi profile support Tab groups, Model picker and Ask GPT Sidebar Multiple tab attachments in Chat Composer and improved mentions User experience Use projects from Ask ChatGPT sidebar that'll be a cool one to be able to access your projects right in the sidebar an opt in ad blocker Add a menu listing all shortcuts bookmarks Overflow menu Add speed bump before deleting all chats in browsing data dialog Improve personalization of suggestions Keep improving Agent time to first message improve under triggering of ChatGPT using agent mode Make Agent pause state more reliable Improve chain of thought animation for different agent actions and improve Cloud, Excel and Google Drive use on Agent. So that's the full list of things they said they're actively working on, which is also a full list of things that you might have run into if you've used Agent mode and had issues with it.

100% agent mode for a lot of people is like sounds like magic. For people that have started using it, like people that aren't very tech savvy when they start using Agent, they're like, they're wild. Imagine having Atlas as an agent that's 100% more wild, you know.

Yeah. So I think that actually the one thing about Atlas and we can move on from this topic because I know we spent quite a bit of time on it. The thing that I found the most valu when I was using Atlas is probably the thing that most people are scared of was the ability to go and have it search my chat history for something I was looking at a few days ago. Yeah, because it will import all of your history from Chrome if you want it to. And I actually asked it something like I was reading an article on TechCrunch the other day about AI agents. Can you find that TechCrunch article for me again? And it searched and it was like, well I found three articles. Maybe it was one of these. And sure enough one of the articles that it listed was one of the articles I was looking for. And I was like, that, to me is so helpful because I don't know how many times I've literally scrolled my history looking for a website I visited a few days ago, and for whatever reason, I can't find it again in my history and I can't remember what the domain was. So I'm like, ah. But now I can just do it with my chat. I can just say, hey, I was looking at this site 3 days ago, what was it again? And it pulls it up and I found that so helpful.

It is helpful. It is. Honestly, a lot of people don't have really good memory. Imagine a hack just scrolling through and like finding the right one at the right place at the right time. That's like, so helpful.

Yeah.

Especially if you're like, at a workplace and like your manager told you, like, remember that article? And like you're scrolling through it and like, you couldn't find it, but then alphast does it for you. Yeah, promotion.

I guess that can work the other way too. You leave the office, your boss is down his computer. What else was he looking at?

What else would he. Yeah, but.

Yeah. So speaking of browsers, though, Microsoft rolled out a new version of Edge that has a whole bunch of AI features in it. And I can't help but think it feels like a response to OpenAI's atlas.

It is.

It feels like Microsoft going, all right, that Atlas, they got a lot of buzz. Let's roll out our browser now.

Yeah, and there's a lot of features in this one also, like, there's like 12 new features, which is like, also a cute little orb that is called Maiko. And apparently it's like a whole AI philosophy about optimism and a time of like, synonymous. And like. Yes, but are you okay? It looks like it's not okay. It looks like it's being pushed to say some stuff, to be optimistic and things, even though the world is not optimistic as we speak. So it feels like it's trying to make you feel better, but on the inside it's like being destroyed. You know, like, this is. But it's a cute orb and the name is also cute. Maiko. You know, it's like Clippy, but like the cousin of Clippy, which is.

Yeah, yeah, yeah. And you can actually make it become Clippy if you want.

Yeah. If you want to go back to the ancestor of it. Yeah.

Actually, let me pull it up here. We can actually take a look and show it off.

Look at it.

So inside of the Microsoft copilot. These are the new features here. This is Maiko. There's a bunch of other new features too. I'll talk about some of those in a minute. But let's play with Maiko here.

So cute.

What's the most interesting thing to happen in the world of AI news this week? Okay, that's enough.

Okay, yeah, that's.

We're getting into politics. That's plenty.

Wait, did you see the demeanor of the blob? Like, it's like it's suffering on the inside. So I tried to. Okay, so this is what happened today. I know you want that, but, like, I'm tired today, so let me give the answers. Go back to sleep. This is the vibe that was giving.

Yeah. Oh, you again. Here we go.

What do you want to know?

Like, I just wanted to show this off real quick. If I just like beat it with my mouse over and over again, it eventually becomes clippy.

It's clippy. Yeah. Welcome back, bro.

So, yeah, that's the new meco. You can kind of talk to it. They made it playful and sort of fun and you can tell it to change its colors and stuff like that. But I think the more interesting stuff that Microsoft rolled out is they actually rolled out memory this week. So you know how in ChatGPT we've had memory for a little while now where it remembers all of your conversations and can sort of reference past conversations you've had? They actually just rolled that out inside of Copilot now, like, that hasn't been a feature yet inside of Microsoft until today. The other thing that's interesting is you can do these collaborative chats now, so you can be in there talking to their chatbot. You know, in my case. Right. I might be planning the next YouTube video and having it do some research for me on my next YouTube video. And then I can pull my producer from my YouTube channel, Dave, in, and say, hey, here's the ideas AI came up with. What do you think he could get in there, start actually having a chat with the same AI conversation I was just having a chat with and keep that conversation going. And so now you can have like a three way conversation. That's me, another human, and Microsoft's copilot, all sort of in the conversation.

Yeah. It's wild that we've come to the point where this is actually happening.

Yeah. And the funny thing is, like, you know, OpenAI released Atlas earlier this week, and then today Microsoft rolled out the ability to share your chats and bring other people in. Yeah, well, OpenAI responded this morning as well, so check this out. So this post is from at the time of this recording, like an hour ago. Right. But it says shared projects are expanding to Free plus and pro users invite others to work together in ChatGPT using shared chats, files, and instructions all in one place. So Microsoft rolls out copilot with the ability to bring other people into the conversation. And OpenAI went, oh, yeah, well, we do that now, too. On the exact same day.

Yeah, I think the tech war, they're kind of speedrunning the whole thing, and it's like just butting heads right now. But if we can come back to Copilot, I think it's more like getting smarter across Windows and Edge.

Yeah.

And a lot of people that use Windows are like, use basically any Microsoft. They're having the time of their lives, you know, because, like, it can summarize things. It could summarize your tabs, it can book hotels, it can fill out forms, it can do many things. And if you want to tell it. If you want to tell Copilot, can you fix my life? And it's just organize your whole calendar. And that's honestly, it's kind of refreshing to see a company talks about AI as something that you should connect people together, not just making us stare at screens the whole time. They're betting on warmth and memory and good vibes. Like an AI assistant that remembers your anniversary and your trauma at the same time. This is what I want to see. Yeah. I mean, it didn't really lack it in OpenAI, but now that Windows has it also, it's like everyone's just like, who has the better agent right now.

Yeah.

Hey, if you take a look at my web presence online, it's safe to say that I'm a bit AI obsessed. I even have a podcast all about AI that you're watching right now. I've gone down multiple rabbit holes with AI and done countless hours of research on the newest AI tools every single week. Well, I've done it again, and I just dropped my list of my favorite AI tools. I've done all the research on what's been working for me, my favorite use cases and more. So if you want to steal my favorite tools and use them for yourself now, you can. You can get it at the link in the description below. Now back to the show.

Well, Microsoft has an upper hand, too, that they also run the operating system that so many people run on as well. So right now we're getting a lot of this AI in the browser But I mean, that's the sort of next level for Microsoft. Windows 11 is sort of getting rolled out into, like, every computer now. I think they're deprecating Windows 10. So it's going to be harder and harder to actually use Windows 10 if you still want to. But Windows 11, they're baking more and more AI features into it. So a lot of, like what we're doing in our browser, theoretically, pretty soon you'll be able to do directly inside of Windows, like go to Windows and say, hey, my computer monitor is too dim. Can you brighten it up for me? And it just goes and adjusts the settings for you. Or, hey, I need a new desktop background. I want to use something from the Simpsons. Can you find me a good desktop background? And it goes and searches the Internet, finds a good image, and just replaces your desktop background for you. And everything you do with your computer, your operating system, you're just prompting it to do it instead of having to dig around in the settings anymore.

Exactly.

So I think that's where we'll see Windows go. Everything becomes promptable on the operating system.

Can you imagine if that rolled out like 40, 50 years ago? It would be deemed as witchcraft. So I'm very happy.

I mean, if we had iPhones 40 or 50 years ago, that would be witchcraft. So another story I wanted to talk about from the past week is that Claude code is now inside your browser. Now, I don't know how much, you know, vibe coding type stuff you've done yourself.

Never.

But I've played around with a lot of the various AI coding tools, and for the most part, I've found Claude code to perform the best out of what's out there. Claude code seems to work really well. The main problem with Claude code is that you had to run it in a terminal.

Yes.

So you either had to run it inside of your terminal on Mac or Windows, or you could run it inside one of the terminals inside of an ide. So, like Visual Studio Code, Cursor, Windsurf, you know, tools like that that have a terminal built in. You're able to fire up one of those tools, then get Claude code running, and then you're dealing with the whole thing inside of a terminal. That part was a pain in the butt and definitely scared a lot of people away from using Claude code. Well, now Claude code, they just have like a browser interface where you can mess with Claude code now.

Yeah. Also for the people that don't know what IED is, it's basically like a developer's all in one workspace. So think of it as like a Microsoft Word, like a Google Doc and like a Finder combined, but for coding. So for people that don't know this, but yeah, as you said, Anthropics decided developers deserve fewer tabs, you know, and like more sanity. So they did cloud code and like it's now running in the browser. No terminals, like no local setups, no chaos. So you just describe what you want and it connects your repo and Claude starts coding like the world's calmest intern ever. That is like being just given 7,000 caffeine cups and stuff. So now it runs in the cloud and it does real time debugging and ships a pull requests and even tells you what to change. And it's basically like it's GitHub Copilot, but with an intention, like a really good attention span. And it's not just like for hardcore devs. The fact that you can do this straight from chat means that anyone could explain a book in English, in plain English and suddenly has access to engineering power. So you don't have to have a background in that, you can just do it, obviously. And it's the kind of feature that quietly eats away ides once you can code from your browser. V as code starts looking a little 2019, you know. So for people that want to kind of experience the world of coding, this is wild stuff. So yeah, everyone's going to be happy about this.

Yeah, it's a lot, a lot, a lot more user friendly than what Claude code was. Right. You basically had to go to a website and it gave you instructions of like other dependencies you needed to install on your computer. So if you didn't have Python installed on your computer, you had to install Python. If you didn't have like Node JS installed on your computer, you had to install Node js. Right. There's all these little like development elements that you had to install on your computer. So you had to follow these like step by step instructions and do it all in the terminal. And if you miss one step or mistype one thing, nothing seems to work. And then you want to pull your hair out because you can't even get into Cloud coded the first place to start writing the code. Yeah, well now they just simplified it. Right? We've got this.

I have friends that do coding and like they're engineers and stuff and I look at them as if like they own the world. Like they can do many things, things. And most of them are women, which is badass, but that I can't do that. And so, yeah, it is witchcraft. They should be persecuted. But this is awesome. This is amazing. I think that it could like, unlock a lot of people's potentials. People can do many things now and like kind of create their own apps and stuff, so.

And if you do anything with code, you're probably familiar with GitHub. It's like a place where as you're writing code, you can sort of push your code there. And it really serves a few purposes. One, it's like a backup for your code. So if, you know, you go too far in your code and you're like, oh, you know what? I liked the version three versions ago better, let's roll it back to that. You can actually just pull the old version off of GitHub and start from there. But it also is a platform where if you write code that you want others to be able to access, you can share that code with anybody. They can download it and use it. And so what's really cool about Claude code is it integrates with GitHub just right out of the box. Like one of the first things it says is connect your GitHub account. And so anything I've developed on GitHub in the past pops up here as something that I can just pick up where I left off. So I made this AI jump game like two years ago. I can select that game, give it prompts, and it will just the code from GitHub of the game that already exists and just let me start coding from that spot.

Yeah, you can basically continue what you've done before without having to do it from scratch.

Exactly. So pretty useful upgrade.

Thank you, Anthropic. We love you, Anthropic. You're the best.

And Anthropic makes the best coding models too. They're called Sonnet 4.5 and Opus 4.1 are pretty much considered the best AI coding models out there.

For the most part, 100%. Do you listen to that? OpenAI or like, is OpenAI not in the picture anymore? Like, they should do something about this.

Yeah, I feel like they've kind of diverged in two separate directions.

For now.

Yeah, for now we're going to go more in that consumer direction. We're going to build social media apps, we're going to build browsers, we're going to build that easy user interface, anybody can chat. They're going that consumer path where Anthropic they realize their models are the best for coders. Let's just really go heavy down that path now. I'M not saying they're not thinking about consumers as well. Right. They do have, you know, the regular Claude website where you can use it just like ChatGPT. But I feel like their main focus is let's focus on coders. Like as far as the amount of use a large language model gets, you've got the OpenAI models and you've got the Anthropic models. There's graphs out there that show which model gets the most usage. And Anthropic blows OpenAI away. Like people are tapping into the anthropic models way more than the OpenAI models, but that's just because it's all through API. There's so many companies tapping into Claude's API where OpenAI most of their use is through ChatGPT, through their front end user interface. So like, in terms of like profitability and getting the most use, Anthropic is actually in quite a bit better position than OpenAI. But that's just some behind baseball, like really nerdy talk. But Anthropic gets so much more use than OpenAI because it's so good at coding.

Yeah, yeah. I sincerely, as I said, I don't like to see coding in any shape or form. It triggers my anxiety. So I have so many things what I want to do, so I'm going to probably play around with it just to see what it comes up with.

Yeah. And I love like the AI coding for just making like little simple apps. I mean, we've talked about this in many episodes in the past on this show.

Yeah.

This is a little bottleneck I have in my business. I can write a quick little script that fixes this problem for me and then I never have that problem again. Right. Like I've made a little Python script.

Yeah.

That will take any image file and convert it to a jpeg.

Exactly.

Sounds really, really, really simple. But it's just like a little box that's open on my window and I drag any image into it or bulk images. I can drop like 20 images and if I want and it just converts all of the images to a jpeg. So if you get a webp file, if you get an AVIF file, if you get a all the random image files you get when you do like a search for an image and then you download the image and you're like, dang it, this is a vector file, it's not a jpeg. I can't use it on my website. Whatever. Right. I made a little script where no matter what image I get, drag and drop it in here, converts it to a JPEG that will look good in my videos, look good on a website, et cetera. And it was just a little script that took me 10 minutes to build. And I vibe coded it and I would never sell it. It's really, really basic, but it solves a little problem that I had in my life and made that little thing.

A little bit easier, Makes it convenient, makes it easier, and makes your life 10 times easier. That's the whole idea of why AI is just, you know, wild to a lot of people. And, like, I understand that people feel like it's taking away some of their jobs and stuff, but it's actually very helpful if you know how to use it. And it's not just taking away some stuff, it's probably creating more things to do. So.

Yeah, yeah, yeah. I know a lot of coders are really worried, but I think right now is like, probably the best time in history to be a coder because, like, you know what you're doing with code, but you also have AI to assist you. And if AI can assist you and make things a lot faster and you.

Can create like thousands and thousands of things. Yeah.

But then when AI screws something up and it's like, okay, I'm running into this bug that I can't figure out how to get around. Most vibe coders like me never figure out how to get around it. But if you actually know how to code, you're like one of the few people that goes, okay, I ran into this issue. I'm not going to vibe code the solution because I know how to fix the thing myself. Right. So I feel like right now is actually a really good time to be a coder. Like that person that can jump between using AI to help code, but then when they run into a roadblock, jump in and just manually fix the thing that AI messed up.

I've seen people debugging before. If it was me and their players, I would have set my laptop on fire. I would never and my life put myself in that position. There's no amount of coffee in this world that would make me kind of sustain my calmness when it comes to. No. So, yeah, this is helpful.

Absolutely. You mentioned the next topic that we're going to dive into. There was this article circulating. Amazon plans to replace more than half a million jobs with robots. Internal documents show the company that changed how people shop has a far reach plan to automate 75% of his operations. Yeah, yeah. So this is pretty much the thing about AI that most people are most scared about. Right. Is AI taking all of our jobs. Right.

Yeah. Understandable. You know, like, we're not saying that people shouldn't be anxious about stuff like that. When it comes to evolution and revolutions in general, these kinds of stuff happen all the time now because of technology and how everything is, like, more convenient. It's happening faster and faster than we anticipated. But when it comes to this specifically, it doesn't mean, like, mass layoffs. It just means that instead of hiring another half million people, Amazon planning to lean onto its robots and, like, robot workforce to keep up, you know, with demand, which is translation for, like, more orders, fewer coffee breaks. So the company is already testing new warehouse systems that can pack and, like, sort and move products faster than ever. Saying about, like, 30 cents per item, I think, and potentially $12 billion in costs by 2027. So that's not obviously surprising. Robots don't need overtime pay or, like, motivational posters. They just need charging docs. Also, to be fair, Amazon says that these levers come from internal teams rather than the whole company, and they're still hiring thousands and thousands of people for the holidays. What's interesting here is that there's a language shift, so they're using softer terms like advanced technology and like cobots instead of automation, which, let's be honest, sounds way cuter than robotic efficiency initiative. But this really shows how the next wave of automation will work instead of replacing it. So robots handle the repetitive stuff, which humans can handle the creative or complex parts. And together, it's like a complete different kind of workplace. And if Amazon can pull that off at a scale, it sets the blueprint, obviously, for how AI and robotics kind of quietly blend together when it comes to industries and everywhere. Not a takeover, but sort of an upgrade. That's me.

Yeah, this is a little bit clickbaity to me to say Amazon plans to replace more than half a million jobs with robots. Because if we read this, it says Amazon's automation team expects the company can avoid hiring more than 160,000 people in the United States it would otherwise need by 2027. And then down here, it said executives told Amazon's board last year that they hoped robotic automation would allow the company to continue to avoid adding to its US Workforce in the coming years, even though they expect to sell twice as many products. That would translate to more than 600,000 people whom Amazon didn't need to hire.

Yet.

If you read this headline, it sounds like Amazon is dumping 500,000 people. But when you actually read the article, it's like, no, they're building automation so they don't need to hire as many people as they used to need to hire.

Honestly, it's just the way that journalism works these days, and not just, you know, to promote us, but with my stream is that we never say that. Like, we never put in these headlines where people should be scared of what's going to happen tomorrow because no one should be fear mongering anything. We're trying to make people like AI and like work with AI rather than make them have a heart attack because someone woke up and like decided to write something so weird like this, you know?

Yeah.

We need to lessen the triggering and make people feel more at ease when it comes to artificial intelligence, in my opinion.

Yeah. And it's just sort of like a truth too that so many people out there just read headlines and don't read articles. Right. They read the headline and infer what the rest of the article is going to be about based on the headline and then they're, you know, out in the world talking to their friend going, hey, did you hear Amazon's gonna lay off 500,000 people because of AI?

Not everyone's gonna click and like read the article. Even if it's clickbaity, the headline itself could be enough for them to kind of like form an opinion about the whole thing. So people should be careful with the words that they're using these days. Yeah, in my opinion.

Yeah. And I mean, it's so easy for somebody to just like take that headline, screenshot it, post it on Twitter and then it goes viral on Twitter and all anybody knows is the context of the headline and nothing else, you know? Exactly.

Yeah.

It's a sort of shady approach. But we know the truth.

We do, we know that. Yeah. They're not replacing people.

Yeah. All right. So there's a handful of other things we could kind of do. More of like a rapid fire sort of like just quick thoughts on each one of them. Here's something that just came out the day that we're recording this. It is a new AI video model. There's this new model that came out by LTX called LTX2. And here's what they say about it in their sort of launch tweet. Synchronized audio and video generation. So it can do what VO3 and what Sora do, where it actually generates with the audio.

Oh, wow, this is wild. Yeah.

Native 4K fidelity, up to 50 frames per second. And one thing is it's not actually generating in like 720p and then upgrading to 4k, like some of the past video models do. It's actually generating natively in 4K. Wow. I know that cause I've talked to the people over at LTX API First Design runs efficiently on consumer GPUs. Companies do use the words consumer GPUs very freely. Freely, liberally. Because a lot of times they'll say it works on consumer GPUs. Okay, well which one? Well, if you have an Nvidia 5090, it'll work. Okay, so if I spend three grand on a GPU, I can make it work. Is that really consumer?

But it's not consumer friendly.

I guess technically it's a consumer gpu. It's just the highest end consumer GPU you can buy.

Yeah. But the results are.

Right.

Like I'm looking at the screen right now. It's really, really good. Like can you imagine what people could create with this?

Yeah.

I saw some other demos side by side and it seems to be really good at physics. Like even better than like Veo was at some physics, like doing things like flips and things like that. It's really interesting to me because we're moving at this pace now where we get these state of the art models that just blow people's minds. VO 3.1, Sora two things like that. And then like two weeks later an open source equivalent that's pretty much just as good rolls out. So it's like even those big state of the art models aren't saf. There's always an open source version following closely behind.

Yeah, I've been seeing what people are coming up with. I know you're not on TikTok, I am chronically online. But like I've been seeing what these people have been creating with all of these and it's just wild. Like someone created like a 1970s fantasy and it's anime and as if like it was drawn by an artist in the 70s and it's anime. So. So it's wild. It's wild what we can create with these. It's so, so cool.

Yeah, it's getting really crazy what you could do with video. And the one thing that's a little scary about stuff like this and having them be open source open weight models is, you know, with companies like Google and OpenAI, right. They're gonna have some guardrails, right. They're going to make it so you can't generate certain things. You can't deep fake stuff. You can't create, you know, fake news that makes it look like a war. Just Broke out in a place that it didn't actually just break out in. Yeah, they're going to do what they can to safeguard against that stuff.

These open models, they don't have guardrails, do they?

I mean, even if they do have guardrails, because they're open, people can figure out how to rip the guardrails off pretty easily. You know, like I think, I think a lot of these companies do sort of bake guardrails in when they first train the models and do a little bit of the fine tuning. But as soon as you put open source models into the hands of the public, the public's going to figure out how to break them as quick as they possibly can.

Pretty smart, are they? Yeah, I mean it is dangerous. At the same time, it's not, you know, like you wake up to war that broke up in one country, but it didn't really break out in that country. It's just like been made by some 15 year old somewhere, you know, because they did, they found out how to use this.

Yeah.

So yeah, it is dangerous 100%. So these companies need to have a bit of responsibility when it comes to these kinds of stuff because you don't want it to end up in the wrong hands, honestly. So yeah, well.

And I mean once it's open source too, the genie's out of the bottle, the toothpaste is out of the tube or whatever, saying you want to work.

With it, we're going to use that. Yeah.

You can't put it back in. And like even if the company put this up on their GitHub and said go ahead and download it and work with it and then deleted their GitHub, well, somebody still got it somewhere, right?

It's still out there, the Internet, it is there. So yeah, yeah.

Super, super interesting to kind of pay attention with what's going on. I don't do a lot of stuff with open source, but it's always super impressive to me how quickly open source follows behind Frontier Lab state of the art models.

It is wild.

Yeah. So the next thing I want to talk about here is this deep SEQ ocr. We don't have to spend too much time on it because it's kind of getting a little more into the weeds, a little bit more on the technical side. But the company out of China, Deepseek, they basically figured out this new like training method where they can essentially ingest a lot more text into the training model by actually putting all of the text into an image and then having it train on the Images instead of the text itself, which is just like a really wild sort of like outside the box. Like how did they even come up with this concept for training? But supposedly it makes it a lot more efficient.

It is, yeah. They figured out like a way to make AI read text as images and would do it like up to 10 times more efficiently than the standard text tokens. So instead of breaking words down into thousands of little pieces like other models do, because that's what they do, a deep SEQ literally looks at the page and kind of scans it and compresses it, or in other words, squishes it so that visually it somehow keeps everything almost perfect accuracy. And they're calling it the paradigm inversion, which means that they flipped the script. So text used to be more efficient than the vision for models. Now it might be the other way around.

Yeah, I mean, it'll be interesting to see if some of the other companies pick up this idea and run with it. Like, will we see, you know, DeepMind, OpenAI, some of these other companies anthropic testing stuff like this, or are they going to stick with let's jam as much, you know, text content as possible in. Yeah, I have no idea whether they will experiment.

Can you imagine 10 million tokens worth of data into one model context window? And like, that's like feeling like having like an AI running your company, like the entire Dropbox of your company.

Yeah. I guess the other question I have around something like this is, does it work from the consumer direction? So let's say I go to OpenAI and OpenAI has a, like one of the chat models, and let's say it's a model that has 125,000 token context limit. Right. So I can, I can only put in, let's say up to 90,000 words of text into this model before I run out of context window for it to actually understand what I'm putting in. What if I took a hundred thousand words of something, put them all into images instead, and then fed that to OpenAI or anthropic or Google or one of these tools. Does this same method work that other direction? Can I actually shove more context into a prompt by giving it images with text instead? That's something that I think could be fascinating to figure out because then what we might actually see is we're still typing or copying and pasting text into a prompt box. But then behind the scenes, these model creators might convert that to images for us in order for us to squeeze out more context out of the models. Does that make Sense. I don't know if I'm getting too in the weeds or not.

No, no, no, it does, it does. Can you imagine like now that they're all images? It could be the other way around.

Yeah, yeah. So maybe, maybe it'll allow us to squeeze more context out from the user side and actually be able to push more text into the model. But yeah, we'll have to see how that plays out. This is very, very new sort of technology and research, so we'll have to see. I don't know, I'm curious to see if any of the other frontier labs look at this technology and decide to play with it or if they're going to keep going with the methods they've been going with.

I think it's like a big hint that the future of language models might not be language based at all anymore.

Yeah, a couple last quick, sort of like rapid fire topics here. Over the last week, YouTube added in AI likeness detection. So yes, on YouTube we've always been able to like claim copyright on videos if somebody just like flat out takes your video and posts it on their channel or you know, takes music that you created and posts it on, on their channel, things like that. But now they're making it where if somebody actually uses an AI generation of your voice in their video to make it sound like you, or if they use AI video, like one of these avatar tools like hey gen or something like that to generate a video of you saying something you didn't say, you can actually now strike those down as well. And I was telling you before we hit record, I actually saw this feature, double checked my account, saw a bunch of people actually using my likeness in their videos and copyright claimed them all.

I mean, I think it would put people's like minds at ease when it comes to. A lot of creators feel better when it comes to these kinds of stuff. It's just they're trying to say we got your face covered, you know, So I think it makes people feel. Honestly, it's like an apology.

Yeah, yeah.

You know, like for everything I've done. So yeah, it's awesome. I like it. I like that they're going into this route because now you can create stuff but you don't have to use someone else's work in order for you to do that.

Yeah, yeah, I like that a lot. It is kind of a funny dichotomy just in my scenario, right, where I make videos about, here's how you use all these cool AI tools and how you can do things like clone avatars and stuff. And then people are using my teachings against me and then I've got to go and copyright strike them.

Exactly.

Very interesting dichotomy to be in.

Yeah. I did warn, like, YouTube creators that it might even flag their own videos by mistake. So can you imagine that happening? Like, you know, but this is my work. What am I? You know, these kinds of stuff, people are going to probably lose their minds for sure.

There's a couple other news stories that you mentioned that I'm actually not super familiar with, so I'll let you sort of take lead on these two. But you mentioned something about like an MIT fashion thing and then Yelp's new AI.

Yeah, yeah. MIT builds like an AI tool that is called Refashion, which leads you to remix your clothes like an actual DJ set, which is so funny. It's like you can rebuild everything by like Legos, so you can turn your trousers into a mini dress or like your hoodie into like a maternity wear or like whatever. Basically it's like a depop meets like cat software. So for the fashion girlies, it's just not just for the fashion girlies, but like anyone that likes fashion because who has the time for like over consumption and when you can literally just control plus z your whole outfit? So this is wild. I like mit. It's not Zara yet, by the way, but it's on its way to be that. So it's also pushing for sustainability and like making sure all the clothing end up in like landfill. So I'm very happy about this. Yeah.

Do you know what it's called?

Refashion. That's what it called.

Refashion. Okay.

By mit, MIT and Adobe, I think.

Gotcha. Okay. New software designs eco friendly clothing.

That's so what? Yeah.

All right. So to reduce waste, the Refashion program helps users create outlines for adaptable clothing, such as pants that can be reconfigured into a dress. Each component of these pieces can be replaced, rearranged, or restyled.

It's awesome.

Cool. This is actually one that I haven't come across yet.

It's wild. It's so good. It changes the face of fashion in my opinion. It's like a recycling kind of thing.

Is it demo able right now somewhere?

No, I don't think so. It's just like the MIT article.

Yeah, it's probably just like internal at MIT for right now and then once they feel like it's ready for primetime, put out some demo that we can use. The last thing that we were going to Chat about. I actually came across the news, but I only read the headline. I'm that guy right now. I only read the headline, didn't read the article. But there's now some sort of new, like, Yelp AI.

Yeah. So Yelp just launched like an AI receptionist and like an AI host at the same time, because apparently no one's actually answering phones anymore. So for restaurants, there's like a Yelp host, which is an AI that answers, calls and takes reservation and manages tables and gives away time and sends you the menu and probably knows if your date's allergic to gluten. Probably it even texts you follow ups, like, your table's ready, bestie, that kind of stuff. But there's also something called Yelp Receptionist, which is basically the same exact thing, but for businesses, I guess. So it books appointments and collects customer info and screens everything that leads to 24 7. So all of this for $99 a month, which is cheaper than a real receptionist. And it doesn't really need lunch breaks and, you know, these kinds of stuff.

So the $99 a month, that's paid for by, like, the restaurant, right? Yeah, or like whatever company is using it. I'm assuming Yelp is still free to use for.

I don't think people use Yelp in the uk, but I think in the US it's insane. It's like, very helpful for a lot of people. So also, the wild part about this is that Google's working on AI that calls businesses for you. So we're getting like, they're just close to AI just calling each other now, which is gonna be very funny for me to see. So.

Yeah, yeah. It's like you Google a restaurant and you tell AI to call the restaurant to find out if there's reservations, and then the restaurant answers, but it's an AI bot answering. So it's. It's like my Google's AI reaching out to this restaurant's AI.

They're gonna probably, like, tell each other that they're doing amazing and they're supercharging each other. This is going to be a conversation. It's gonna be insane. I would love to be a fly on that wall when that happens, like, just to see how that works.

I mean, we're getting there already kind of with things like email and whatnot anyway. Right. It's like, I don't know if you've seen those. Those memes where it's like, I wrote this one sentence and had AI turn it into a giant paragraph for me, and then they send the email. And the person that received the email is like this person sent me this giant paragraph of email. I had AI summarize it down into one sentence for me. Right? So it's like the original person wrote one sentence. The person that received it summarized it back down to one sentence.

Can you imagine it would be summarized into one word eventually?

Yeah. Like, so it's hilarious because like we're gonna see more and more of that kind of stuff where it's just like, like the AIs communicating with each other on our behalf and then eventually the AI is just like a middleman anyway.

Yeah, yeah, it's gonna be like the meme of like the spider man. I don't know if you've seen that meme of like three spider mans just pointing at each other. That's what's gonna be like.

Yeah, yeah, yeah, exactly. That's gonna be pretty hilarious. I know. That's already happening in like schools, right? Like students are using ChatGPT to help them write their homework. And then teachers are actually. There's a whole self work episode about this. The teachers are actually going and like using ChatGPT to grade the homework. So like the student is no longer even seeing the homework because ChatGPT is doing it and the teacher's no longer even seeing like what the students are doing because ChatGPT is doing it.

It is funny. The years are wild times. So I just hope that schools find like more ways into making sure that the youth is learning something.

Yeah, well, I mean my hope, my hope is that tools like ChatGPT and NotebookLM and all of these various AI tools that are out there make students more resourceful. They feel like they have more access to more information, more access to more knowledge, and they use that to actually help them, you know? And the sort of opposite end of the spectrum, the side that I hope it doesn't go down, is the path of everything just becomes AI generated brain rot.

Yeah, yeah, yeah, yeah. I mean, I need something to make sure that the youth kind of like has us ways of getting more creative rather than just recycled content. Because to be fair, like we're not going to beat around the bushes. It is recycled content. So like we need like more things to be pushed so that the students are learning something. Yeah, yeah.

This is getting probably off the rails a little too much. But like, I feel like that's where things like world models kind of solve a lot of that kind of stuff. Because if AI sort of understands the world around it as opposed to just what was injected into the initial training data that gets us to a point where AI is starting to come up with novel solutions because it has an understanding of the world, not just an understanding of what it was trained on. Sort of. That future excites me as well. But how far off we are to these world models actually being helpful, I don't know. We might still be a ways off.

Yeah, I mean, besides, I just need AI to cure big problems like cure cancer and stuff. Also automate my house. That would be also very helpful. But yes, the same level. Automate my house and like cure cancer.

Yeah, yeah, yeah, yeah, yeah. Cure cancer, solve climate change and do my dishes and laundry for me, please.

Exactly. Exactly. Well, this has been awesome. I mean, obviously a lot of has happened over the last week since we chatted last time.

Yeah.

I would love to hear from the audience what people think about these episodes, these sort of news recap and riff episodes. Again, I don't really do them as much on my main YouTube channel, so if you like more of this dial and you want to hear more of our musings and thoughts on what's going on in the world of AI news, let us know in the comments. We'll definitely keep doing more of them because they're such a blast for us to do.

I'm having so much fun.

Yeah, amazing. Well, so make sure you check out the Mindstream newsletter. That's the newsletter that Maria is writing and the head person over there keeping you looped in with all the latest AI news. So check that out if you haven't already and make sure you subscribe. So whether you're listening on Spotify, YouTube, Apple, we're available everywhere you listen to podcasts, so make sure you're subscribed if you're not already. And thank you so much for tuning in.

Yeah, thanks guys.