AI News: New ChatGPT And Gemini Images; The Incredible Meta Muse; AI Wearables and More

Vídeo em Inglês, publicado pelo canal Paul J Lipsky.

Video transcript

Transcript from available captions; it may contain errors.

Guys, I feel like I'm losing my mind. We've seen such leaps in what AI can do over the past couple of weeks. It's both exciting and scary. After getting several big model releases last week, this week we got a new image generation model from OpenAI and two of the best image generation editing tools I've ever used. We also got a surprisingly incredible AI agent from Meta, as well as the best AI wearable I've ever seen from Apple. We'll talk about all of that and a whole lot more in this week's weekly AI news recap. My name is Paul Jay Lipsky and every week I bring you the AI news that actually matters. So, make sure to subscribe to stay on top of what's new. And now, let's get to the news. We're going to start out surprisingly talking about one of Meta's new AI tools. Now, Meta does make good AI models and tools, but I rarely cover them because compared to what I see coming out of other companies, Meta's products haven't really excited me until now. This is the new Meta Muse. It's a personal AI agent with its own cloud computer and browser. So, this is my Muse on my iPhone. This is currently available for Android and iPhone. There's also a web app, but I found the best experience is here on my phone. And if you look at this, in many ways, this feels similar to Grockbot because you get this little character up here. You can give it a name and an image. But unlike Grockbot, there's only one agent. If you open up the side menu here, you can't create multiple agents. There's just one. Now, you can have side chats with that agent, but primarily you're just going to be interacting with it inside this one main chat. and then using it feels just like you're messaging someone on WhatsApp or iMessage. So, the whole experience is incredibly simple. And I think it's that simplicity that is going to help this do very well. So, let me just show you some demos of how I've been using this. So, I just set this up yesterday and I've been getting it to help me search for a new laptop because the one I'm using just isn't keeping up with my workloads. So, I've been sending it questions back and forth, having it do research for me. You can see right here, it has a built-in browser. So, it's using its browser to do the research. And all this is happening in the cloud. So, I don't need to have my app open or something open on my computer for this to run. It's just going to run automatically for you, even 24/7. Like down here at the bottom, this is a message right here that it sent me proactively. I didn't prompt it here. Just randomly, it sent me this message telling me it's going to share a few tips with me this week. But let me show you some of the cool things I like about this. So, I'm going to continue that conversation I was having about buying a new computer. I'll say to it, actually, I'll use the voice dictation for this because that's the primary way I've been interacting with it. What if instead of buying a MacBook, I buy a Mac Studio instead. And you see as it's thinking, it has those three dots just like a human when they're typing. So, it's very, very fast. And here's a good answer that it gave me. yada yada yada. But the trade-off is portability. So really like solid answer. Now another cool thing you can do with this is of course just like any other agent you can connect it up with thirdparty apps. So coming into the menu on the left we'll come into the settings and you see the connectors here. There aren't a lot but there are the basic ones. So we have calendar reminders healthly Gmail all the Google Google stuff and a few others. So they definitely have to add more but this is a start. And then by default, all these ones are connected. A browser, Facebook, Instagram, and threads. It comes with a lot of built-in protections to kind of protect the user from the agent taking action without you wanting it to. So, for instance, let's say I'm going to tell it, hey, that laptop we were looking at from B&H, I want to purchase that one. Let's go through with that purchase. Here, it gave me a a really great response saying, hey, if you order it from there, you're not going to be able to return it once you open it. So, just a heads up about that. That's proactive on its own. I think that's really cool. It did that. But I'll say, "Yeah, I want to purchase it anyway." And notice what it did right there. It gave me a thumbs up. And that's something else I've noticed is is really kind of fun about this is that randomly it will give you a thumbs up. There are a few of them here. It doesn't do it for all of them, but it does it every once in a while, which again kind of makes you feel like you are interacting with a real person. And you can also thumbs up or react to its comments in any way that you want. It doesn't actually do anything from what I've seen, but I found myself doing it anyway, just out of habit. So, now it's opening up the browser to make the purchase from B&H, but it's not going to make the purchase on its own. First thing I would have to do is link up uh use link, which is how you can make the purchases, and then it will ask for my approval before doing that. So, any critical step is going to ask for approval, like purchases and sending out emails as well. but it can make purchases for you if you connect up link. I also had it put together a presentation for me. Now, that has gotten lost in the chat. So, down here at the bottom, if you click the last tab right here, this will show you all of your artifacts and everything that is created for you. So, I can open up this this PowerPoint presentation. I asked it to put together a PowerPoint presentation convincing me to buy the MacBook I was looking at. It was just a silly demo to see how well it could handle presentations. It did a pretty good job. Like, it pulled real images. It's nothing crazy good, but it gets the job done. But let me show you the other tabs down here. So, the first one is the chat. This is the main chat where you interact with the agent. The next one right here is a feed. So, as your agent learns more about you, you'll notice that things start to appear in here that it thinks will be relevant or interesting for you. So, yesterday I was talking about it to it about MacBooks. So now it says in this feed, this is like a news feed saying, "Hey, you know, Apple expects there to be new MacBooks coming out next month and Apple revealed a new foldable iPhone." The third one are ideas. If you're not sure what to use your Muse for, these are some ideas and some of these are generic and some of these will become personalized for you based on the conversations you've had. The next one are goals. You can create goals in here or do it in the chat and it will just come up with some on its own like this one right here telling me to upgrade my MacBook. Overall, I just find this to be very pleasant to use. The whole experience feels very natural and the language it uses when it responds feels very accessible. It doesn't use a lot of those AI terms that feel unnatural. The responses are a good length that's very readable without giving you too much information. I think they did something really nice here that is going to make it very easy for regular people to use this. For instance, if the average person out in the real world told me they wanted to start using an AI agent, but they had no idea where to start, I might actually recommend this just because of how easy it is to use and it's simple, but it works and it works really well. There are some advanced settings. So, if you click on your avatar on the top, you can see some of those in here. So, this is showing you everything the agent has done. Right here are all the pending approvals. And then right here is a schedule, I guess. And then this last one is where you can really dive in and change its soul or its memory. Its soul is basically its personality. So, you can come in here and make edits here. And the memory is everything it remembers about you. But those are the advanced settings. And I don't think most people ever need to come in here because they're just going to interact with the agent directly inside this main chat. This is currently rolling out in the US, but they do expect it to come to other regions soon. And before we move on, let's take a quick moment to hear from the sponsor of today's video, Abacus AI Agent. Abacus AI Agent is a general purpose agent that can work across websites and other tools to complete complex multi-step jobs. For example, I asked it to build a live dashboard that tracks the price of five competing products over time. I told it to use computer use to visit each pricing page, collect every plan and price, save everything in a database, graph how the prices change, and highlight any increases or newly added pricing plans. Then I told it to publish a dashboard, refresh it every Monday, and email me whenever something changes. From that one prompt, Abacus AAI handles the research, builds the app and database, and sets up the recurring automation. So, this isn't a one-time report. It's a working dashboard that keeps collecting new data, and becomes more useful over time. It can also build other full stack apps, test and fix bugs, create presentations, and connect with tools like Gmail, Jira, and Slack. Abacus AI agent is part of Chat LLM by Abacus AI. The subscription costs $10 a month, includes access to chatbt, Claude, Gemini, and Grock, plus image and video generation tools. Check out Abacus AI agent using the link in the description. And now back to the news. Next, I want to talk about two image generation tools that came out this week that are blowing my mind. The first one is called Google Pix, and the second are a bunch of upgrades to the images inside of Chat GBT. Both of these are a huge step up for what we can now do with image generation and editing. So, I want to look at these side by side. I want to start though with Google Pix. They actually announced this back in May at Google IO. So, they've obviously been working on this for months now, and it was rolling out starting last week, but I finally got access to it on Tuesday, which was the same day that the new chat GBT images 2.5 came out. So, everyone was focused on this, and no one noticed what Google was doing, but I have to show you this because it is so cool. So, this is the new Google Pix. And just like you would expect with any image generation tool, we have a prompt box here. I'm going to give it this really long dense prompt telling it to create an ad for a fake product. And it's going to use Nano Banana Pro or Nano Banana 2 for this, which is nothing new. And it's not the image generation that I want to show you. What hap it's what happens after the image is generated. All right. So, here's the generation. And you'll notice over here on the left, we have four different options. So, you can choose exactly which one you like. I think I like probably this last one the most. All right. So, I will select this one. I will confirm it by checking that one off right there. Now, this is where it changes everything because this is no longer a static image. If we hover over this, you'll see you can select any element here and it's really good at identifying what those elements are. So, here we have the man wearing earbuds. Here we have the earbuds charging case. Here we have the wireless earbuds. And you can select any of these to tell Pix to make changes to that specific element. So let's say right here, this is 4.9 out of five. I could select this and just say let's change this to five out of five. Or maybe we'll do 4.8 so it feels a little bit more accurate. So we're editing the text there. Save. And then I'll apply that one edit. All right. You see that has now been updated. Let's try something more complex though. So up here we have this image of this guy. So I'll select this. Select edit image and I'll say change the man to a woman. But I don't just want to make one edit at a time. I will select multiple elements. So I'll select the charging case and say make it a white charging case. Add that one. And then I'll say I want these earbuds to be white. Make the earbuds white. Add. And then make this one white too. Add. And then we'll apply all four of those edits. And here's the updated image. And you'll notice that even though it made changes to the elements we asked it to, it did not make changes to the elements we told it to leave alone that we did not select. And that's the big unlock here. Because in the past, this was so difficult to do with image generation tools. You would tell it to make one change to one part of the image and it would change another part of the image or distort it. But Google Pix fixes that. There are some other things you can do with it and you can also edit real images. But I want to hop over to Chat GBT first and show you this in comparison to what ChatGBT rolled out. So they rolled out a new image generation model and new tools that you can use inside of Chat GBT that are very similar to what we see with Google Pix. So, I'm going to send it the same prompt and we'll send that one along and see what it creates in comparison to what Google Pix created. And here you go. And you see this is a very similar look and that's because I gave both of them a very long detailed prompt. Now, unlike Pix, it's not isolating elements, but you can still make edits in here. And this is another big change for Chat GBT. It's so much better at making those edits the same way that Pix is. Instead though, it uses comments for that. So up here at the top, you can click on comment and select any element that you want changed. So we'll try that again. The same stuff. So I'll select this and say make these white and check that one off. And you'll see up here it saves that one, but it doesn't send it yet, just like Google picks. So here, I'll select this guy and say make him a woman. And what we change this one to change to 4.8 stars. And then the earbuds make these white. And I think that's everything. So, we'll send that along. And then here is the updated image. And you see it mostly got it right. It updated the man to a woman, changed the 4.9 to 4.8, and changed the earbuds to white, but it didn't change the case, even though I did comment on the case itself. And that's something I have found where this comment tool isn't super accurate and can miss what I'm talking about. Which is one reason I prefer the way that Google Pix does it where it actually isolates the object and you'll know for certain that you are telling Google Pix in this case exactly which element you want changed. But overall what changes it did catch it did very well with. And notice just like with Google Pix, it only changed those things and nothing else. Everything else stayed exactly the same. Next, let's check out how well each of these can edit real images. So, what's cool about Google Pix is you can just throw an image in here by selecting right here where it says upload. So, I have this image of me right here with Josh Woodward, who is a VP at Google. And I want just want to make some changes to it. So, I will select that one and open it up. And you see it brings it in without making any edits or changes. And now it's detecting the objects. so we can select them to make the edits. There's no real way to do that inside of chat GPT. You have to give it the image and then tell it what changes you want to make right away in order for it to actually bring something in like this. And then you can click on it and leave the comments. But it did change my shirt color. It did make things a little bit blurry, but that's only noticeable if you zoom in. And then you can make some changes. So I might say, let's get rid of Sorry, man. We're going to get rid of you. say remove this guy. And maybe let's get rid of this this projector up here. Remove this projector. Am I cheating by telling it? Maybe I'll just say remove this. And let's see how it handles that. And then I'll come back over to Pix because this is going to take a little bit of time. Chat GBT, the new image generation model is faster, but it's still not as fast as what we're getting from Google. So now I'll do the same thing. Say remove and remove. So you see Google is already done. It did a great job. Chat GBT is still working over there. And you can mess around with the thinking level, but I've always consistently found it to be slower than what Google's doing. So coming in here, if we zoom in, can we zoom in a little bit? I think this looks really nice. I don't really see a lot of distortion there. This is gone. It looks like it was never there. And the rest of the image is untouched. So it's really good at editing images. So, let's see how chat GBT handled it. I think it did a really good job. You can't tell that there was something here, that there was a person here that was removed. And interestingly, chat GBT left the chairs. And I'm just realizing that now. If you go back to what Google Pix did over here in Google Pix, it got rid of the table and chairs. So I think actually chat GBT did a better job here because it left the elements that were real that were really there and chat and uh Google Pix removed too much. I was playing around with this a little bit more and I ran into another weird issue with Google Pix where I had selected this person in the background and said turn this person into an alien and it did but it left the selector box around it. You can see that right here. So that's a little bit of a bug. But if we come back to chat GBT and look at the edit it made, you can see obviously it's not going to have that box around it because it doesn't use that selector method. And the guy was successfully turned into an alien. But notice what else is happening. If we look at the first edit and let's look at what my face looks like. It's like slightly blurry, but it still looks pretty much like me like the original image. Coming into the second one, there's a little bit more distortion. I look different there. And then this most recent edit, I look a lot different there. There's a lot of distortion happening and this is something specifically that Open AI said they fixed with this new image generation model that obviously is an issue. You still you also see that with Google Pix like coming in here if we scroll into my eyes they look like I have like crazy eyes in there and there's definitely some distortion. Um I would say I guess both of them have it. So neither one of them is perfect. Obviously, they both still have that issue, but in terms of the editing itself, it is a step up. Just has a little bit more work to go. Personally, I think I like what Google Pix is doing a little bit more where you can select the exact element more than using the comment tool inside of Chat GBT, but OpenAI also released a lot of other cool things you can do with images inside of Chat GBT this past week. I made an entire video on that, so I'm not going to cover it in this video, but it's linked up up here, and I'll remind you and have it linked up at the end of this video as well, so you can check that out after the rest of the news. Let's talk about some updates from Apple because they may have just released the best AI wearable of all time. I've had my eye on the AI wearable space for a while, and I'm not quite sure where it's headed. Is it going to be glasses? Is it going to be some sort of wristband? I'm not sure, but this may be it. So, this is the announcement. Apple released or announced two new Apple Watches. The first is the Apple Watch Ultra 4, and the second is the Series 12. Now, there's a bunch of updates. I'm not going to cover all of them, but I want to highlight one particular feature. Both these watches will have ambient listening. That means if you want, they'll be listening all the time. Now, they're not recording. I guess they are recording, but they're not storing the recordings. The recordings are just buffered. And then they may take action based on what they're hearing. And all of this is being done on your watch and on your phone. So you need to have your phone with you. It's all being done locally. Nothing is being sent to the cloud. But here's what it can do. So first we have sound recognition. So if you, let's say, for instance, a doorbell rings, you can have it notify you that the doorbell is ringing. This could be really helpful for someone who's hard of hearing. But these are the two features that I'm most interested in. This first one is called live rewind. So, as it's listening, if you doubleclick on the crown, it's one of the buttons on the side. You'll get a little recap of what was just said. It's the past or the last 15 seconds. So, let's say you're in a loud restaurant and a waiter just told you all the specials and you kind of missed it. Double tap. You can see what all the specials are. Let's say someone just told you their name and you instantly forgot it. Double tap and you can see what their name is. I think that's pretty interesting. The other thing that they have is called Siri recap. So what happens with this is throughout the day if this passive background listening picks up that you are having an important conversation of some kind maybe a meeting it will create a summary of that conversation of that chat and then you can come back to it at the end of the day to refresh your memory and see all the key points. Now again it's not recording according to them. So, I think this is how they get past the uh the requirement for consent when you're recording because it's not actually recording. You cannot access the recordings. No one can because they don't exist. It's just transcribing what is being said. And if the AI detects that it's something important, like an important conversation, it will create one of these summaries. They've also said that if it seems like it's something very personal, it's not going to make those summaries for you. We've seen other products like this before. This is a device called OMI. I've had one of these. I've tested it out and I didn't particularly love it. The idea is you wear it around your neck and it's listening all the time. Sending that over to your phone which then transcribes it. Same thing. And you can get reports at the end of the day. My experience with this was that it had a really hard time knowing uh the context of what was happening. So it would maybe hear the radio music or someone talking and it would think that was me and everything just kind of got convoluted. So Apple hopefully they found a better way to do this. And if they do, well, this may be the best wearable of all time because it's not something new. It's already on the wrists of millions of people. They don't have to try to bring a new product that people are unfamiliar with like a pin or something. This is super familiar and they're just adding AI functions into it. And it seems like this could actually be a useful feature if it works well. But it's not out yet. It's rolling out in the fall. So, we'll have to see what it's like when it actually comes out. We're not done with the news yet. There are a few other updates. We'll be able to move through these ones pretty quickly. And we'll start with what's up with Chat GBT. So, there's been some improvements to Chat GBT voice. So, now with voice, you can use GBT 5.6 Soul or GBT6 Astra. And that way if voice needs to, it will use those models when it needs to search or reason. This should make voice a lot more capable. In addition to that, they've increased the limits for chat GBT voice. So on the go plan, you now can use it for up to 3 hours per day. That's 3 hours with GBT Live 1 mini. On the plus plan, it's 3 hours with GPT Live 1. For the pro plan, you get up to 15 hours with Live 1. And on the $200 pro plan, you get unlimited usage of GPT Live 1. So you should be able to use it a lot more than you could in the past. In addition to that, ChatGpt has released something called writing styles. So when you set this up, it will look at your messages, your connected documents through something like Google Drive or Notion, and your email to set up a writing style based on your tone of voice. Now, I opted not to do this because a lot of what I have inside my documents and emails was written by AI. So, this wouldn't be accurate and I would prefer to be able to feed it documents and say these are ones that I've actually written, but it doesn't give you that option. So, I just use skills for that instead. So, I'm not using this writing style, but it is there if you want to use it. And then finally, a little bit of a public service announcement. Apparently, there is such demand for the new GPT model Astra that it's possible that the team may need to pause new signups for pro users uh until they can match that demand better. So, if you've been on the fence about signing up for a pro subscription with Chat GBT, you may want to do it now before they lock that down or pause it for a little bit. Or this is just a publicity stunt. I'm not sure. Next, there were a couple of updates to Grockbot. So now, if you have your bot, let's say, browsing the web for you, like it wants to purchase plane tickets for you or log into your in in this case, your united.com account, instead of having to open up the browser, which can be kind of clunky to do that, you can just sign in directly in the chat instead. You can see that looks like right here where it pops up and says you put in your username and password. And this is something the AI won't see. It won't see your passwords. And you can even use your password manager to fill that in as well. So that's nice. And there's now a bot marketplace where you can browse bots that other people have shared and add any of them to your existing Grockbot account. So just search for Grockbot marketplace and you should be able to find that. And then finally, we have some updates about Google Gemini. So first, there's finally a Windows app for Gemini. There's been one out for a while for Mac. It's pretty nice. So, if you use Gemini a lot, you may want to download the Windows version and Google released LIIA 3.5, which is their music generation tool. So, if you come into Gemini and click on the plus icon here, then select down here where it says create music, and you can choose from one of these templates or just give it information about what you want generated, and it can generate music for you. A lot of people don't know that, but they do have music generation, and they just updated the model, so it's even better now. So, I don't know about you, but I felt like there were a lot of cool things that came out this past week to play around with. It's been a lot of fun messing with the new image tools. The new Muse is really nice, and I'm getting pretty excited about some of that Apple stuff. But what do you guys think, though? What are you most excited for from this past week? Let me know in the comments section down below. Then if you want that full video I mentioned before where I go over all the updates that OpenAI released in terms of their image tools and image model, check out this video right here. One of the updates that I'm most excited for is you can now sketch things right inside the app and it turns your really bad sketches into images and it's surprisingly really good. So check out all the updates right here and I'll see you over there in one second. Bye for now.

Assistir ao vídeo original no YouTube

Assuntos: Agentes de IA, Muse. Publicado no YouTube em 2026-09-11.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.