AI NEWS: The Next Wave of AI Tools Is Here

This past month in AI has been wild. OpenAI dropped a full on browser. Google upgraded VEO again. Microsoft and Anthropic both launched major new tools. And for some reason, even toilets are getting AI features now. There's been so many big announcements lately that most people are struggling to keep up, but I got you. I went through everything from the past month and pulled out just the updates that actually matter. So whether you're curious about where this tech is headed or you build with it every single day, I'll get you completely brought up to speed and I don't want to waste your time. So let's jump right in, starting with what people are calling the AI browser war. In October alone, we got a whole bunch of announcements about AI native browsers. The biggest announcement was probably ChatGPT's Atlas, which they announced back on October 21st. I did a full deep dive breakdown of this browser and tested all of the features. You can find that breakdown in this video called ChatGPT launched a new browser you'll actually use. But here's the quick TLDR of this browser. When you open it up, it defaults straight to Chat chatgpt. If you type anything in the URL box at the top, it actually sends that prompt to Chat GPT. So instead of doing a Google search for you, it will basically plug in whatever you typed and send it over to ChatGPT. You can choose between your models just like you would inside of Chat GPT. They gave us an Ask Chat GPT sidebar which opens up Chat GPT and has the context of whatever windows we have open. They gave Pro users this agent mode where you can actually let ChatGPT take actions in your browser on your behalf. So as an example, I could give it a prompt like filter this page by Translation tools and sort them oldest to newest. Now again, this is just an example. It's probably a lot faster for you just click on the things yourself. But if I go ahead and do that, you could actually see that it takes over my browser. That little preparing to filter page, that little arrow that's there, it's doing it on its own. You can see it click the translation filter and then it's sorted by oldest to newest for me without me ever touching anything. So it can actually start to take actions on your behalf. They also have a feature where if you're writing something like an email, you can select the text in your email and it will give you a little OpenAI logo next to your highlighted text. If I click on that, I can actually Tell it to improve this email so that it actually gets me a job. Here's my original email. Let's go ahead and use the updated one that it wrote for me. Probably much better. But that wasn't the only browser news we got this month. Back at the beginning of the month we actually got an announcement out of Perplexity's Comet. We can see back on October 2nd they released the Comet browser to the world for free. So originally the Comet browser was only for paid users of Perplexity and I think it was only for max users that paid the higher end 200 plus dollar plan. But now they rolled it out so anybody can use this browser if you've never used Comet before. Just like ChatGPT's Atlas, this one is built on Chromium, so it's got a lot of overlap with Chrome and all of your Chrome extensions will work. It also has an Assist assistant feature where if I click this assistant button it opens up a right hand sidebar. It could pull in the website that you're on for context with the messaging. What is this website for? And we get a quick description of this website. Great for summarizing articles and things like that as well. It has a cool feature called Shortcuts where over in my assistant if I type slash I can have pre written sort of workflows and prompts for me that I use often. It's real easy to create a shortcut. Just click the button, create a shortcut and give it the additional instructions you want that shortcut to prompt so you give it the trigger word. Let's say the trigger word is slash mat prompt and the instructions are tell me a joke. We can also tell it what mode we want it to operate in Search, Research or Labs. Pick our model or let it just select the best model for us and we can select sources. It can search the web, academic, social, finances, and various connectors. If you make these connections now, anytime I go here and type slash mat prompt and just hit enter, it's going to tell me a joke. You can also use this assistant to search past history. I was looking at a website that shared a bunch of AI tools. What was it? And look at this. You were recently looking at future tools and there's this cool quality of life feature called Split View where if you've got a few tabs open and you want to see them side by side, you can actually right click on one of the tabs. Click on new Split View with Current tab and we can see that it actually split the two websites. I have open into side by side browser windows, which is pretty helpful. And if I close my assistant, I even get a better view of the two windows. And in response to ChatGPT's atlas and perplexity Comet, well, Microsoft decided, hey, we also have these AI features over in Microsoft Edge. So they now have Copilot mode in Edge similar to the other two. You can open Copilot in a right sidebar here and it'll use the context of whatever's in your main window. It's also capable of searching through your browser history. Show me that blue hoodie I was looking at last week and it pulls up a bunch of history of what you were looking at. So even though we have Atlas and Comet and even the browser companies, DIA browser, which are all AI native browsers, all of the normal browsers everybody's used to using is also getting a lot of those AI features baked into them. We got it in Microsoft Edge. Google Chrome is starting to roll out some of these AI features as well. Brave is starting to roll out AI features. So pretty soon all of your browsers are going to be AI native. And one thing that I find exciting about this, and I did talk about it in my video dedicated to Atlas, is that we're gonna get to a. We don't even have to open up our browser anymore. Maybe you just use your phone audio prompt something you want done and these agent using browsers will go and just do the thing on your behalf. You don't need to be sitting there staring at your screen anymore. It makes the entire Internet promptable. Hey, go and look up when the dates of this event are. Book me a hotel, book me an airline, and I want to keep my budget under two grand. Go do it for me. And this is me holding my phone, by the way. You give it a prompt into your phone and it goes off and just uses your browser to go and do all those things on your behalf. That's where I think all of this is headed. That's where they want to get it to where you don't even need to be sitting in front of your computer anymore to go and achieve things you want achieved. We also had a bunch of news come out this month in the realm of AI video. So we got a new version of Veo called Veo 3.1. This is another one that I did a deep dive breakdown video on. If you check out my YouTube videos and you find a video called Testing Every feature in VO 3.1 with unhinged results, the title might be slightly different. We're currently actively split testing titles. But again, here's the TLD VR of what they released. They released ingredients to video where you can give it three different images. In this example, they gave it an image of a person, a room and some clothes. And it generated a video of that person wearing those clothes in that room because it had all of those ingredients. Here's one where they gave it two images. This person riding a motorcycle through some dunes and this like sparkly crystal Y looking thing and that was blended together to get this motorcycle guy riding through the glittery stuff. They also introduced the first and last frame feature before you can give it a starting frame and it would make a video based on that frame. But now you can give it a starting frame and an ending frame and it will figure out how to animate between the two. Definitely watch my video on VO 3.1 if you haven't already. Because we got some hilarious results testing this feature.

I really got the thumbs up right. We can see in this example they gave it an image of a barn here and then they gave it an image of this guy on horseback. And when the two are blended together, it makes a video of going through the barn to see the guy riding horseback. They also rolled out a feature where you can add objects to a scene and they're rolling out soon a feature where you can remove objects. But the remove hasn't been rolled out yet. The ad has. So this example, there's like a white snowy scene and if we play through it, you can see they added somebody carrying a torch through the snow and then removed it. Here's one where they remove a giant Death Star looking thing from the scene. But again, the add feature is available now. The remove feature says coming soon. A pretty significant upgrade to veo. But again, we tested it all out in this video right here. So check that out if you haven't already. Sora 2 was actually, if you can believe it, released last month back in September. But in October we've had a few new updates to Sora 2. They added the ability to do storyboards so you can generate multiple scenes that are back to back to back to create a longer video. And they made it so all users can generate videos up to 15 seconds. Now if you're a pro user, you can actually generate videos up to 20, 25 seconds in the web version if you're inside of Sora, you can actually click on the little storyboard button if you're on one of the accounts that has it. Clicking on storyboard opens up a new window where you can prompt multiple scenes into your video. OpenAI also rolled out a feature where you can add character cameos. Are you Tony's stank? In the past you could do cameos where it was you or your friends that gave approval to use their likeness in their videos. Now you can actually upload cartoon characters or stuffed animals, animals or even pets, and will actually create characters that you can use those cameos for. And in the last bit of Sora news that came out this month in October, they're launching the ability to buy extra generations. So if you're on a Sora plan and you find yourself reaching the limit of what you can actually generate, you can actually pay to get more generations. They said we thought 30 free generations a day would be more than enough, but clearly we were wrong. This will let our pro creators get as much usage as they want to pay for. There's also some talk here about monetization and a way for new generations of Sora creators to make money. Not a lot fleshed out there yet, but apparently you'll be able to make money on Sora in the near future. If you want to hear my more in depth breakdown and thoughts on Sora, definitely check out this video. Sora 2 is changing everything, but not how you think. Again, the title might be slightly different, but look for the video with Casey Neistat on it. And since we're talking about AI video, I have to mention this new LTX2 model. This is actually an open source model that's pretty much generating videos at the same level as what we're seeing out of like VO3, but higher quality and longer videos. So LTX2 synchronized audio and video generation, native 4K fidelity, 50 frames per second and it can run on consumer GPUs. Let's take a quick peek at their demo here and get an idea of what this model is capable of. Welcome to the world of audio.

So look, most wild thing to me about this whole thing is that it's open source and we get models like VO3 which are insanely good state of the art models, and Sora 2 insanely good. And just a few weeks after those models are made available, we get an open source model that's basically just as good. Like open source is only following behind by like a month which is just wild. Now Lightrics, the company behind LTX2 did set up a playground where you can go and test this model. If you want you can find the playground at app ltx studio/ltx2 playground. So that's the URL right there, it's not quite perfect yet. I gave it the prompt a monkey on roller skate. I got what looks to be a toddler on roller skates.

It did do a pretty decent wolf howling at the moon. But I'm going to spare you the audio because it didn't do a very good job on the audio for that prompt. And since we're talking about AI video, I want to mention some of the AI video features that Leonardo has rolled out in October. Now quick disclaimer. I do hold a very, very tiny amount of in Leonardo. Well actually now canva, but I was an advisor for Leonardo, so just keep that in mind. But here's all the video updates that they rolled out. So October 3rd they added start and end frames using cling 2.1. October 16th they added VO 3.1 as an option to generate videos inside of Leonardo. October 21st they added Sora 2 and Sora 2 Pro as options to generate AI videos. October 28th they added Haylou 2.3 and Halo 2.3 fast as options for generating videos. And on October 30th they rolled out cling 2.5 turbo standard. So tons of options for generating video. If you're ever inside of Leonardo and you click on video under AI Tools and you click on model, you have a ton of options for video models. Now pretty much any video model you could want to use you could use over in Leonardo now. And since we're talking about AI video, I felt this deserved a quick mention. YouTube rolled out a likeness detection tech technology for YouTubers. Now in my back end of my YouTube account, I can see when people are using my likeness, not just taking my actual videos but using like an AI generated version of my face or voice and I can actually copyright claim those videos. Now for you using my likeness, a feature that I think is going to be much more and more needed as time goes on and more of these AI video models roll out. All right, so we talked about AI browsers, we talked about AI video news. Now let's talk about AI image news because there was a little bit of that in October as well. Here's you. Oh geez. And you got the stink lines and everything. Starting with the new image model that was rolled out by Microsoft called Mai Image 1, which is a pretty solid image generation model. It could get texted images, it does realistic looking people, realistic looking animals, and looks pretty on par with what you'd expect from most state of the art image models. Now if you do want to use it, there is kind of a workaround in order to be able to use it right Right now I think it's only available inside of LM arena. So if you want to use this model, you'll head over to lmarena AI, come up to the top and click on New Chat in the top left menu. Up at the top where it says Battle. We're going to switch it from Battle to Direct Chat because we want it to specifically use this model. Down in the bottom, make sure you press the button that says Generate Images. And then up at the top it'll present you with a new dropdown with all of the available image models and you'll see Mai image 1 as one of the models that you can select from here's some of my obligatory tests. I did a wolf howling. The moon looked pretty good. I wanted to test if it would generate real people, so I did Sam Altman shaking hands with Elon Musk. It got close to Elon Musk, but this does not look like Sam Altman at all. I then wanted to test if I can get a lot of elements into a image. So I did a three headed dragon wearing cowboy boots watching TV while eating nachos. If you look at the image, it nailed it. It got literally every single thing from this prompt into the image that I wanted to test. Well, can I get away with trademark ip? So I did Mickey Mouse High five Super Mario. No problems with that. And then I did SpongeBob SquareP, Batman and Spider man taking a family portrait together. So I wanted to get three different studios because SpongeBob I think is Nickelodeon, Batman is DC and Spider man is Marvel. So a bunch of different studios all shoved into one image. No problem generating any of those. Just a little bit weird that we can only use it in LM arena right now. Also interesting because this model was put out by Microsoft. Microsoft is a 49% owner of the nonprofit that owns OpenAI. OpenAI has their own image model and they also have Dall E3. And now Microsoft is creating a new image model that competes with OpenAI's image models. So maybe Bing in the future is going to be using this image model instead. Or maybe it'll be a selector who knows what they're doing. But in other AI image generation news, we got some updates from Nano Banana and Nano Banana went viral last month when everybody was able to do Photoshop like edits with any image they want by just giving it text prompts. And well, Google decided people like this, let's roll it into literally everything. Nanobanana was added into NotebookLM. So when you generate video overviews, it's actually generating images. With Notebook lm, if you're on search and you open a photo with lens, you can instantly transform your images. They're adding it into Google Photos where you can go and edit any of your images in Google Photos with Nano Banana. So pretty much like every Google app that has anything to do with images whatsoever is getting Nano Banana added in, it seems. And as I just got done talking about how Nano Banana is able to do a lot of stuff that people would do with Photoshop and seems to be kind of disrupting Photoshop. Well, in the final week of October, Adobe had their annual event called Adobe Max, where they show off all of their announcements and pretty much all of their announcements were AI related. They've added new AI features to Photoshop Lightroom, Premiere Illustrator. They're adding a new AI assistant into Photoshop which which enables creative professionals to scale their work and free up time by instructing the assistant to help on a series of repetitive tasks and Surface personalized recommendations, all while maintaining creative control. So it sounds to me like Adobe is adding Nano Banana like features directly into Photoshop where you just kind of tell it through typing what you want the image to do and it will change based on what you type. That's what it sounds like they're rolling out here. They've also seemingly stopped trying to to force you to use Adobe's models. When you're in one of the Adobe products. We can see that generative fill inside of Photoshop, which is one of the best Photoshop features they've ever rolled out, in my opinion, now lets you use other models. We've been able to use Nano Banana for a little bit now. You can now use Black Forest Labs Flux 1 context and of course you can use the Firefly image models. They've also added Generative Upscale, and not only using their own Upscaler, but you can actually use Topaz Labs Upscaler, which is probably the superior Upscaler, as well as a bunch of other features that they've rolled out in Adobe products. I personally haven't watched the entire Adobe Max keynote yet. I plan to get myself caught up on it, but it sounds like they're just rolling out AI features into all of the Adobe products. It more seems like most of the features they announced at Adobe Max last year are finally starting to roll out into Adobe products. If you're a coder or a vibe coder, someone that doesn't know how to code but tries anyway because of AI, you've got some new tools at your disposal this month as well. Did you see see the memo about this? Yeah, yeah, yeah, I have the memo right here. Earlier this month, Claude code went to the web. In the past, if you wanted to use Claude code, you had to do it inside of a terminal. Well, now if you just go to Claude AI over on the left, there is now an option for code. If you click this, it opens up Claude code and it connects directly into your GitHub. So you can pull in any GitHub repositories that you've built and continue to build off of them directly inside of your browser without needing the Claude code terminal. So this makes it infinitely easier to use Claude code. If you're not comfortable with the terminal, you just do it all right here on Claude's website. Now, Google also wanted to help out the Vibe coder, and they added Vibe coding features into Google's AI studio. So if I head on over to aistudio google.com right at the very top there's a new option that says Vibe code Gen AI enabled apps in build. If you're not seeing this link, there's also Build over on the left sidebar, but let's go ahead and click into here. And not only is it a sort of Vibe coding platform where you can code right from here, it also gives you tons of ideas for prompts to help you on your Vibe coding journey. You can also select from other Gemini models, but you're probably going to want to use Pro if you're writing some code. You can also give it instructions on whether you want to use Angular or React. And you can turn on the speech to text here and also just talk out exactly what you want it to code for you. October did see some new large language models being made available for us to use. And since we were just talking about stuff for coders, I figured I'd mention this one first. But the company Cognition, which created Devin and then acquired Windsurf, just released their own agent model called SWE 1.5. And if we look at this benchmark here for agentic coding tasks, it actually comes in just below Sonnet4.5, but also insanely fast compared to the rest. 950 tokens per second compared to Sonnet 4569 tokens per second. And when it comes to the score, it just barely underperformed this one, but way, way, way faster. So if you're looking to experiment with another agentic coding model, maybe SUI 1.5 is one to check out. We can see here that they used Cerebras for their inference provider, which explains why they're getting such insane speeds here. But in other large language model news this month we also got haiku 4.5, which is a new model out of Anthropic. It's a small model designed for when you need really really fast responses, and we can see that despite it being a smaller, faster model, it actually outperforms Sonnet 4. It just barely underperforms Sonnet 4.5. So if you want a model that's like a good halfway point between super super fast and pretty dang Smart, Cloud Haiku 4.5 is a pretty solid model and I believe it's fairly inexpensive to use as well. It's about a third of the price as Sonnet 4.5 both on input and output. We can see output here is $5 per million tokens, where on sonnet4.5 it's $15 per million tokens. Obviously this will likely have a much higher context window as it's giving us prices for larger context as well. We also got another model out of OpenAI. This one is another open model called GPT OSS Safeguard, and these models are basically models that are fine tuned to follow your safeguards. The GPT OSS Safeguard models use reasoning to directly interpret a developer provided policy at inference time. It says the policy is provided during inference rather than being trained into model. So it's easy for developers to iteratively revise policies to increase performance. So you could keep on tweaking the safeguards that you've built into it until it gets you the results you're looking for. Not a model I've played with yet, not really a model that I have a ton of interest in, but if this is something that you need for your business or your use case, it's available.

And I have a few more quick like rapid fire pieces of news that I'm not going to go super deep on because, well, I've already talked about some of them or it's not really big enough news to go into. But Also this month OpenAI introduced Agent Kit and this was a pretty big deal. However, I also did a deep dive on this one called Did Chat GPT just kill Zapier and you can get my in depth breakdown on on exactly what Agent Kit is and why it's significant. But the TLDR here is you can build these workflows that look similar to what you might get from Zapier, N8N MindStudio Tools like that and you could do it directly inside of ChatGPT. Now, they're not nearly as in depth as what you can do with those other tools that I just mentioned. It's still very, very basic at the moment, but it's built into ChatGPT and kind of shows us that ChatGPT is coming to disrupt some of those workflow builder tools as well. If you do want to try Agent Kit yourself, you actually need to do it in the OpenAI playground and not chat GPT. So in order to find that, head on over to platform.OpenAI.com log in and then click on Dashboard at the top. And then over on the left side you'll see Agent Builder. This is the new Agent kit that they were just showing us. And you can create these logic workflows that start with a trigger and if something happens, then it goes to one agent. If something else happens, it goes off to a different agent and you can build out these in depth workflows again. You can see there's not a ton of options yet or a lot of integration, but this shows us what is coming. On the same day they showed us the Agent kit, they also announced apps inside of Chat GPT. And I talked about this a little bit in that same video that I just mentioned, but basically they're trying to build like an app store where people can develop stuff straight inside of OpenAI's chat GPT. Kind of reminds me of what they were trying to do with plugins which never played out, but this looks to be a little bit more deeper integrated than what the plugins were. Some examples of apps that have already been built inside of ChatGPT. Booking.com, you can have it book hotels for you directly inside of ChatGPT. Canva, you can have it generate images for you by giving it prompts inside of ChatGPT. Coursera, Expedia, Figma, Spotify have it generate music playlists for you and Zillow. Those are ones that are currently built in, but they're opening it up so anybody can build apps that run inside of ChatGPT and that'll most likely be pretty powerful, especially if people figure out ways to monetize it. People will build whole businesses on top of Chat GPT similar to the way people built whole businesses on the back of the Apple App Store when the iPhone came out. So pretty, pretty exciting development. And I can't wait to see what sort of apps other people build outside of the like partners that they launched this with. Jumping back to anthropic for a second this month, they Introduced agent skills where you can teach it how you essentially want it to handle certain tasks for you, and then you can turn those skills on or off. So we can see here how skills work. While working on tasks, Claude scans available skills to find relevant matches. When one matches, it loads only the minimum information and files needed, keeping Claude fast while accessing specialized expertise. They're composable, so they can stack together portable, so they use the same format everywhere. You built it once and it'll work across cloud apps, cloud code and the API. They're efficient, they only load what's needed, they're powerful, they can include executable code. It says, think of skills as custom onboarding materials that let you package expertise, making Claude a specialist on what matters most to you. So you're basically teaching it very specific information so that when you need to call on that information in the future, it has it readily available and can easily access those skills that you created. And once those skills are created again, you can use them in Claude, you can use them in the API, you can use them in Claud code, you could use them across the entire sort of anthropic ecosystem. In the final bit of news that I wanted to share this week, Kohler wants to use AI to look at your poop.

Kohler's new toilet camera provides health insights based on your bathroom breaks. Because everybody wants a camera in their toilet. That's something I know for a fact. Dakota is a health tracker that analyzes what's in your toilet to assess your hydration and gut health. Now, it's not actually a toilet, it's this attachment that attaches to your toilet so it doesn't look like you need to buy a full new toilet. You just need this little gadget here. It's designed to peer into your toilet bowl using sensors and analyze what it sees, using algorithms to provide insights into your hydration and gut health. It will discreetly notify you when blood is detected, which can indicate more serious medical issues. It's available for Pre order for 5.99, and it looks like it's already started shipping by the time you're seeing this video. So maybe it's not cameras, maybe sensors looking at your poop isn't so bad. But that's what I got for you today. October has been a crazy month and I'm hoping to do these monthly breakdowns. I've slowed down on the weekly news breakdowns and opting to focus more on individual pieces of news and make videos that kind of go deeper on the stuff that I think's important as opposed to making a weekly video where I just share all of the AI news. Even if there wasn't anything that interesting this week, to me, it makes more sense. Let's focus on individual videos that dive deep on the real important news and then maybe once a month I'll give you this breakdown of here's what sort of shook out to be the most important news of October on my Future Tools website where I break down all of the latest news here. The month of October I had 220ish news articles and I narrowed this video down to just about 20ish, so less than 10% of the news that I came across this month. So I want to make these videos that share more of the signal and less of the noise and then they might be a little less frequent, but it's just the stuff that I think was really important for the month and that's what you got in this video. Hopefully you feel more looped in, more tapped in. Thank you so much for nerding out with me. If you like stuff like this, give this video a thumbs up. Maybe consider subscribing to this channel. It will ensure more stuff like this shows up in your YouTube feed. Once again, I really, really appreciate you and hopefully I'll see you in the next one. Bye bye. Thank you so much for nerding out with me today. If you like videos like this, make sure to give it a thumbs up. Thumbs up and subscribe to this channel. I'll make sure more videos like this show up in your YouTube feed. And if you haven't already, check out FutureTools IE where I share all the coolest AI tools and all the latest AI news. And there's an awesome free newsletter. Thanks again. Really appreciate you. See you in the next one.