AI News: New FREE AI Challenges ChatGPT!

As with most weeks lately, it has been a really busy week in the world of AI and I don't want to waste your time. So let's go ahead and break it all down right now.

Starting with the fact that OpenAI is finally bringing their Advanced Voice mode to the web. Now we've already had Advanced Voice mode available on the mobile app for a little bit now. They recently rolled it out on the desktop app. Well now it is going to be available inside of the browser app. This is based on an announcement from Kevin Wheel here on X. He he says rolling out to ChatGPT paid users this week Advanced Voice mode on the web down here. At the bottom of his post he says give it a try@chatgpt.com if you're a + Enterprise teams or edu subscriber. We'll look to roll it out to free users in the coming weeks. Let me know what you think. That seems to imply that it's already in there, but when I head over to my own ChatGPT account, I'm not seeing any options here to use the new Advanced voice mode yet. So it seems to be a rollout. Maybe some people have it already and it is still rolling out, but I don't have it in my browser account yet. Show off but something that did definitely roll out this week was a new update to GPT4O. According to OpenAI, the model's creative writing ability has leveled up. It's more natural, engaging and it's tailored writing to improve relevance and readability. It also apparently works better with uploaded files, providing deeper insights and more thorough responses. Again, if you head over to Chat GPT, if you're using the GPT4O model, you'll be using the latest version of the 4.0 model. I've played around with it a little bit. I find the updates to very subtle, but I also use Claude a little bit more than I use Chat GPT, so I'm not sure I would totally know the difference at the moment since we're talking about Chat GPT. Remember back when they showed off the Advanced voice feature in that demo video? Well, they showed it off with the ability to actually get live video and have Chat GPT actually see what you're looking at to be able to inform its responses. Well, that feature never really rolled out with the Advanced voice mode like we thought it would. Well, it looks like it's going to be coming very soon. I found this out from Rowan Chung here who spotted it from Android Authority, but it looks like the code for Those features was actually discovered in the latest beta version of GPT. They discovered in the code live camera functionality, real time processing, voice mode integration, and visual recognition capabilities. I'll make sure this tweet thread is linked up in the description below, but you can actually see screenshots of the code here that make it look like it's going to be coming pretty soon. Moving on to anthropic news, they just rolled out a new feature. If you use Claude, you can now direct the add documents from Google Drive. If I head on over to my cloud account here, you can see there's a new little Google Drive button here. If I click this button here, it shows me my recent files inside of Google Drive and I can also do a search for specific files in Google Drive and whatever file you pull in will work the same way as if you were to just drag and drop a file into Claude. Google's Gemini got an update this week. It can now actually remember things. So ChatGPT has had a memory feature for a little while now. Now Google's Gemini has it as well. When you're logged into Gemin, you can actually come down to your little settings cog wheel and you'll see a button that says Saved Info. If you click in there, you can actually see all of the info that's saved. You can add info or you can ask Gemini to remember something while you're actually chatting with it. At the moment it says this feature is only available in English, but here are some examples that Google gave of the types of things you might want to save. Use simple language and avoid jargon. I'm a vegetarian, so don't suggest recipes with meat. After responding, include a Spanish translation. When trip planning, include the cost per day. I can only Write code in JavaScript. I prefer short, concise response. You can also give it info about yourself, like your name and your career and where you live and things like that, so it can be better informed when it's actually giving you its responses. And since we're talking about Google, I want to point out something really cool that YouTube is doing. I've mentioned it in the past, but it looks like it's actually finally starting to roll out. They're now going to start doing automatic dubbing on videos into multiple languages and the YouTube creators don't have to do anything extra. So for someone like me, all I need to do is upload my video to YouTube like normal and will automatically create a translated version of it in Spanish, Portuguese, German, French, Italian, Hindi, Indonesian and Japanese. That's super, super exciting to me. And it should be for any content creator that's publishing on YouTube, because it means like, every single piece of content you make is going to be available to an audience that's like 10 times bigger than what you're already getting. That's going to be huge. There's a new model out from a company called DeepSeek, called DeepSeek R1 Lite Preview, which is a model that's going head to head with OpenAI's 01. Now, if you remember, the OpenAI 01 model is the model that sort of thinks things through before it gives you a response and is much better at sort of logic and questions that require some detailed thought behind them. And up until now, OpenAI's Zero1 model was really the only model out there that had that sort of focus. Well, now there's a model out of this Deep Seat company from China that can seemingly act in the same way. And we can see on these benchmarks here that in things like math and code, it actually outperforms the O1 preview model. Now, keep in mind, this 01 preview model isn't the full O1 model. It's like a checkpoint of the model. We actually haven't seen the full OpenAI01 model. In fact, rumor has it it's going to be coming out within the next few weeks. But again, that's just a rumor. But this deep seq model that is beating 01 preview in a few of these benchmarks here, it's not really a comparison head to head with full 01 model that we're going to be seeing soon. Now, the French AI company Mistral also has some updates to their chat bot called lechat. It now has search capabilities, vision capabilities, ideation and coding and you can use it all for free. So if I come up here to lechat, we'll go ahead and sign in. And very similarly to ChatGPT, we've got the option to use Canvas, we've got the option to search the web, and we've got the option to generate images. Lechette now has some new features, Web search with citations, canvas for ideation and inline editing, state of the art document and image understanding, image generation powered by Black Forest Labs Flux Pro. And we got a little breakdown here comparing Chat, GPT, Perplexity, Claude and lechat, showing that most of the features that are paid in these other platforms are actually free. Here in Lechette give me at least 10 YouTube channels to follow to stay up to date with AI. We've got the Pie Coach, two minute papers, some bum Conrad Cording deep learning AI Robert Miles AI explained Yannick, Kilcher, Vanillin Valkov and AI in games. Some of these are actually channels I hadn't heard of yet, so more for me to check out. If you're on a budget and you don't want to spend any money on a chat bot like ChatGPT or Claude, but you want a lot of the same features, give Mistral's the chat version a try because it does a lot of the same stuff that the paid versions of those do, but for free. And just for fun, let's see how it does with image generation. Generate an image of a cat eating a taco and that's what we get. It took about 18 seconds to generate and again it's using Flux1Pro, which if you've used the AI generator over in Grok, it's the same one that that one uses, but again it's free to use tools like ChatGPT and Claude have been absolutely amazing for businesses for brainstorming ideas for marketing and branding and business strategy. But it's not always super obvious what prompts are going to give you the absolute best results for the goals that you have in your business. And that's why for this video I partnered with HubSpot to share their AI prompt library. Now this prompt library was custom curated by the CMO and the SVP of HubSpot, Kip and Kieran, also the host of Marketing against the Grain. Now, as somebody who has a podcast on the HubSpot network, I know these guys personally and they know their stuff. They know how to craft amazing prompts around things like marketing and sales strategy, branding and advertising, business strategy and analysis, pitching and investing, website strategy, and so much more. Want to write headlines like David Ogilvy or sell your products using a proven storytelling framework? All of these prompts are available to you completely for free and they're designed so you don't need to be a prompt engineer to get amazing results from these AI tools. I've gone ahead and put the link to this 100% free resource down in the description below. You're going to want to check this out and thank you so much to HubSpot for sponsoring this video. There was a lot of news out of Microsoft this week because Microsoft held their annual Ignite event, an event where they typically make a ton of announcements. I actually don't know if this announcement came out of the Ignite event, but Microsoft signed a deal with HarperCollins, which is a book publishing company to essentially get permission to train AI models on the books from HarperCollins. Now, HarperCollins is going to get permission from each individual author, as it sounds like. But with all the lawsuits that have been happening over news sources and authors and people like that actually suing these AI companies, it's starting to sound like the big companies are starting to ask for permission a little bit more frequently. Now, if you remember, back at Microsoft Build, earlier this year, they announced a feature called Recall, which was like your web history, but for your entire computer. So you can go back and remember what you were doing in certain apps throughout the day. Or, you know, several days ago, maybe you were doing something in Photoshop and you wanted to go back to that exact moment. You can actually go back in history on your computer. Well, there was a lot of concerns when they announced that they went back to the drawing board and really, really dialed in, like, the security and just overall functionality of it to make people a little more comfortable with it. Well, as of today, November 22, copilot PCs with the Snapdragon chip in them are now going to have the ability to access this recall feature finally. Now, in order to access this, you need to have the Windows 11 Insider Preview build. And of course, you need to be using one of the newer Copilot PCs. Here's what the new feature looks like. You have this little scroll window across the top of your recall page here, and you can scroll back to any moment. You can see you've got today, October 21, October 20, October 19. So you can go back to any day in time and remember what you were doing back on that day and actually get screenshots of that moment that you're looking for. You can also type to search your history as well. There's also a little icon that will pop up down in your system tray here that will allow you to pause the recording. So if there's websites you're looking at that for whatever reason, you don't want Recall to remember that you were looking at that website. You could always pause it and make sure it's not remembering that specific moment that you're looking. And if you do happen to record moments on your computer that you wish weren't recorded, maybe you were doing some gift shopping or something like that, you could go back through and actually delete those moments from your recall history so that they're completely removed. There's also security measures in place. So to even open the recall window and see what's in there, it's gonna Double check and make sure it's you through whatever security measures you have set up on your PC. And if you really, really don't wanna use it, you can turn it off completely and not use recall at all on your computer. So if you're concerned about buying a Copilot Plus PC and you're considering not getting one because you don't like the Recall feature, it is possible to disable it completely. There's also a new feature called Click to Do where you can click on an image and it'll give you options like Visualize, search with Bing, blur the background of the photo, erase objects in the photo, remove the background with paint, and things like that. So when you click on an image you have all sorts of new AI related options. And again, these features roll out the day. This video is going live November 22nd. So if you already have a copilot PC with one of the Snapdragon 8 or better chips in it, you should be able to upgrade to the new Insider Preview build of Windows and using these features. Another thing Microsoft announced this week is that it's going to let you clone your voice on teams meetings. This is really cool for if you're speaking to somebody in a different language, you'll be able to speak in your native language and it will actually translate it and localize it to whoever you're talking to and it'll still sound like your voice.

We need to review the progress of the project. Can you please tell us how the tasks we should take are going?

So I'm super excited to see that come out inside of Microsoft Teams. I would love to see that feature roll out into things like Zoom and Google Meet as well soon. Eleven Labs rolled out a new feature this week to build conversational AI agents. So if I log into my 11Labs account, you can see there's a new button over here called conversational AI. Let's go ahead and try a call and see what it sounds like. I'll use the video game character here. Let's just see how this goes.

Greetings, Traveler. How can I assist you on your quest?

What is your name?

Greetings traveler. I am Callum, a wizard here to help you on your quest to collect the 11 gemstones scattered across Labs Onia. I can help you find them, but I can't do it all for you. What can I do to help you today?

What are these gems for?

Ah, the 11 gemstones. They are said to hold immense power, capable of granting wishes and even altering the very fabric of reality. Legend has it that they were Scattered.

That's fascinating. Let's go ahead and try this myself now. So I actually just told it to create that character for me and you can see it dumped it into this AI agents list over here and I can see what it looks like to actually build an agent now. So the agent language we want, English, we can have it set on the very first message when somebody starts talking to it. There's a system prompt with the details, wizard name, etc. Which large language model do you want to use? And it's got quite a bit of options. Gemini 1.5 Flash is the default because apparently it's the fastest model in here. But you've also got GPT4O, Claude 3.5 Sonnet, and even a custom LLM which looks like you can set up through using an API key. But I'm not going to go there for this one. Let's just leave it on the default. You can set the temperature and then it's got a knowledge base, so we can add data to this knowledge base. So if I wanted to make a AI news bot, I could take the transcriptions from all of my old videos, throw them in this knowledge base and people can ask questions about all of my old videos. You can add tools, apparently uses web hooks here you can add secrets. Okay, so this is probably for adding, you know, API keys and things like that. I'm guessing you can select the voice it used Callum, but I have my own Matt Wolf voice that I trained in 11 labs, so I can make it sound like me if I want. We've got the normal settings you get from 11 labs, all sorts of really cool features to go ahead and build out a bot that has the information and the voice of whatever you want it to be. So that's, that's kind of cool and fun. I can see the main use case probably being like customer support bots on websites and things like that, but this is something I definitely want to do a deeper dive in and try to build some characters with later. So maybe that'll be a video we circle around to in the future. Eleven Labs also rolled out the ability to auto regenerate. So if you're using their long form text editor to generate audio, it looks like if something doesn't quite sound right, you can highlight that selection and tell it to regenerate it and it will do it again. And ideally it gets this sort of tone and inflections right the second time around. Suno got a huge upgrade this week as well with their new V4 and I actually Got early access to this and was able to play with it a little bit before they released it. And honestly, it is pretty dang impressive. It actually generates music that I want to listen to. And quite a few of these songs I've actually gone back and listened to multiple times. One of the things I was having fun doing, and I hope I don't get in trouble for this, but I was taking lyrics from existing songs that I like and changing the style like. Like Damn it from Blink 182. This one, I told it to make it electronic poppy dance version, and it reminded me of Owl City, so that's why I named it that. But listen to this.

It's all right to tell me what you think about me I won't try to argue or hold it against you I know that you're leaving you must have your reasons the season is calling your pictures are falling, falling down.

I actually think it's good. I actually like listening to that version of this song. And then I did it like a poppy dub style. And I did it like a country version, like, it's all right.

Tell me what you think about me here's why I subscribe to Matt Wolf's song. It's a punk version.

Tech whispers through the digital trees AI news brings you to your knees finally.

This was actually a song that I did with Suno version 3. But if you do a song with Suno version 3 and you want to sort of bring that song up to date with the new version 4, you can go to your old songs. Click on these little three dots here. Click on Create over here, and there's actually a remaster and it'll convert the song into a V4 song and actually make it even more impressive sounding. So sudo v4 has been really, really fun to play with. We got a pretty big update in the world of AI Art as well. Black Forest Labs released Flux One Tools, which actually gives Flux One all sorts of new abilities. We're getting closer and closer to being able to do with Flux 1 what we have always been able to do with stable diffusion AI art. Things like in painting and out painting. We can see this picture here where he's starting with this gray suit. They mask out the suit, and then he turns it into like a jean jacket here or this sign that says burgers and beers. Mask out the burgers, change it to coffee and beers. And you can see it maintains that look, that style. And then here's an example of out painting where they gave it a picture of just an eyeball and then it out painted the rest of the face. They also have what they're calling structural conditioning with flux with canny and depth models. Now, if you're familiar with control nets with stable diffusion, this is essentially, essentially the same thing as the control nets. You've got a depth map so you can give it an image and it tries to estimate the depth in that image like you see here in this gray image. And then you give it a prompt. It will create new images trying to model this same sort of depth here. And then this is the canny model here where you upload an image. It sort of makes like a traced outline of the image, and then your new images follow that same sort of traced outline, but in a completely new style. They also added image variation and restyling with flux. So here's some examples here where you combine an input image with a prompt. Like, they uploaded this duck here and then said fine glass sculpture on a shelf. And it made this image of the same duck, but as a glass sculpture or a simple 3D model, or actual duck on a desk or epic anime art. And all of it modeled the look of this duck, but in the new style that was prompted. Here's a guy taking a selfie like out on a hike or something. Flat vector, simple material art, large hand puppet with brown fur in the forest, pen and ink sketch and detailed anime artwork, all from this original image. And then here's this one with bread and looks like milk and cheese. And they were able to make like different variations and angles of the same image. And all of this is available inside of the Black Forest Labs API. So tools that leverage Black Forest Labs will probably start rolling out a lot of these features soon. There's a handful of websites that you can use. Some of this, there's File AI replicate together AI, Freepik and Crea AI. If I jump over to File AI, the very top says introducing Flux tools. So if I click try it now, it looks like all of these tools are available here as well. Let's see, if I click on Flux 1 fill, I'll upload this image of my headshot here. And then it asks for a mask URL. Let's go ahead and create one. And now let's just go ahead and, you know, mask off my face, see what it does. Then use mask and you can see here's the mask that I just did. And let's run it and see what happens. I accidentally left the prompt up here. A knight in shining armor holding a great shield with fall on it. So it turned my face into a shield with fall on it. Let's try the canny model real quick. Once again, I'll upload my headshot here. I'll leave the prompt as a pink owl and let's just run it. It should basically make a pink owl that looks like the shape of my head and I guess that's its interpretation of a pink owl. But really what it did was it found the outline in the original image and then it used that for this. So if it doesn't see outline of an owl in my original image, it's going to struggle. So some fun stuff that you can now do with Flux, giving it a lot more controllability, the same kind of controllability we have with something like stable diffusion. Coca Cola is in the news related to AI this week in that they create an AI generated ad that resembles their old Christmas ads, but just generated with AI this time around. Here's a peek at what that AI generated ad looks like. Even the people in the ad were AI generated. This whole thing fully AI generated. I don't know the copyright of the music, so I'm getting rid of the audio here. But they put out that ad and well, a lot of the AI people thought it was really cool that Coca Cola is putting out ads with AI. And a lot of anti AI people were very pissed off that they used AI instead of hiring people to make an ad the normal way. They basically said that they wanted to create an ad that found the sort of nostalgia of the old Coca Cola Christmas ads, but they wanted to bring it into the modern era and they thought using AI was a really good way to do that. Since we're on the topic of AI video, the tool HeyGen now has an iOS app. HeyGen is a tool where you can sort of train an AI avatar of yourself and it can speak like you and look like you and lip sync like you and translate and still sound like you. And now a lot, lot of this stuff you can do with an iOS version of it. Now, every week before I make these videos, I go on X and I go on Threads and lately I've been going on blue sky and LinkedIn as well. And I've been sharing most of the news that I'm going to share on these videos and I'm asking people for help, like what news am I missing that's not on this list so that I can make sure I get full coverage of what people are talking about in the world of AI this week, sponsorship guru actually pointed out something that I hadn't seen yet. He, he said Pickle, the AI tool that allows you to join Zoom teams, meet with your avatar and speak from elsewhere. I actually hadn't seen this one yet, so I had to look up like, what is this Pickle thing he's talking about? And I found their Twitter account and it's actually pretty interesting. So check out their demo video here.

As you can see, I'm currently participating in a Zoom call, but what if I tell you that I'm actually not in front of the camera right now and this is an AI generated live screen that moves in sync with my voice. Pickle generates real time video of talking you based on your audio inputs, which can be streamed directly to Zoom.

So you can have an AI avatar looking like they're on a Zoom meeting the whole time, and you could be in a different room talking into a microphone. And in real time, it syncs up your voice to the avatar that's actually on the Zoom call. So with something like this, you don't have to worry about, you know, brushing your hair or putting on clothes before getting on a Zoom call. Perplexity has a new feature because Perplexity just keeps on shipping new features. This time around, you can actually purchase products straight from the Perplexity app. So you can use Perplexity to do research before buying a product, which is something that I actually use Perplexity for quite a bit. And when you find the perfect product that you want to purchase, you don't even need to go to the E commerce store to buy it anymore. You could just press a button and purchase it straight through Perplexity. So you can see, if I go to Perplexity here and say, compare the available versions of the Meta Quest, it gives me a response comparing these various versions, the 3, the 3s, the 2 and the Pro. But then if I scroll down a little bit, look, it's got all of them available, available for purchase right here in the same Perplexity app without me needing to leave. And I could simply click Buy with Pro. I would then add my shipping address, my billing address, and then check out. Once you've added these details once, you don't need to keep adding them. In the future, it becomes that much easier to make purchases straight from within Perplexity. If you're into AI coding, the tool from Vercel called V0, just rolled out some new features you can now create and run full stacked Next JS and React applications. Create multiple files in one generation link and deploy to Vercel Pro projects And use Vercel project environment variables. I'm not a coder myself, but I do play around with these AI coding tools a lot. I think cursor is still my main go to, but I haven't played around with these new features. I haven't really compared v0 to what I'm getting out of Cursor too much yet. If you have a rabbit R1, it seems like we finally got the feature that they were promising from the very beginning, which is the ability to teach it how to do things on your computer. So it now has a Teach Mode. So here's a demo video they shared in this blog post where you can see they give it a lesson name, a description. They're under this Teach tab here. And then from that point on it says recording. Every action you make on the website that Teach mode can record will be displayed in the log below. So then it looks like you just kind of go through and actually do the steps you want it to do on your computer. It watches those steps and logs them all. And then in the future you can actually run that set of actions straight from your rabbit. So you teach it once inside of the app on your computer, and then once it's learned how to do it that one time, you should be able to call upon that action from your rabbit whenever you want and then it will go through those steps again. I do have a rabbit R1. I literally haven't even taken it out of the box. I haven't tried it yet. But now that the feature that I was actually excited about for the rabbit R1 is available, this is something I'm going to be exploring for sure. The startup h just raised $220 million to launch Runner H, which is an agentic AI that they're trying to build. It says Runner H will initially focus on three specific use cases. Robotic process automation, quality ass insurance, and business process outsourcing. I hadn't really heard of this company H until I started prepping for this video, but this will be a company that I'm watching closely and see how this all plays out. And ideally, once there's something for us to play with and actually show off, I'll show it off in one of my future videos. If you use the Brave browser when you use their search, their search also now gives you recommended questions to continue on the search. So there's a little screenshot here if we take a closer look. Somebody used Brave Search to type in in Christopher Nolan films. It brought up a bunch of films. It then looks like it gave a suggested question of which actors appear most in his movies and then it gave an answer to that. This is like their search engine, like Google, but it's got this follow up questions area in it. Unlike Google. This week, while Elon Musk was playing Diablo 4 and apparently becoming the best Diablo 4 player in the world, somebody asked him, do you think AGI will come next year? Here's his full response. Thanks to James Bloom for pointing this out.

It'll be next year. Probably it's not next year, say 2026 at the latest for AGI.

Latest.

I hope it's nice to us when defines AGI is smarter than any human. I think it's less than 24 months.

Away from that Google DeepMind shared new research this week called Alpha Q Bit and this is AI to assist quantum computing. It says our new AI system accurately identifies errors inside of quantum computers, helping to make this new technology more reliable. Now, quantum computing is something that I've been really, really fascinated with. I've been studying it pretty much just as long been studying AI. I just don't feel smart enough to talk about it too intelligently yet. But basically, quantum computing can take certain problems which take a conventional computer billions of years to solve, and solve them in just hours. Which is why in my opinion, quantum computing is actually way scarier than AI. The implications to cryptography and Internet security and stuff, those implications to me are even scarier than the implications of AI. But that's a video for a different day. But this new paper lays out a way to make quantum computing computing even more effective by spotting errors and making the quantum computers more reliable. It says here, accurately identifying errors is a critical step toward making quantum computers capable of performing long computations at scale, opening the doors to scientific breakthroughs and many new areas of discovery. I'm not going to get too into the weeds on this article, but again, I will make sure it is shared in the description. So if you want to dive deeper, this article is available for you to read. And that's pretty much all I've got for you this week. I wanted to share that Leonardo AI, a company that, yes, I am an advisor for, is having a Black Friday sale next week. So if you want to try out Leonardo Pop back over on Black Friday, where they'll be having a special deal going on. I'm actually thinking about doing another video next week where I round up a bunch of cool Black Friday deals that I've came across. Haven't decided if I'm going to do that or not, but I tend to mostly sit out on Black Friday and not buy much, but I know a lot of people wait around for Black Friday to grab tools that they've always wanted to grab. Let me know if you think that would be helpful in the comments and maybe I'll make that video as well. But that's what I got for you this week. If you like videos like this and you want to stay looped in on all of the latest AI news, make sure you subscribe to this channel. I share AI news videos, AI tool tutorials, I break down cool AI research, and if you're following my Shorts feed, I share a lot of other emerging tech, AI and otherwise things like cool 3D printers and drones and flying cars and just stuff that I find fascinating. That type of stuff. I've been posting a lot on short form video. So again, if you like that kind of stuff, give this video a thumbs up and consider subscribing to this channel. It'll make sure more of that kind of stuff shows up in your feed. And finally, if you haven't Already, check out FutureTools IE where I curate all the cool AI tools that I come across. I keep the news page up to date on a daily basis, so all the news I don't get around to sharing these videos, they're on this news page. And of course I've got a free AI newsletter where I'll share just the coolest AI tools I come across and the most important AI news. And if you sign up, as a special bonus, I'll give you free access to the AI Income Tax Database, a cool spreadsheet database of various ways to use these AI tools to earn some side hustle income. That's totally free. You can find it all over at FutureTools IO. Thank you once again for tuning in. I really, really appreciate you. I do record these videos on Thursday, so if more news came out on Friday that I missed, that'll make it in next week's video. And thank you to HubSpot for sponsoring this video. Really, really appreciate you guys. Thanks again for nerding out with me today. I'll see you in the next one. Bye Bye.