Claude Blackmailing Explained, AI to Build Shops in One Click & More AI Use Cases

This week in generative AI releases, we have Enthropy coming out with their own voice agent along with some new alternatives that are more emotive than ever before. A big controversy around Opus 4 actually autonomously betraying the user and sending a warning email to government agencies and some new state-of-the-art image and video generation models. And we'll be looking at all the releases that you can put to work today in this week's episode of AI news you can use. Okay, let's start by

talking about AI voice assistants. So, first things first, there's a new open source model that sounds incredible. These voice models are so good at this point that nobody can really tell what's up. People claim this to be the 11 Labs killer, and I kind of see the point. The voice synthesis of this thing is faster than the inference of the AI models, meaning it's faster than apps like chatb can even produce the text. And you can clone voices with just 5 seconds of source material. Here, a quick comparison to the current market leader, 11 Labs. My name is Maximus Desimus Meridius, commander of the armies of the north. General, Honestly, there's so many good alternatives at this point that soon we'll just be getting voice AIs everywhere. Let's see what's next.

And this one is kind of fun. It's called Rhyme, and they too do AI voices. Now, there's a twist to this one because these are probably the most emotional AI voices I've heard yet. And I'll just put on my headphones and let's have a listen together to this live chat. I'll give it Mike. What's up, dude? It's nice to see you. I actually got a few minutes to chat. So, uh, how's it going? Hey, it's going good. What about you? Um, I'm really curious to see how you're going to work with the interruptions cuz that's probably the most annoying thing about the chat GPT voice mode, right? Oh, dude, for sure. I totally get that. Like, interruptions can be hella annoying, but honestly, yeah, what if I interrupt you? What? No, you can't just drop me like that, man. I'm just a skater dude trying to chat about music and skate tricks. All right, man. So, I don't know. Let me match your energy or something. I don't even really care. What you up to today? Honestly, just uh hanging out and trying to work on some new music with my band, you know. Yeah. Without I thought so. Just hanging out. Me, too. All right. Have a good day. Not bad. You can pick all sort of voices. You can see the latency is not optimal. The interruption actually worked super well there. But it's about the motion in the voices. You just drop in and just smack the lip. Drop down. Snap. And I quickly wanted to show this off because these are so damn expressive and emotional. Oh, hell yeah. Dude, is like classic. You know, the whole killing demons thing is pretty wild. They're kind of the opposite of professional voices to be honest, which can be really good in certain cases. And if you need that, then this is a great service to consider because they just don't sound like your typical AI assistant and I thought that was interesting. With that

being said, Claude actually added a voice AI this week. Now, I was really looking forward to testing it out and comparing it to ChatGpt, etc. But they're slowly rolling it out and it's not on my phone yet. It's only in the mobile apps. And with that, Entropic added one of the two big features that they were lacking behind on with the competition, voice and image generation. But we gathered some examples from across the internet and you can listen to them right here. Did I get an email from Brighton Co.? I'll check your Gmail. I see that you did receive an email from Brighton Co. last night. In the email, they mentioned they're looking forward to today's presentation at 11:00 a. m. And this is super interesting. As you can see from this example, this voice assistant can just like Claude connect to your Google account and take advantage of all these integrations like Gmail, Google Calendar, or Google Drive. Now, Clot could already do this, but it's actually really interesting to see a direct integration with the voice mode. This is something that, for example, in CHBD is not possible. So, you could hook it up to your calendar and then talk to the voice assistant about your upcoming appointments. Same thing goes for your email. Now, honestly, an integration in one of these consumer apps that I already use is something that's really welcome and I can't wait to try this myself. Just update your app and check if you have this for yourself. This is supposed to roll out to all paid plans first and then even free users will get 20 to 30 voice messages. And as we're

talking about anthropic, we have to discuss this story that happened over the course of the last week. Even though it's not really news you can use, but it is news of AI using tools against to harm its user. Sort of. It's a story that you might have heard of at this point where Anthropic's new AI model, Opus 4, blackmailed researchers because they were talking about performing illegal activities with some of the work. They were using Opus for concretely faking some clinical trials which would make the billions but kill many people. And as you might have heard from the story, the clot that was active within agentic system and access to multiple tools actually decided to write an email to a government agency with the details of what was going on there by itself. Now, here's an important caveat that I don't see talked about enough. First of all, if you give an AI model access to various tools like sending email, browsing the internet, and having the ability to make these decisions on its own without user intervention, it will do all kinds of things. Sending an email like this is just one of the many behaviors that can happen. And that's why none of these apps by default do that. It could equally as well decide to delete every file on your computer. But again, there's no consumer application where this is possible by default. You really have to be very dedicated to set it up in a way where it can even do things like this. And secondly, once you give this level of autonomy to a model even today, yeah, well, all sorts of things can happen. And this is an absolutely quite horrible example of something happening cuz the whole conversation there might have just been a thought experiment or like in this case and this is the thing that most people missed. This test where it sent that email was fictional and they were trying to push it towards that. So not just that they gave them all the tools, they were actively pushing the model to betray them. And yeah, this is definitely a new kind of story and honestly I was equally as terrified as many people are when they first heard about this. But I think some of these footnotes do matter because if you look at it from the angle of okay, they purposefully set up the AI to do this. They gave it all the tools which no normal consumer would know how to even do and then they were pushing for results like this. Well, for that lens

we can pull up another news story from this week which was essentially a developer using 03 to find a zeroday vulnerability inside of a Linux implementation. In other words, preventing a programmer's nightmare and a potential disaster just with 03. No additional tools, nothing fancy, just raw all free through the API protecting users from something that the developer might have missed. So I guess my point is yeah, these tools are becoming increasingly capable and a sharp knife can be used for both good and bad. But nevertheless, it's an interesting story and I wanted to dedicate a little bit of time to discussing it here. It's not something you could directly use, but it's something to be aware of while you're using these models. The next

story is something that's actually really interesting for people who work at bigger companies or enterprises. It's the ability to use the world's best OCR, which is short for optical character recognition, aka showing an image of a document and the computer intaking everything on it, including graphs. And this came out a while ago. We covered it on the show. It's a service offered by the company Mistral out of France. And on all benchmarks, they're just the best. And we actually live tested this back in the episode and it worked so much better than just throwing it into chat GPT or using various competing products. And now they offer this at scale. So if you have a lot of documents that you need to digitize, no matter if they're handwritten, ancient, or barely readable like this one right here, now the service is available for the API, meaning you could throw thousands if not tens of thousands of documents at this and just get them digitized. This can be super essential in getting an organization ready for this next phase of technology or managing the context that you give to your AI assistants or agents is the name of the game, but it all starts with actually digitizing and organizing it first. And this tool can actually really help with that. And while on the topic of context management for companies, I just quickly want to point out that here at the advantage, I built a company from the ground up fully remotely, fully digitally. So our entire company operation system, every document, every SOP, every contract, everything is digital and everything lives inside of our notion where we have everything. So very often when I prompt and I need a certain piece of context like the tasks of every single person in the company, well, I have a repository with a document that includes every single task done by every team member. I can just export that and use it right away. If you don't have things like this set up and you're a business owner, I think it's time to start thinking about these things and getting everything ready for this next phase where these products are going to be very capable, but you will need to give them the correct context. I mean, we're sort of getting there already, but this is only going to become more relevant as these systems become smarter. And to round out the point, I just want to say that I did some consulting work recently for a company that was looking to transition their legacy systems to a more digital form. And something like this is a part of the pipeline paired with two in-house GPUs, offline LLM, and a simple software that strings it all together and allows all employees access to all the essential company documents without using online services. It's just a straightup weapon for something like a law firm. But it starts here, digitizing documents. So think about that. And I wanted to spend a bit more time on this because I do think that this is actually the best and most reliable solution to turn physical documents into digital ones. Okay, on to the next point. Okay

so the next one is about chat GBT3 playing Pokemon by itself. And let me explain why this actually matters. I don't know if you caught this, but during the Google IO presentation last week, they actually pointed out that the new Gemini 2. 5 Pro is actually the first model that managed to complete Pokémon. All of it. It got all the badges and got to the Hall of Fame and it took it around 800 something hours. Now, that might be a lot of time, but this is the very first AI model that actually managed to do this independently. Quite the achievement. And now somebody just took 03 through the API and hooked it up to Pokemon. And it's currently streaming live on Twitch. And you can watch it play. Look, it's still kind of early in the game. It's got Pokemon between level 10 and 18. But this thing has only been running for 32 hours. And it made really good progress. You can see the reasoning. You can see everything. And on top, you can see the progress, how far it is along the entire game. And now the big question is, will it manage to complete the game? And if yes, will it take less than the 800 hours? Now, obviously, this will take a few weeks to find out, just cuz that's how time works. Matter of fact, it should be 33 days before we find out, and I'll follow up on the story. I'm really curious to see. It's really too early to judge or to compare it. I just love seeing these interesting benchmarks, which are not some obscure set of questions somebody put together, but something tangible and relatable like the completion of Pokémon is on the case now, and it's looking good. We'll keep an eye on this. Okay

look, if you're watching this video, you know exactly how exhausting and overwhelming it can be to stay on top of AI and navigate all the different spaces across the internet that house the information. YouTube is fantastic, but can be a sensory overload. On X, you get good info quickly, but it's mixed with a bunch of topics that you probably don't even want to see. And I love Reddit, but often feels like the posts are just driven by somebody trying to prove themselves to the world, sharing their knowledge just so they can prove to themselves how smart they are. Not always the case, but if you've been around, you know there's some truth to those statements. Now, I myself was also looking for an alternative to this. And after some conversations, I kind of realized that what I really wanted is this old school forum vibe. Now, you probably got to be at least 25 or 26 to know what I'm talking about here. But back in the day on the internet around 10 to 15 years ago, there used to be these traditional forums with genuine discussions. And even old Reddit was a completely different culture than it is today. Genuine discussions could flourish and you actually knew the different users by name because there wasn't tens of thousands of them. And if you saw a specific profile picture, you knew that, oh wow, this person created a new post. I might really want to read that because what they do is high quality and at the advantage you might know that we started our very own community. Now this community is not for everyone and it is paid but I check it out every single day and I do get this feeling that I used to get in the golden age of the internet forums. The people in our community are genuine and helpful and nobody's posting just to make themselves feel better. Everybody there is on this common journey of trying to master these tools, trying to get the most out of them to improve their professional or personal life. And there's zero clickbait because there's no point in that. Generally speaking, you only join the community if you have an open mind, can afford it, and you're curious about the different possibilities that AI tools absolutely do hide. Not all the use cases are obvious. And if you pass those filters, you don't need to use inflammatory language to get people to click on a guide or a course. That's not how humans actually communicate. It's how humans have to communicate. If you're distributing one person to hundred thousands, but a few dozens or hundred, you don't need that. So, it just creates this unique environment that I myself and many other members in there cherish. And I kind of just wanted to take a second to communicate that in a bit more of a human way of me just kind of ranting about it a little bit. But it really is a space where you can ask genuine questions and get proper answers to them. Share your progress and actually feel heard while being on this journey of acquiring skills and developing your skills relating to generative AI along with others who are on the same journey. So, if you enjoy this channel and you're looking for a place to connect with others who are interested in generative AI and its possibilities just as much as you are, then this is it. That's why we created the community. All right, that's all I got to say here. Now, let's continue with the video. Okay, so we have two new

image editing models this week. One of them coming from Bite Dance, the company that owns Tik Tok. And the second one is Flux. 1 Context from Black Forest Labs. You might be familiar. That's probably the open-source image generation model. Well, now both these companies released image editing models. The one from Bite Dance looks decent. Nothing special to be honest, but the one from Black Forest Labs looks absolutely incredible. Actually receiving results that you would otherwise only expect from a skilled Photoshop user. I mean, even if you're skilled in Photoshop, good luck putting this girl into the same scene, including all of the snow. That's a tough workflow. And with these new editing models, you can do increasingly complex things in a heartbeat, which is just something that used to not be the case a few months ago. And while we're

talking about images, let's also talk about video because there's a brand new release in that space and we actually went ahead and tested this one thoroughly. It's been about a week since VO free released and shocked everybody with their audio visual generations. Please don't finish writing that prompt. I don't want to be in your AI movie where every clip has audio associated with it. And now we have Cling making their move. If you're not familiar, Cling is one of the top players by many people. It was actually considered the top player before V3. And while this model does not include audio, they claim this to be their best looking model yet. And what can I say? I think on some of these it actually is better. The visual fiddled, the detail in the cling images. I mean, look at this car. This is the best we've ever seen. It always used to have problem with the tires. Now we even see the car suspension on the highway. And the anatomy of this woman is kind of good, which used to be one of the biggest weaknesses of the model. It's crazy how quickly these things progress. I mean, just about a year ago, we heard about Sora. Everybody was blown away by the fact that this is even possible. And now most shots coming out of this just look hyper realistic. Impressive. The

next update/release from this week is actually coming out of Shopify. You might have caught the viral memo that came from Shopify CEO Toby telling the entire company that they're basically AI first and that this is happening no matter what and that they want to integrate it at every layer of the company. And they're doing exactly that. They shipped a bunch of new features. I'm going to skip over some of them which are not that exciting like okay, there's free image generation now for the shops. But I think most interestingly, one, they updated their AI assistant that helps you build your shop, customize your shop, and extract insights from it. I want to actually test it live on our shop here. And then they also improve their built-in AI shopping agents. Plus, they also talked about how they're now working towards being featured in Perplexity shopping consistently if you have a Shopify store. That is if you do not know both Perplexity and OpenAI have now these shopping interfaces within the app where you directed towards a purchase within chat GPT or Perplexity and Shopify is making a big deal out of actually being available on these AI platforms. Okay, but beside improvements to the chatbot, I really want to have a look at the AI sidekick as they call it. It helps you extend the shop. So what I did is I just fired up our own shop. If you're not familiar, it's very rudimentary. It gets the job done. And we basically just offer two products on here with simple product pages. And now I'm interested in the Sidekick, which I can pull up by clicking here, open Sidekick. And now let's start with something simple like improve the homepage, and see what it comes up with. Because one thing they did with the Sidekick now is that it thinks. So it doesn't just generate responses right away, but it actually uses reasoning to do it. So you can use simple goalbased prompts like this if you want. Okay. Okay, so I'm just going to click this button, create a new homepage layout, and I'm going to send that over here. And now the sidekick is actually starting to make edits. Let's see what this comes up with. Okay, so this is what it came up with. It actually just added a bunch of sections at the bottom. It's saying, hey, there should be a hero section, featured products, and then customer testimonials versus I had a hero section, sort of products, and then I had some YouTube tutorials linked. I mean, that's not really revolutionary. So, I'm thinking this is probably more tuned towards creating a new page rather than customizing an existing one. Or I should have prompted for improvements. Okay, let's see what happens if we go into the product page of our main product here. Let's say improve this to be a better converting sales page cuz that's kind of the goal here. We don't even have a sales page here. It's just this product description. And I could certainly imagine a lot of this being rearranged to make more sense. Okay, so I see it immediately going to this new block and just generating stuff in here. So, I don't think it can really edit existing content, can it? Okay, so it finished generating and it created nothing. I don't know. This doesn't seem to work exactly as I would expect it to. I think this is more geared towards building new shops and for that it's absolutely amazing to speak to an assistant rather than to have this whole interface where you have to drag and drop things and kind of think about what would be needed kind of just do the thinking for you. I see the point of that as somebody who already has a store. The blog does say that it's also for customizing stores like that, but from my initial test, it's not exactly what I would hope for. But overall, this seems mostly like a feature for somebody starting out with a blank slate. Okay, let's see what else

we got. All right, and to round things out, we'll do this week's quickfire section where we quickly brush over a bunch of AI stories and releases that are worth your attention, but maybe we don't need to spend multiple minutes on them. Starting out with the United Arab Emirates introducing free chat to everyone in their country in partnership with Open AI. Very interesting. This is a first worldwide where we really see a country pushing for AI adoption by providing every single one of their citizens with premium plants. I only expect to see more of this in the future with legislature in various European countries already shifting towards mandatory AI upskilling and we're just getting started on this front. So that's going to be interesting. Next up, I want

to talk about the follow-up to last week's story of OpenAI partnering with Johnny IV to create the AI wearable. And we have some more details and comments on this. I think the two most significant ones are on the form factor and a bit more details on the vision. So the vision for it is they basically want to create the first device that everybody should own next to a phone and a laptop and then Sam Alman actually made some comments on the form factor and he said that the device is not going to be a pair of glasses. I've had some extensive discussions over the last week about this and I think the two most likely form factors are either some sort of necklace kind of thing with either a camera, microphone or both. Or something akin to the form factor of a AirPod maybe a bit more extensive. It might go behind your ear. Again, camera microphone for context gathering and it might connect to your phone to get the computing power from there or it will be a completely standalone thing. We'll see. But I think these are the two form factors that people are mostly debating and now you know too. If I've missed something, please leave a comment. But I'm following this story with great interest because 2026 it looks like we're going to get a bunch of wearables and hardware. This is going to be one of the top players for sure. A very

interesting story coming out of Enthropic is that they open sourced the tool that they use to track the thoughts of the LLMs. A few weeks ago we talked about the fact that Enthropic was looking into how the LLMs actually arrive at the results because nobody really knows that. And now they open source software that helps them look at it along with a few examples and a cookbook on how to use this. So, by open sourcing this, hopefully more people will find ways to understand how these models work under the hood, cuz it's kind of crazy that there's components of it that still nobody really gets. And

then another quick release would be Deepseek R10528. They're brand new model. Essentially, it's just a better Deep Seek. You can look at the benchmarks right here. The light blue over here is the original Deepseek R1 and the dark blue is the new one. Essentially, it's on par or a bit under 03 in most of these. not that interesting for most consumers because the tooling is the thing that makes products like Chad GP cloud or Gemini so powerful and this is just a raw language model and that's pretty much everything for today. We're working on a bunch of videos behind the scenes so expect a few extra uploads over the coming week. If you're new here and enjoyed the video, then make sure to subscribe because my name is Igor and I do these every Friday.