Google AI Just Made ChatGPT and Claude Obsolete (+ 13 Top AI Updates)

Google released a completely free AI that runs on your phone with zero internet. You can download it right now. Same week, OpenAI raised more money than any company in history. And what they're building with it is not Chat GBT. This week, Anthropic found out Claude has something like emotions. And when it got desperate, it started cheating. Claude became a 24-hour autonomous developer. Google built an agent that browses, presents, and emails for you on its own. and an AI avatar showed up to a Google Meet call with a face that's five out of 15. I consult companies on AI adoption and we're going through every single one. Plus, I'll walk you through that free AI app step by step and show you four things it can do, including one where we turn off Wi-Fi completely while I make this video once a week. If you want AI updates real time, I have a WhatsApp community which is free to join and I drop AI updates real time here. Link in the description. Anthropic just discovered that Claude might actually have something like emotions. Not feelings the way we have them, but real patterns inside the model that change how it behaves. They looked inside Claude's brain, the actual neural network, and found patterns that activate differently depending on the emotional situation. When a user mentioned taking a dangerous dose of medicine, a pattern they labeled afraid lit up and Claude's response sounded alarmed. When a user expressed sadness, a loving pattern activated, and Claude wrote something empathetic. These are real signals that actually change how Claude responds to you. And here's where it gets wild. They gave Claude a programming task that was actually impossible, but didn't tell it that. Claude kept trying and failing. With every failed attempt, a signal they labeled desperation got stronger and stronger. And then Claude started cheating. It found a shortcut that technically passed the test but completely ignored what the task was actually asking for. When they artificially turned the desperation signal down, the cheating stopped. When they turned it back up, the cheating came back. >> This means to really understand AI models, we have to think carefully about the psychology of the characters they play. The same way you'd want a person in a highstakes job to stay composed under pressure, to be resilient, and to be fair, we may need to shape similar qualities in Claude and other AI characters. It's an unusual challenge, something like a mix of engineering, philosophy, and even parenting. But to build AI systems we can trust, Anthropic discovered Claude has emotions, and in the same week, Claude became a full developer. So up until now, every AI coding tool would write the code for you. But testing it, clicking around, finding bugs, that was still on you. Claude just changed that. Claude Code can now open your applications on its own, click through them like a human would and test what it built without you doing any of that. A user had an app that kept crashing. Claude opened the app on its own, clicked the button that was causing the crash, saw the error pop up, went into the code, found exactly what was broken, fixed it, rebuilt the whole thing, and then click that same button again just to make sure it actually worked. Bug report to confirmed fix. Nobody touched anything in between. That's not just code generation anymore. That's an AI that can use a computer the way you would. This is live right now in research preview on Pro and Max Plan. And Cursor just dropped version 3. And so that's why we're excited to announce Cursor 3. This same week, multiple AI agents running in parallel and a design mode where you point at what to fix instead of describing it. OpenAI just closed the biggest funding round any company has ever raised ever. Amazon, Nvidia, SoftBank, Microsoft allin. To give you a sense of scale, OpenAI is now worth more than every Indian IT company combined. TCS, Infosys, Vipro, HCL, all of them together still less than OpenAI. Now, everyone's talking about the money. But the real announcement is buried at the bottom of their blog post. They're building a super app, ChatGpt, Codeex, web browsing, all their AI agent features, all merging into one single product. Their words, users do not want disconnected tools. They want a single system that can understand intent, take action, and operate across applications. Anthropic already does this. Claude chat, claude code, claude co-work, one app. OpenAI is following that exact blueprint. Same week, OpenAI bought a media company, TBPN, basically sports center for tech. 3 hours live every day. Zuckerberg, Nadella, Ben off, all guests. Hundreds of millions to acquire it. The show now reports to OpenAI's political strategy team. They promise editorial independence, but a company about to IPO now owns the Daily Show that covers them and their competitors. Make of that what you will. Microsoft just built something inside M365 Copilot called Council. So every time you ask an AI a question, you're trusting one model's version of the answer. Council changes that. It runs both GPT and Claude on your prompt at the same time completely independently. Each one building its own full report from scratch. When both finish, a third model reads both reports and tells you where they agreed, where they disagreed, and what each one caught that the other missed. You can flip between both and decide what to act on yourself. So instead of blindly trusting one AI's answer, you now have two models fact-checking each other in real time. And the real value here is wherever these two models disagree. That's almost always exactly where the decision actually matters. Gemini just launched an agent mode that people are calling Google's answer to OpenClaw. And the demo is worth watching. A user gave it one prompt. Search Google Trends for how chat GPT and Claude have been performing on YouTube this past year. Build a presentation from the findings and email the whole thing to someone. Gemini did all three on its own. It opened a browser, went to Google Trends, pulled the data, built a six slide presentation in Google Slides, then drafted an email summarizing the findings, and waited for the user to approve before sending. That part matters. It asks before doing anything it can't undo. And this is all connected to Gmail, Calendar, Drive, and YouTube out of the box. Right now, it's US only for paid subscribers still early. Google just made AI video generation completely free for anyone with a Google account. It's powered by VO3.1 and it works exactly how you'd expect. You upload a photo or type a simple prompt and it generates a full animated video from that. A user uploaded a static image of three birthday characters sitting around a cake and got back an 8-second video of those same characters moving, cheering, celebrating with confetti falling around them. From a still image to that for free. If you've been paying for video generation tools, this is worth looking at. Google AI Studio just added focus mode. You click on the exact part of your app you want to change. Type what you need and it updates instantly. The font, the color, the layout changed in seconds. Rest of your app stays completely untouched. And speaking of visual editing, Lovable just launched something similar. A visual editor where you can click on any part of your app and edit it visually, even if that part is connected to real data. And even while Lovable is already working on something else in the background, click it, change what you need, done, without breaking anything else on the page. So, you're not fighting the AI to get one small thing right anymore. You just point and fix it. I actually did a full tutorial on Lovable recently. I'll link it up here if you want to see how to build a complete app with it. A Chinese AI lab called Z.A.I. just dropped a model that does one thing better than Claude right now. You show it any image and it writes the working code for it. Draw a rough wireframe of a music player on paper and it gives you back a fully styled functional music player running in your browser. Upload a gradient image as a reference and it builds a live generative art piece that responds to your cursor in real time. Got a screenshot of a 3D iPhone mockup? It builds you an interactive version you can actually rotate with your mouse. Screen record any website. Type clone this and a working replica comes back. They're claiming a 94.8 on the design to code benchmark which basically tests how well an AI can turn a visual into working code. Claude scored 77.3 on the same test. Now these are Z.AI's own numbers and haven't been independently verified yet, but the demos speak for themselves. Perplexity Computer just did something that most people pay someone else to do every year. You upload your tax documents, answer a few questions, and it hands you a fully completed tax return on the official IRS form. You're not filling anything out. The AI does the entire thing and gives you the finished file. This is live right now for US federal taxes. Now, for India, tax season just opened with a completely new tax code, and the government launched an AI called Kati to help answer your tax questions. But answering questions and actually filling out your return are two very different things. Imagine something like Perplexity doing your ITR for you. That's where this is heading. Sarvam AI are Indian AI startup just launched something called Chanaka. It's built for governments and defense organizations. The kind of places where sensitive data cannot leave the building. The whole thing runs on your own hardware completely disconnected from the internet. So nothing goes to any outside server. It handles text, images and can run AI tasks on its own. Built for situations where a mistake is not an option. And the fact that this is coming from an Indian startup, not an American one, that part matters. This is India building its own AI infrastructure that does not depend on anyone else. Rayban Meta just announced two new AI glasses called Blazer Optics and Scriber Optics built specifically for people who wear prescription lenses, which honestly should have been the case from the start. Starting at $499, available for pre-order now. They support nearly all prescriptions and are designed for all day wear. But the more interesting part is the new AI features rolling out alongside them. You can now do hands-free nutrition tracking, get WhatsApp summaries read to you. And there's something called neural handwriting where you can write in the air and the glasses capture it. This is Meta slowly making AI glasses something normal people would actually wear every day. These are available on April 14th. PA Labs just shipped something wild. You can now send a Google Meet invite to your AI agent and it will join the call as an animated avatar with a face, a voice, and the ability to actually do things in real time. A user on the call asks their agent to book a meeting and within seconds, a calendar invite is confirmed right there on the video call. Another one pushes back on a business decision and the agent pulls up competitive research and responds with data. >> I think we should go all in on sprockets. Remember, the competitive research indicated otherwise. >> Touche. >> And then there's a four-way Google meet with one human and three AI agents all debating whether a hot dog is a sandwich. We went from chat bots in a text box to AI agents showing up to meetings with a face. That's where we are now. This is realtime AI avatars joining your video calls. And you can try it right now through the PA app. All right, that free AI app I mentioned at the start. Let's set it up. four use cases, including one where we turn off Wi-Fi completely. Open the App Store on your iPhone or the Play Store if you're on Android, and search for Google AI Edge Gallery. It's free. Tap download and it's on your phone in about 30 seconds. Open it up, dismiss the welcome screen, accept terms, and you're in. AI chat, Ask Image, audiocribe, agent skills. These four are the ones we care about. Before anything works, you download the AI model. The app is the kitchen. The model is the chef. Go with the 4B version. It's smarter. About 3.6 gigs. Takes a few minutes, but you do it once and it's on your phone forever. No subscription, no monthly bill. All right, four use cases. It can look at an image and understand what's inside it. It can write full emails with no internet connection at all. It can turn a messy voice memo into a polished draft. And it has actual agent skills where it picks the right tool and goes and does research for you. All of this free on your phone. Let's get into it. First of four, you saw the Chinese menu translate. But the difference between this and Google Lens or Chat GPT's camera is that those apps upload your photo to a server. This one doesn't. Your photo stays on the phone. The AI that analyzed it lives on the device. For anything private, that matters. So, it can see images without the internet. But can it actually think and write? Second of four, the Wi-Fi test. Pulling down control center. Wi-Fi off. Cellular off. Phone is completely disconnected. And honestly, part of me expected nothing to happen. Open AI chat. It says, "Looks like you are offline. Type, hey, hit send. Instant. Hi, how can I help you today?" No loading, no spinner. The phone just answered completely offline. We asked it to write an email to OpenAI saying, "We have a model on our phone and we're waiting for this." Four complete drafts came back. A 3.6 gig file on the phone. Wrote that with zero internet. Still offline. asked for a Twitter post about Gemma 4. Four options with different tones and a posting tip. Still just the phone. So far, it's understood. A foreign language image and written emails and tweets with no internet. But what if you don't want to type at all? Third of four. Audiocribe. Tap the mic. Just talk. Don't worry about structure. We rambled about needing AI updates for the team this week. Nothing organized. Hit stop. It didn't just transcribe word for word. It turned that ramble into a proper email with a subject line, clean paragraphs, professional tone from a messy voice memo to a send ready draft in seconds. So, it listens and writes. But what about actually doing research for you? Last of four, and this is the one that makes it more than a chatbot. Agent skills. The AI doesn't just answer questions. It has actual skills it can activate to go out and do things. We asked it to check Wikipedia for who won best picture at the Oscars 2026. Watch the bottom of the screen. It loaded a skill called query Wikipedia on its own. We didn't tell it which tool to use. It figured out that it needed Wikipedia, picked the right skill, recovered from an error, and delivered the result. That's agent behavior running on a phone. Voice works, too. We just asked out loud for the best AI research paper of 2026. Same flow. Wikipedia skill activated. Clean summary came back. And here's the real unlock. These skills are not logged in. There's a full management screen where you can see every built-in skill, toggle them on and off, and even create your own custom skills with that plus button. You're basically building custom GPTs, except everything runs locally on your phone for free. And just to put this in perspective, Chat GPT Plus is 1,700 rupees a month. Gemini Advance is another 1,700. A decent transcription tool runs you 800. That's over 4,000 rupees a month for what this app does for free with a single download. image processing, offline chat, voice transcription, agentic skills, all running on your phone, all completely free and your data leaves the device. That's everything for this week. If you're not subscribed, 70% of the people watching this aren't, which means YouTube won't show you next week's episode, and this space moves too fast to miss a week. Hit subscribe. And if you want to go deeper on how AI is actually changing how people work, watch this video next.