AI News: Claude Can Now Control Your Computer!

News, news, news. This is really uncomfortable. Oh, what am I doing with my life? This is my first news video of the year since I've been home from ces and it's been a pretty interesting week. There's been tons of updates from Google, some fresh drama from OpenAI and Anthropic. Even DocuSign got in on the action. But we absolutely have to start with Claude's launch of their new Cowork product. Let's check it out. So what is Claude Cowork? Cloud Cowork is essentially Claude code for people that aren't interested in coding. I guess they say when we released Cloud code, we expected developers to use it for coding, and they did and then quickly began using it for almost everything else. That prompted us to build Cowork a simpler way for anyone, not just developers. Basically, you can give Claude coworker access to different folders on your computer, and it can do things on your computer on your behalf. Taking a look at their demo video here, you can do things like create a file file prep for the day, crunch data, organize files, make a prototype, and send messages. They give an example where they give a folder containing meeting transcripts and say, summarize meetings from this week and find action items. Where do you think I can be more efficient? It looks through all the transcripts for them, asks them how detailed they want it, and then it creates a checklist that it then follows through on. They also ask it to check their calendar for what's urgent tomorrow. It adds that to the task list and builds this at a glance page for them. Here's another demo example they share where they ask it to please help organize my desktop. You can see an insane desktop there and it processes the prompt, creates a little task list, and we can see that it starts actively going and reorganizing the desktop and making it pretty clean. Now this is something I want to test real quick. A couple caveats about it. It only works on Mac right now and you have to be on the max plan to use it. The lowest available max plan is 100 bucks. So yeah, it's not for everybody yet, but this is still kind of like a demo bas beta tester mode. So they're rolling it out to their highest tier plan to let those people test it, and then eventually it's going to roll into the pro plan and then the free plan. Hey, real quick, future Matt here. When I recorded that thing about Claude Cowork, it was only available on the max plan, but since recording that, it actually has come out now that it's also available on the 20amonth plan. So wanted to make sure I shared that because now more people can use it. Yay. I am on the max plan. So I'm going to try it myself real quick. I can't really have it organized. My desktop because I'm kind of a psychopath, keeps my desktop clean all of the time. However, I cannot say the same about my downloads folder. I just kind of happen to never clean this. So as of right now, this is only on the Claude Code desktop app and again only on Mac. But you get these three tabs up here where you can use the normal Claude chat. You could use Claude code. And then we have this cowork option here. Let me kind of pull this to one side. We'll keep our downloads folder here and it actually pre gives me an option under what needs tidying up and I'll go ahead and select my downloads folder. It then asks me if I can allow Claude to change files in my downloads. So I'll go ahead and say allow. This time I'll click Allow again on the actual Mac dialog box and we can see the prompt that it wrote for me. Help me organize my downloads folder, scan the contents and propose a plan. Categories folders to create how files should be sorted. Any naming conventions to apply files to flag for review or deletion. Show me the plan before making changes. Only proceed after I approve. All right, let's go ahead and do that that and see what happens. So we can see that it's running some commands here, reading some files. Now let me also check the contents of the existing subfolders. All right, so it says, now I have a complete picture. Let me create a comprehensive organization plan based on my scan. You have roughly 270 items in your downloads folder, including nine existing subfolders. Here's my proposed organization 3D models STL files for 3D printing installers dmg pkg apk files so it's suggesting new folders for the various files. Naming conventions. Files to flag for reviewer deletion. So there's some duplicates. Apparently I've downloaded Zoom multiple times, so summary of actions create 10 new folders. Move 150 files. Flag 15 duplicate sets for your review. Flag 35 gigabytes of old installers for potential deletion. Yeah, that sounds like a smart move. Please proceed. Now let's see what happens. So my downloads folder is over here. So let's see if we can watch it do this thing in real time. All right, so it's Running some commands, so just looks like it did some sorting. So I think it created all the folders that it said it was going to create. I mean, I'm seeing files move around, so it's definitely doing something. All right, so moving install files, we can see on the right. There's a little checklist here which is showing me the progress as it does these steps. So it's moved the installer files, now it's moving font files. If I click show 10 more, you can see all of the steps that it's going to work through. What's nice about this is that I am completely hands off right now. I could just walk away from my computer and let it just keep doing this thing. And theoretically when I come back, my downloads folders organized or, you know, open up YouTube and watch a video or something. I recommend my recent CS videos. They're pretty fun. Look at that. Almost completely cleaned up now. And now everything's been moved into folders. It's still working here and now it's generating a summary for me. So essentially what this is doing is just running terminal commands on your computer on your behalf. It knows what terminal commands it needs to run to add files, remove files, create folders, reorganize things, and it's just running those terminal bash commands behind the scenes for me. There we go. Organization complete. And also it gives me a pretty clear picture of where I can probably free up some storage. Right, videos. I've got 66 gigabytes worth in there. The projects folder has 43 gigabytes. Archives has 47 gigs. AI models, 9.2 gigs. That's pretty impressive. So now you can actually use this new Claude coworker to essentially manage your computer for you. Again, only on Mac, and only if you're paying over 100 bucks. But they will eventually roll it out for Windows and roll it into the less and less expensive plans over time. But this is probably one of the more genuinely useful tools or features we've got out of one of these big model makers recently. We also got a ton of updates out of Google this week, and I'm going to share them all with you. But the biggest by far was Gemini's new personal Intelligence. Personal Intelligence makes it so you can actually connect the Gemini chatbot with a bunch of your different Google accounts. At the moment, it connects to Gmail, photos, YouTube and Search in a single tap. I imagine in the future it's going to connect to your calendar, which is one of the ones. I'm kind of surprised it didn't Connect to right off the bat as well as Google Drive. And I mean, you know, the Google ecosystem, it's going to be able to connect to everything at some point. Most likely. One of the best examples of this personal intelligence in use actually came from Josh Woodward, the VP of Google Labs, Gemini and AI Studio. His example was, what if he's at a tire shop and going to purchase new tires, but didn't want to run outside to his car to figure out what tires he actually needed? Well, because Gemini is connected to his Google Photos, there's probably some photos inside of his camera feed. So he gave it the prompt. Gemini did its thing, did all of its analysis, and it said, for a vehicle like the Honda Odyssey, using two 3560 R18 tires, etc. Etc. But the idea is here, all he said was, I need to replace the tires for my car. Which ones would you suggest for me? He didn't say what his car was. It used his camera roll to figure what tires he would need. He then asked it, what's my Odyssey license plate number? Again, because it has access to its photos, we could see it used the personal intelligence and then it shows his license plate number. He actually did a whole video that he posted on X where he broke this whole thing down. And we can see here he actually says it was able to reference the photos from Google Photos. So it's like now, if you just want to remember stuff, you can just take photos, save them to Google Photos, and then Gemini will be able to pull that information for you at any time when you're chatting with it. This sounds great for like screenshots and quotes and things you want to remember in the future. Just save them to your Google Photos. Now they say you can choose to turn it on. It's not on by default. You can decide exactly which apps to connect to and you can turn it off whenever you want. Now, Gemini does access your data to answer your specific requests, but things like your photos and emails and Gmail do not directly train their model. So none of that is being saved as training data. So your license plate number or the type of car you drive isn't being saved in their systems and getting trained into the models. You sure about that? Now, this isn't available on the free plan yet. You do have to be on a Google AI Pro or AI Ultra plan and you do have to be in the US for right now while they're sort of testing it out. But it does work on Web, Android and iOS, and it does work with all of the various Gemini models. I do have one beef with this right now, a little bit more of a frustration because it's not really that useful to me yet, and that's that it's available for personal Google accounts right right now, and not for Workspace business accounts. I pretty much exclusively use Workspace business accounts, so my main Gmail that I use, my Google Photos account, my Google Calendar, all of those types of things are all actually on my Google Workspace account connected to my mattwolf.com domain. Now I do have a personal Gmail account, but I only use it for personal emails. There's like no business happening in that email account, so this isn't really that helpful to me yet. But let's give it a try anyway. Real quick, before we get back to the I get a lot of questions about how I keep up with the crazy fast paced AI world every single week. And to be 100% honest, I use a lot of AI. I mean, duh. One of the things that has actually helped me make this process so much faster and easier is Comment. It's Perplexity's web browser and it makes it so much easier to use the power of AI in my everyday work. So while I'm already doing what I would normally do, like read articles and scan social media and watch YouTube videos and dig through documents and press releases, Comment is there to just help me do it better and faster. For example, I'll be reading three or four different AI news articles at the same time, but instead of taking notes manually, I'll ask Comet to summarize the key takeaways across all of my open tabs and tell me what actually matters. Then I'll jump over to YouTube and I'll have Comet scan a 40 minute video reviewing a new AI platform and it'll pull in the most interesting timestamps and give me a quick breakdown without having to watch the whole thing. And when I'm researching tools or companies, I'll have It organize my tabs into a single workspace and compare claims across sites and even export a clean summary straight into Google Docs. For me, the big difference is that Comet can actually see the page that I'm currently on, so it knows the context when I ask it to do something for me. And then I can just keep doing whatever I want while all of this runs in the background. So if you're interested, Comet is free to use and easy to get started on, so check it out in the link in the description. Thank you so much to Comet for sponsoring this portion of today's video. Now let's get back to it. Okay, so here I am inside of my Google Gemini account. This is a personal account, not a work account, and it is an ultra plan. So I should have access to personal intelligence. Now, full disclosure, I have had access to this for a couple weeks now. They gave me early access to play around with it and offer feedback. So I have tested it a little bit. But if you do want to use it, you do need to go and turn it on. So in order to turn it on, you click on the Settings and Help button down in the bottom left, select Personal Intelligence, select Connected Apps, and flip the switch on pretty much everything. And as of now, it looks like it does actually connect to Google Calendar and Google Drive already. I don't feel like that was made super clear on their article about it, but from this screen, it looks like it will actually connect to those things right now. So now if I head back over to Gemini, I don't actually need to do anything else. It's now just connected to my personal intelligence. I'm going to actually test one of the things that they showed off in that previous video. And if it works, I'm going to have to blur it. But what is my license plate number? All right, so it does say personalization in progress. Connecting. Pinpointing the answer. All right, so it found a few license plate numbers here. Let's make sure we actually blur those. What type of vehicle was each plate number on? We could even see that it actually says as the title identifying license plate numbers from photos. So it was looking in my Google Photos. It's searching Google Photos again. And, well, yeah, it did actually find my actual license plate numbers among my photos on my old Google Photos account. I don't think there's been any new photos added to this account in, like, years. So two of the license plate numbers it found were on vehicles I don't even own anymore. But it also found the license plate number of the vehicle I currently own among my photos. So, yeah, this definitely works, which is why I have to blur so much. Let's see. Are there any urgent emails in my inbox right now. Now. All right, so it's connecting to the Google workspace, and I pretty much have to blur everything on the screen right now. But it did find some emails in my personal inbox. But quite honestly, none of it's actually super urgent because it's not connected to the email that I typically use. But that's Gmail's personal intelligence. It's going to connect to more and more of the things that you use in your life so that when you chat with it, it has all of those areas at its disposal to help you get to the perfect answer for yourself. So pretty cool. But even more useful once it's available in the workspace plans and also once it's available on free plans. What these companies tend to do is they roll them out on their higher tier paid plans to get people testing it and finding bugs and sort of helping them work through all of the issues. And once those issues are worked out, then they start rolling it into other tiers and it appears that's what's happening right now. But come on, give it to us in the workspace plan. I really, really want it in the workspace plan. But I mentioned we got a bunch of updates out of Google. That was just the beginning. We also got some upgrades to Google's VO 3.1 video model. They improved the Ingredients to Video feature. They say even with short prompts, you can generate dynamic and engaging videos based on ingredient images. You'll now see richer dialogue and storytelling, making your videos feel more alive and expressive. They give it the prompt documentary style. A raccoon manages a coffee shop dialogue. That was the whole prompt. I always wanted to contribute, dude, I'm just here to make the best coffee possible. And you've achieved that? Absolutely. They've improved the identity consistency, which makes it easier to keep your characters looking the same even as the setting changes, making it easier to tell a full narrative. By having the same character appear across multiple scenes, you can achieve background and object consistency. Control the scene by maintaining the integrity of your setting and the objects within it. You can also reuse an object backgrounds or textures across scenes. You can seamlessly blend textures, characters and objects. Combine disparate elements like characters, objects, textures and stylized backgrounds into a cohesive high impact clip. They now have native vertical output for Ingredients to Video. So for the first time, Ingredients to Video supports generating videos at 9.16and they added the feature to upscale to 4K resolution. These features are available in YouTube Shorts and the YouTube Create app, and you can also try the enhanced 3.1 in portrait mode for VO in the Gemini app. It's also available in Flow, Gemini API, Vertex API and Google Vignette Kids. All right, so let's test this out. I'm gonna go ahead and upload a picture of this dude right here. Drop it in and we'll say make this character doing jumping Jackson in 9:16 video. And apparently I can't do that. So I don't think it wants me to upload faces of real people. So I wonder if I give it like a fictional character. I have this like AI generated Jedi type character character. Maybe it'll do it. Put this character in a battle video. Should be 9:16. And well, it didn't give me the 916. Let's see how the video looks though. The video looks good. I mean, it got the character consistency, but it didn't do the 916. So maybe that feature doesn't work inside of Gemini yet. So let's try it over in Flow instead. Although it is experiencing high demand, so hopefully this will work. But we'll do a new project. Make sure we're in portrait mode here. We've got 3.1 quality. I've got this pirate image that I've generated in the past, so I'll just pull that in first. I gotta switch this over to ingredients to video. Now we can add our image in. So let's upload this guy here. My real question is, if I upload this as a16.9 image but the output is a916, how is that gonna work out? I'm gonna give it a short prompt. An epic SW fight between pirates. Make them talk like pirates. Okay, this feature isn't supported on VO 3.1 quality. Push to V 3.1 fast. Okie dokie. And let's see what it gave us. So here's the first one. You'll never get me treasure, you scoundrel. Over my dead body, you build rat. Take this. Not today. I feel like he just growled like a lion or something. All right, here's another one. Yes. Scurvy dog. Prepare to meet the character. Consistency is there. It converted the wide image to a 9. 16. So yeah, that's the new version of VO 3.1. All right, moving on to other Google updates, because again, there was a few. Google also updated their search trends. It's now using Gemini behind the scenes inside of Google Trends. So if I head over to trends.google.com let's go ahead and explore a trend here. Artificial intelligence. I mean, it does look like a cleaner user interface here, but let's go ahead and suggest search terms. This was kind of the new feature here. And what is your area of interest? I'm going to put artificial intelligence again, find search terms. And it used artificial intelligence to search terms for artificial intelligence. We could see some of the terms. Chat, GPT, deep learning, machine learning, AI and healthcare AI, ethics, et cetera. And it gives us other ideas to explore. So this is all being pulled in using Gemini now. Now, one thing to keep in mind, it does say down at the bottom, the new experience is initially launching on desktop roll out gradually. I tried using this on one of my Google accounts and it didn't work. And then I logged into my personal account, the same one I had to use earlier, and it did work. So it's still rolling out, I guess. Google also rolled out some new tech and tools for retailers. This week they launched the Universal Commerce Protocol. This UCP will soon power a new checkout feature on eligible Google product listings in AI mode in search. So it doesn't sound like it's rolled out yet, but if you talk about product names and the AI search mod pulls up those products, well, they're going to put buy buttons basically right there to make it easier for you to purchase immediately. Here's an example right here where they give it a prompt looking for a modern stylish rug. It gives them some options. Here they click on one of the rugs and we can see that there's actually a price tracker that uses AI and the ability to purchase directly from within the AI search mode. Google also released a new model called Translate Gemma, which is an open source translation model. It's using Gemma 3, one of Google's open models and it works across 55 languages. Right now they have various model sizes depending on the device that you're running it on. So the 4 billion parameter model is optimized for mobile. The 12 billion is designed to run on consumer laptops, and the 27 billion parameter model runs on the higher end H100 GPUs or you know, Google's TPUs. This is an open model. I don't believe there's a place where you can use it directly at the moment, but you can download it on on Kaggle and Hugging Face and run it locally on your computer or your preferred cloud GPU service. And since we're Talking about translation, ChatGPT actually quietly rolled out a new translation feature. It's weird to me because they sort of hype up everything. So the fact that they rolled this out sort of stealthily is kind of fascinating. But it's a translation tool inside of ChatGPT that supports over 50 languages. They've positioned this as a direct challenger to Google Translate and I believe you can actually use it for free inside of the free Chat GPT plan. If you head over to chat GPT.comtranslate, that's where you can access it. Now it doesn't show up just in your regular Chat GPT account. So you actually have to go to this URL to access it, at least at the time of this recording. But as we can see here, you can have it detect a language or pick a specific language and then translate it to any one of these languages. So if I type some text over here, it will translate it to Spanish for me. Hola, mi nombre es Matt Wolf E ago videos in YouTube. Sobre intelligencia artificial, por favor, subscribite. I'm sorry, those of you who speak Spanish for totally butchering your language, let's switch it to Russian and we can even have it speak it out. Nope, we can't. It says failed to play message. No wah wah. So that feature does not appear to be working at the moment. All right, let's move right along because this week we have some a high drama that I want share with you. The first bit of AI drama comes out of Thinking Machine Labs and OpenAI. So if you're not familiar with Thinking Machines, the CEO and founder of Thinking Machines is Mira Murati. Now, Mira Murati used to be cto over at OpenAI. In fact, she was, for a very, very, very short window of time, actually CEO of OpenAI. When Sam Altman got ousted, not too long after all of that drama, she actually left OpenAI and started this new company, Thinking Machines. Now, the CTO of Thinking Machines was someone named Barrett Zof. Barrett was somebody that Thinking Machines essentially poached away from OpenAI. When she went and started this company, he followed well this week, Thinking Machines basically fired him for unethical conduct. On the same day that he was fired for that unethical conduct, he was rehired by OpenAI. Well, a source told Wire that Zoflex here was passing confidential company information to competitors. OpenAI announced that Zof was returning in an internal memo on X saying the company doesn't share Marathi's concerns. So just to tie this all together, Barrett zof worked at OpenAI. Mayor Murati also worked at OpenAI, but left to start Thinking Machines. She brought Zof with her. Zof eventually got fired for unethical conduct and on the Same day rejoined OpenAI. And the story behind the scenes was that he was passing confidential information to competitors. What most of us are gathering from this was he was like a double agent for thinking machines and OpenAI. Like he went over to Thinking Machines to sort of keep an eye on what they were doing and feed that information to OpenAI. Now, this is all just rumor and speculation. The source isn't actually known, but this feels right out of like a crazy drama movie. And oftentimes what's happening in Silicon Valley feels like that. But this is one of those stories that's just kind of wild. And another bit of drama that's happening right now is coming out of Anthropompy. People who do the whole vibe coding thing are not really happy with them at the moment. And here's the reason you can sign up for a Claude Max plan. It starts at 100 bucks. I'm on a cloud Max plan. I think I'm actually paying $200 a month for that Claude Max plan. But if you're on one of those plans, well, you get pretty much like unlimited access to Claude code and their latest model. Inside of Claude code, however, there's some other ides that people are coding in. The most notable one being one called Open Code. People were taking their Anthropic API key and using over on Open Code instead of Claude code. That was something that was allowed on your Claude Max plan. You got access to the API and you got a ton of usage from this Claude Opus model. Well, apparently Anthropic did not like people not using Claude code with their API key and using other IDEs. They didn't make any big announcements or anything like that. But this guy here, I think it's pronounced Tariq, who works on Claude code for Anthropic, said on Twitter yesterday we tightened our safeguards against spoofing the Claude code harness after accounts were banned for triggering abuse filters from third party harnesses using CLAUDE subscriptions. Basically saying people aren't allowed to be using their Claude code subscription over in other tools outside of Claude code. The hilarious part was notice how this was on January 9th where this went down. People were really, really frustrated with this. Comments like you killed an entire ecosystem. This is a losing battle. Those safeguards won't work. Opus 4.5 is dominating right now. It would be a smarter move to work with the Open Code team to make this work and well, I'll scroll down. You could read some of the comments here. People were not happy about this and well, OpenAI used this as an opportunity that same exact day, January 9th. Thibaut here, who works on Codex at OpenAI, which is a Claude code competitor, said, we are working with Open code to allow Codex users to use their Codex subscriptions and usage limits on Open code directly. Also exploring how to support other awesome actors in the space. Yeah, people were happy about that and people were really unhappy about the whole anthropic thing. So much so that people wrote entire write ups about how unfortunate this is and how it's a major misstep for Claude. They're basically telling their customers that are paying, you know, hundreds of dollars a month, you must use Claude code. And if you use this API key somewhere else in a different platform, then you're dead to us essentially. All right, almost done. I've got a few more news stories I want to share with you, so let's jump into a rapid fire and go through them real quick. We'll start with another really big story that came out this week and that's that Apple is going to use Google's Gemini in Siri. They released a joint statement this week saying that Apple and Google have entered into a multi year collaboration under which the next generation of Apple foundation models will be based on Google's Gemini models and cloud technology. So not only does Google have Gemini in Android, Chrome in Google, Drive in Gmail, in, you name it, Google owns all the platforms. They now are also getting it into Apple phones. So basically, Gemini is going to be the AI on your phone because whether you're using an Android, you're getting Gemini, or whether you're using an iPhone, with Siri, you're going to be getting Gemini. Apple Intelligence still has its own sort of smaller on device model, so it's not always going to use Gemini whenever you use Siri. But it sounds like it's going to be using Gemini when it has a more complex query where it needs to tap the cloud to get a response. Huge, huge win for Google. I found this one to be interesting. OpenAI partnered with Cerebras now. I found it interesting because Cerebrus is probably Grok's biggest competitor and when I say Grok, I'm talking about Groq Grok, the company that makes the chips that Nvidia just acquired. Both Grok and Cerebras make these ultra, ultra powerful fast chips that are designed for inference. So most of Nvidia's chips are designed and work really well for training, but these are optimized for once you type a prompt, getting that response, response back. It's interesting to me that OpenAI partnered with Cerebras because OpenAI has a really strong relationship with Nvidia who just acquired Grok. So I would have thought OpenAI would have gone and used Grok through their relationship with Nvidia, but no, they're going to the competitor of the company that Nvidia just acquired. I don't know. It's all just fascinating to me. Most people watching this probably don't care, but I nerd out about this stuff. I also teased at the beginning of this video that DocuSign's getting in on the action. Well, they're adding a new feature that translates legalese and does the contract busy work for you? I don't know about you, but every once in a while I get a contract and DocuSign that I need to sign for whatever reason, whether it be like an NDA or a contract for some sort of partnership deal I'm doing with like HubSpot. Because my podcast is hosted by HubSpot and my first course of action is to take that entire document, copy it, paste it into Claude or chat GPT and have it analyze it for me, look for any gotchas for me or anything that they're getting too good of a deal out of me for. I see what it tells me and then I use that feedback as part of the negotiation process before signing the contract. Well now you don't need to do that copying and pasting anymore because those functionalities, those features are just going to be built right into DocuSign. And finally, I'll wrap up with this one. There's a new image model on the scene called GLM Image. It's from the company Za AI and it's an autoregressive for dense knowledge and high fidelity image generation. Basically what this means is it's trying to be a competitor to what Nano Banana is doing and what the latest ChatGPT image model is doing, where when you give it a prompt it will sort of go and research the information and then that information can be worked into the image. Except this one is an open model. We can see some examples on the screen of the type of images it made. I don't really feel like it's up to par with like Nano Banana, but again this is an open weight model here, so you can actually Download it on GitHub and Hugging Face here and run it locally or in your favorite cloud GPU service. It's always fascinating to me that like you get these state of the art models like the Nano Banana model and then you've got the ChatGPT image model and like literally weeks later, like six weeks later, we're getting open source versions of that same thing. The speed of which that gap is closing blows my mind. Just wanted to share that one with you. Haven't talked tested this model, haven't really wanted to make the effort of downloading it and running it locally yet. If you guys are interested, you could let me know and maybe it's a model we'll play around with in the future. But that's what I got for you today. Hopefully you feel more looped in on what's going on with the AI news. I do my best every single week to keep my finger on the pulse of everything that's happening in the world of AI and to some degree, the world of tech in general. And then every week I turn around and make a video recapping what I think is important to talk about. I try to sift through the noise and find that signal for for you and share what I think is most important to the most amount of people. So if you like that kind of thing, make sure you like this video and subscribe to this channel that will ensure more videos like this show up in your YouTube feed. Thank you once again for tuning in. I really, really appreciate you hanging out nerding out with me today. Hopefully I'll see you in the next one. Bye. Bye.