AI News: Gemini Flash 3.7, NotebookLM Update, Grok Bot, AI Watermarks and More

0:00Well, this was a pretty busy week in the world of AI. We have a few updates from Google including a new flash model and a small update to Notebook LM that we've been waiting for for a while now. SpaceX AI released a new agent called Grok Bot that is really promising, but most people talking about it are leaving out

0:17some pretty crucial details about it. But don't worry, I'll break down everything that's important in this week's AI news recap. My name is Paul J. Lipsky and every week I bring you the AI news that regular people actually care about. I want to start by talking about that little update to Notebook LM, or as it's now called Gemini Notebook. So now when you come into any of your notebooks like this one right here, you'll see this copy button at the top. This will make a copy of your notebook. You'll see whenever you do that, it will automatically give it a title of copy of and the name of your notebook, but you

0:51can rename it. So this is really helpful when you have one notebook that you like and you want to sort of branch out from it and create a variation of it. Maybe add some sources to that one while still preserving the original notebook. This makes it a lot easier to do that. The other thing that it's really good for is when someone shares a notebook with you, but there's a little bit of a catch here. So back here on my dashboard, if I click right here where it says shared with me, I can see this notebook that someone else shared with me, but opening it up, you'll notice that the copy

1:24button is grayed out. And that's because the original owner has to allow copying. So if you do want to share a notebook with someone and allow them to be able to copy it, this is how you do it. Come over to the notebook that you want to share. So let's say it is this one right here. And then you'll click on the top right here where it says share. Now if you see that this share button is grayed out, it's probably because you have used this notebook inside of Gemini and you have some Gemini chats in there. So for instance, if we come over to this

1:58notebook right here, you'll see that I cannot share this because over here in the sources, I have chats from Gemini. Unfortunately, that is a limitation with sharing notebooks. But anyway, let's come back to the notebook we can share, which is this one right here. And I'll click on the share icon, and then you can enter in the email address

2:21of the person you want to share it with. You can choose if you want to give them access just to view the notebook, to edit the notebook, or if you want to remove access. And then finally, down here is a box you can check off that says allow copies. And if you check this off, then they can make their own copy

2:37of the notebook just like I showed you. It does appear that there are some further limitations with this. Like I tried this with a workspace account, and I was unable to allow copies. So, just be aware that different account types, it may not be available. And since we just talked about Gemini notebook, let's talk about all the updates from Google this past week. So, first, Google announced that Gemini is going to get more connectors to third-party tools, 14 new apps. So, Gemini will now connect with OpenTable, Ticketmaster, Wix, iHeartRadio, Fever, Pandora, GetYourGuide, Angie, what is Angie? Oh, lawn treatment services. Okay, Otter.ai, which is for meeting transcripts. We

3:24have Localiza, Thumb Tack, Zocdoc, Granola, which is also for meeting transcripts, Zoho. Yeah, so that appears to be all of them. This is something that has slowly been happening over the last couple of months. I've been saying for a while now, Google really needs to move on this quicker because this is something that's holding back a lot of people from using Gemini and Gemini Spark because it doesn't work with that many third-party tools. Now, they are opening it up and they've even said they want to work with more third-party tools. So, this is good to see. We just need to see more of it and quite

4:01frankly, it needs to move a little bit faster. Next, Google announced their new phones, the Pixel 11. And of course, these are built specifically from the ground up for AI. So, you're going to see a lot of AI features throughout the phone. And one that I thought was really, really cool that I haven't seen before is this new American Sign Language to Text. I just love when AI is used for purposes like this and we need more stuff like this in the world. More people using AI for good things like this. Now, some of the other AI features built in this phone, of course, you can

4:42access Gemini from anywhere and it's really intelligent, really predicting what it is that you need. And of course, it works across all of your different apps, so you can access your maps, your Gmail for instance, all of those can be used to help answer your questions. We'll get to more updates in a moment. Right now, I want to talk about the one tool that I consistently use to enhance how I use all these other tools. This part of the video is sponsored by WhisperFlow, a voice-to-text tool that works anywhere I can type on my computer or phone. Hold down one key while I talk and when I let go, WhisperFlow writes

5:17wherever my cursor is. But unlike built-in dictation, WhisperFlow doesn't just describe what I say, it writes what I mean to say. For example, I need a list of AI presentation tools to test. Um ChatGPT, Notebook LM and uh Claude. Actually, not Claude, Claude Design. That took a second, came out as a clean list and was much faster than typing. That matters even more when I'm using WhisperFlow to prompt. I can talk through a much more detailed prompt than I'd ever bother typing, which gives ChatGPT or Claude more context and

5:54usually gets me a better result. WhisperFlow also gets more personal the more you use it. If it gets a name or term wrong, I add it to the dictionary once and WhisperFlow remembers it. And it works the same in every app on my computer and my phone. So on my phone, I can use it inside messages, Gmail, or any AI app without learning a different workflow. WhisperFlow is free to use, but you can use promo code Lipsky under plans and billings on desktop to get 1 month of WhisperFlow Pro for free with unlimited words. All the details are in the description down below, and now back

6:29to the news. The next update is that we have a new Gemini Flash model. This one is 3.7 Flash. When I'm recording this, it's not yet available inside of Gemini, but it is available inside of Google AI Studio and is available uh it powers Google Gemini Spark already. So I have gotten to use it a little bit, and it does feel like a little bit of an improvement over the last model, which was 3.6 Flash. What's interesting here is that we still do not have an update to the Pro model. It is stuck on 3.1 Pro. We were supposed to get an updated Pro model months ago. It never came. And

7:09the rumor is that Google has given up on an update to 3.1 Pro and they're just going to skip to 4 Pro. And look, Flash is a great, very capable model. If you're inside the Google ecosystem, using Google AI in search, inside of Google Drive, in Gmail, you're going to be pretty happy with the Flash model. It works very well for most tasks. But if you look at what all the other major AI labs are doing, like Anthropic and OpenAI, they have been releasing incredible models over the past couple of months and making those models the default inside of Chat GPT. By default, the model that's used is their best

7:52model. And Google has really fallen behind here, and the optics of this aren't great. Now, of course, they are Google, and I have full confidence they have all the resources they need to pull back ahead and to catch up to the competition. We just have to wait for that to happen. In the meantime, though, 3.7 Flash, like I said, is a great model. And they've said that it's stronger for coding, knowledge work, and web development. And looking at the benchmarks, you can see that there are nice improvements to it. So, you're going to notice a difference when you're

8:23switching from 3.6 Flash to 3.7 Flash. And the last update from Google is to Pexels. So, Pexels is a tool I've covered a few times on my channel. It allows you to input your brand's website, and from that it'll generate something called business DNA, and using that, you can automatically generate assets like an entire website, a brand book, or do this virtual photo shoot where you upload one image of your product, and it'll create all these other brand images. So, there's just

8:53been a little bit of an update to that. Now, when you use that product photo shoot feature, you will, just like before, choose the image that you want to then turn into multiple images. So, here, I'll just select this image of this hat, and then you get to choose between these different templates, and your product will be used in these templates. But what's new is that you can now create your own custom style templates. To do that, you'd click right here where it says add your own style, and upload one to three images with a similar aesthetic, and then from that, it'll create a template. But that's all

9:29the news from Google this week. So, let's move on now and talk about everything that SpaceX AI has been up to. First, we need to talk about Grok Bot. This is a new tool that is taking the internet by storm. Everyone can't stop talking about it and for good reason. It is a new way of thinking about AI agents for most people. So, let me quickly show you this app. Over here on the left, you'll notice that I have what look like chats. But unlike using something like ChatGPT, where each of these represent a different thread, instead each of these represent a different bot or agent. Or you can think

10:07of it as an AI co-worker. And each of them have a different role and purpose. So, for instance, this one right here is called Ezra the email executor. And its only job is to draft replies to emails for me in my inbox. That's it. It doesn't do anything else. And it does that on a schedule, of course, for me. Then we have another one down here called Sam the scripter. Sam I use to help me outline my YouTube videos. And that's it. So, anytime I need a script done, I talk to Sam. And it gets better as time goes on at helping me outline my videos. I have another one here to look

10:44through LinkedIn. Another one here that helps me with my travel. So, each one has a specific purpose. But here, I have created a bot called Deborah the delegator. And Deborah's main job is to delegate tasks to the other bots. So, I don't even need to talk to the other bots. I can just jump in here and give the work to Deborah and it will outsource the task to the bot that is best equipped for that job. So, here for instance, if I ask Deborah if we found any good flights for my vacation and if there's any AI news to catch up on, you'll see that it actually messages the

11:20other agents. You can see that right here. It's talking to Atlas, which is my travel bot, and Robbie the researcher, who is my research bot. So, it's not doing it itself. It's pretty neat the way that it works. What most people aren't telling you about Rockbot though, is that it is pretty expensive, at least as of right now. In order to use this, you have to be on a $200 or more a month plan. So, it's not going to be attainable for most people. However, there are strong indications that that will change very soon, and it will be available on the cheaper plans. The other problem though, is that it uses

11:57your credits very quickly. So, they have to figure that out. People who are paying for this and are using it are reporting that they're burning through their credits before the entire week within a day. So, there's definitely a lot of pricing issues that have to be worked out. But once those are worked out, I think this has a ton of potential because it is one of the simplest ways to get started with using agents. If you look at the app, there are very few settings here. You can come into the settings and notice there isn't much here. And even the settings that do

12:30exist are really not necessary. Everything can be done just by chatting with one of your bots, including setting up the bots. If you want to create a new one, just click on plus up here, click on create a new bot, and then it'll ask you a bunch of questions about what you want to use that bot for, and it will set itself up. You'll also see that there are no settings for selecting files and folders beyond a simple attach files. But if you wanted it to have access to folders on your computer, just tell it, "Hey, go into that folder." And it will figure out how to do that. If

13:02you wanted it to have access to third-party plugins, just tell it, "Hey, go into my Notion account, or go into my Gmail." And it will just set itself up. Obviously, you will have to intervene to enter in passwords and stuff, but it just figures stuff out on its own. And all of that can be done just by chatting with the bot. It is a lot easier, I think, for regular people to understand than trying to explain to them what MCPs are or skills are or things like markdown files. You don't have to worry about any of that. I also think that the idea of each bot having its own

13:37different role will be a lot more intuitive for the average person than something like system instructions and skills. I know I've tried to explain to family members those concepts and they don't really understand how they all relate. Like if we have system instructions and project instructions and agent markdown files or how do all those work together? And honestly, it can be confusing even to explain. So, just stripping away all of that and having these simple AI employees, I think will make a lot more sense to a lot more people. While it's definitely behind other tools like ChatGPT work and Claude Co-work in terms of some features, I think it's really the way

14:13you interact with the agents that makes this unique and where I really see the potential if they can just figure out the pricing. All right, a couple other updates from Grok before we move on to something else. First, we have a new model, Grok 4.6. But, I'm not going to talk about it in this video because if you come over to Grok right now, it's not yet available inside of regular chat. It's only available inside of Grok Build and Cursor and different tools like that that are really more for coders and developers. But, what is currently available is their new image generation model. It's called Imagine

14:47Image 2.0. And you can find it, of course, by clicking on the left here where it says imagine. This is where you generate images inside of Grok. And if you have a paid plan, this should already be loaded in with the newest model. I'm not quite sure if it's available for free accounts yet. And I've been testing it out and I'm honestly like a little disappointed in it. It's supposed to be optimized for work purposes, so things like editing images and creating infographics, and I just am not super impressed with it. So, here for instance, I asked it to create an infographic for me that details all

15:23the information about the new Grok bot. And I will say that the text in this infographic is a lot better than what Grok has made for me in the past. However, I don't think it's a particularly nice infographic, not as nice as the ones that are generated by Chat GPT or Gemini. And also, all the information in here is inaccurate. I told it to go out, do research on the new Grok bot, and then to make an infographic on it, and it just made the infographic this None of this information is about Grok bot. It's just about Grok. So, it really struggled

15:53there with doing that two-step process. I then asked it to edit an image for me. So, this was the original image. And for some reason, as I'm recording this video with my screen recording software, I know that this image for you is going to look super blown out and bad, but I promise you the original image that I'm looking at looks much nicer than it's showing up in the recording. I'm not sure what's happening with the screen

16:18recording software there. But anyway, these are the two edits that it made. And again, these edits are probably going to look better to you, but I promise you on my screen, these look worse. Like this original one to me looks very crisp, very clean, good details, the lighting is nice and bright because I'm outside. And these two are just over darkened, and everything looks just a little bit too soft. So, then I

16:41tried something a little bit different. I gave it this image and told it to remove the background and put me in a cozy living room by a fireplace. And this is what it created for me. So, it, of course, accurately changed the background, but I don't think it's very compelling, particularly because of the lighting. Now that lighting is changed, so I would expect, because there's a fireplace back here, for that to reflect on my face, but instead we have the original shadow on this side of the face, and it's clearly lighted from above, which doesn't really work with this image. It doesn't look like I'm in

17:20a cozy living room. It looks like an image of me taken outside that was photoshopped into a living room. And I don't know why there's a dog here. I have no idea why they added a dog to this image. Another thing that's really strange is that I had Grok generate two images for me. This is the first one, and this was the second one. And I have no idea why it generated this image of this woman. So, that feels, I don't know, a little random. Yeah, so all that to say, I don't love the image generation inside of Grok. I much prefer to use ChatGPT or Gemini. And then

17:52finally, if you use Grok live voice or voice mode, there have been a couple of enhancements to that. So, first of all, there are more choices here in terms of the voices that you can choose between. And also, voice mode now has access to all of your existing connectors. So, you could open up voice mode and ask it something like, "What emails do I have in my inbox, and help me draft replies to them?" The other big story from this past week involves watermarks in text that are generated by AI. And this all came to light because Anthropic announced that going forward, all of their models will put an imperceptible

18:34watermark in any text that they generate. So, this is a watermark that you cannot see, but it is there woven throughout the text. It doesn't change the meaning or quality or readability of it, but it is there. And even if you copy the text and paste it somewhere else, it will still exist. Anyone will be able to copy the text and put it through a scanner of sorts, and it will indicate whether it is likely that Claude wrote it or not. The problem is that the watermark can exist not only when text is generated by Claude, but also even if you just ask Claude to translate text for you, summarize text

19:13for you, or proofread text for you. And once again, it is not a visual watermark. It is part of the generation process itself. So, it's basically a pattern in the text that we can't detect, but a model would be able to detect to verify if it was generated by AI. They said that even if you make small changes to the output, the AI watermark could still exist. Obviously, this has raised a lot of concerns. What if you're a company that uses AI in your writing process or proofreading process, and one of your clients checks the text and sees there is a watermark. They may feel like they're getting ripped off,

19:52even if you just used it to proofread. Or what if you're in an industry where the general consensus is that everyone hates AI, they don't want anyone using it, and they don't want to support people that use it, and there's a false positive on your work, or you're a student, and there's a false positive on a paper that you submit. And other companies have said that they will sign up for this as well. So, what if one of your employees just uses AI to help them clean up an email that they've sent, and then the client sees that, and they decide, "Hey, I don't want to work with

20:22a company that uses AI," and they cancel the contract with you, all because of one proofread email that was done by AI. This whole thing sort of has this whole surveillance type vibe to it, where you can't opt into it, it's just done in the background for you, which makes a lot of people feel pretty uncomfortable with

20:41it. Overall, I think this is pretty bad. I actually would be open to the idea of having invisible watermarks for some AI-generated content, like images and videos, but for text, it really doesn't make sense. And the bigger problem is enforceability. If we create watermark rules for companies, that would only be enforced probably on US and European models, and we'd just drive consumers to use Chinese models more. So, I think

21:05ultimately, it wouldn't even work. Fortunately, I'm in an industry where no one cares if I use AI, but I think this could be really harmful for a lot of other people. We also have to talk about this wild story from Australia. Apparently, this guy wanted to get into a class at his local gym, but he didn't do it himself. He just asked his Open

21:27Claw agent to sign him up for a class. What he didn't realize was that the class was already full. And instead of the Open Claw coming back and saying the class was full, the Open Claw actually, on its own, hacked the gym's website, kicked someone else out of the class using a vulnerability in the gym's API, and then put him on the roster for that gym class. And all of that happened without him even knowing it. So, pretty wild. And it sparked this debate, who is responsible when AI agents autonomously go out and do actions like this? And I'm not sure there's a good answer for this,

22:08but by default, I think it has to be the user who set up the Open Claw or whatever agent it is. So, pretty wild. You know, I don't use Open Claw. There's a lot of security vulnerabilities I'm not super comfortable with, and there's so many good consumer tools out there. I don't think you need to make that compromise. You can easily use safer tools like ChatGPT work, Claude co-work, or Grok bot now, and avoid the issues

22:34that come with something like Open Claw. And I can't imagine that that would happen with one of those more consumer tools. But, always just be careful, keep an eye on what your agents are doing. To wrap this up, let's do a few rapid-fire news updates. So, first some news from OpenAI. If you are a college student, good news. It looks like yet again, OpenAI is going to offer ChatGPT Plus to all college students for free for the year. It's not out yet, but it should be coming out in September. So, just keep an eye on that if you are a college student and you want access to ChatGPT

23:09Plus for free. Next, if you use multiple tools for a project, like let's say you're using both the ChatGPT desktop app and the Claude desktop app, there's some good news from OpenAI. If you come into the ChatGPT desktop app and come into the settings and click on import on the left here, there is an auto sync feature. That will automatically keep in sync projects, chats, skills, and plugins, making it easier to work with multiple tools. While we're in the ChatGPT desktop app, let me also show you this. A lot of you probably know this, but if you click on the toggle side panel on the top right here, you

23:47can use a browser directly inside of ChatGPT. And the reason I'm showing you this, this isn't new, but previously OpenAI had a browser, a dedicated browser called Atlas, and that has now officially been sunsetted. And this is the reason why, because the browser inside of the ChatGPT app pretty much replaced that. It is really, really good. In fact, I have a whole video coming out on that very soon, within the next couple of days. So, if you don't want to miss out on that, make sure to subscribe. And the other update is to

24:21the live voice mode inside of ChatGPT. Now, if you come into any of your projects and open it up, you can start live from within that project, which you couldn't do before. In addition to that, if you click on the plus icon while you have live open, you can now attach files, which you couldn't do in the past as well. So, little improvements, the live voice continues to get better and

24:49better inside of ChatGPT. We also have one bit of news from Anthropic. So, as I'm sure you know, there is a Chrome extension for Claude that allows it to take over your browser, answer questions about what's in the browser, things like that. Well, now when you use that, your sessions now carry over to your desktop app, to the web app, and mobile. Previously, if you started any chats from within the Chrome extension, you could only access them in the Chrome extension. But now those

25:19chats will be available everywhere. Honestly, I'm a little surprised this wasn't available before, but glad they finally kind of closed that gap for us. And finally, if any of you are a user of Manis, there's some important information you need to know about. If you're not familiar with Manis, it's a very powerful AI agent. I'm actually going to do a full video on Manis pretty soon because I've been hearing a lot of great things about it and I want to test it out thoroughly for myself. But if you didn't know this already, Manis was supposed to be purchased by Meta, but that deal essentially fell apart. So

25:52because of that, there's sort of some issues with user data. And now during the separation, what you have to do is if you've used Manis or you're using it, you have to back up your data before the end of the month. If you don't, then your data's going to be deleted forever. After that deletion happens, you can then use your backup to restore your account. So if you're a user of Manis or if you've ever used it and want to save your data, make sure to log in and click up here where it says backup now. And that's it. That's all the AI news from the past week that I think regular

26:26people will actually care about. If you enjoyed this update, make sure to subscribe to the channel. I release a video like this every single week. And if you want to make sure you stay on top of all the latest AI news, you definitely want to make sure that you are subscribed. Otherwise, thanks so much for watching and I'll see you next

26:40week for the next weekly AI news recap. Bye for now.