Big AI News You Probably Missed This Week

This week started off as a really slow week in AI. But as the week progressed, more and more news started to come out that got more and more exciting. And instead of breaking it down in chronological order as the news happened, I want to start off with the good stuff. On Thursday, midjourney announced version 5.2. It got quite a few updates, including new esthetics, more variation in the generations. You can use a new shorten command to find out which words are not really impacting your prompt at all. But the most interesting new feature in my opinion of the new Midjourney is the zoom outfit feature. So if we jump over to Mid journey here, I generated a handful of images of wolves in the snow here and here's how you'd use the zoom out feature. I really like this image number three here. Let's go ahead and upscale number three. And as soon as I upscale it, you can see some new options. Here you have very strong, very subtle zoom out 2x, zoom out 1.5x custom zoom or make square. So let's Click zoom out 2X and see what happens. You can see it took that original image and actually made a zoomed out image with the wolf even further back. And. And you can continue to zoom out. Let's say I like number three again, let's upscale number three. Let's zoom out two more times. You can see once again it further zoomed out on the image. If I take one of these images and click make square, you can see it turned them into squares and filled in the top and the bottom. This is Midjourney's current answer to Generative Fill in Photoshop and Uncrop from Stability AI. Let's also quickly test the shorten feature. Let's take one of these longer prompts that I've used in the past. Get these really detailed images. Let's copy this and let's use the new shorten prompt and paste this whole prompt in and you can see that it tells me that I don't need to use the word features, I don't need to use the word lens, accent, lighting or global. If I cut all of those out, it shouldn't really impact the result of the image. So not only would this new update help you zoom further and further out on any image you want, which will be really helpful when you get those images with like cropped shoulders or cropped head. It also helps turn you into a better prompter. Now this midjourney version 5.2 news drops on the exact same day that stability AI announced their stable diffusion XL 0.9. Now a lot of people were speculating that the full version of Stable Diffusion XL was going to be launched today. They gave us 0.9, basically telling us that they don't feel like the full SDXL 1.0 is fully ready yet. But it does claim to produce massively improved image and composition detail over its predecessor. Here's some examples they give on the website of the beta version of SDXL versus the new 0.9 version of SDXL. You could see a little bit of improvement in realism and quality. Here's a hand holding up a coffee cup. SDXL 0.9 has one of the largest parameter counts of any open source image model. Now they say that the API and the ability to run it from Dream Studio will be available to access on Monday the 26th. You can actually play with it right now inside of Clipdrop which is one of stability AI's free front end user interfaces. So if we look at Clip Drop here, we can actually test the new version right now. You can find it over at Clipdrop Co Stable Diffusion. Let's do the ultimate test. A woman holding out her hands. It seems like a lot of people are probably using it right now, so it's loading kind of slow. But the first image that it generated less than desirable results. Here are some other examples that were presumably generated with the newest version of Stable Diffusion and for the most part they actually look pretty good. In fact, if I scroll down here, you can see see an image of silhouettes of hands that actually have the proper amount of fingers and don't look too deformed. This car is looking really good. The people are looking really good. This is getting much closer to the types of images that we would expect from like a mid journey. Now Stable Diffusion is closing that gap and getting better and better with these types of images. Now unfortunately, since I'm recording this the day that the newest version of Stable Diffusion actually dropped, the site seems a little overwhelmed right now and I'm struggling to generate more images myself. But I can get it to generate generate a few decent images from time to time. Like this pretty realistic wolf that I just generated. It's getting much, much better. Now. This was actually announced last week, but it was announced after I recorded my last video. So let's talk about it now. Meta introduced Voicebox. They claim it's the most versatile AI for speech generation. And to me it sounds very similar to what 11 labs can do right now where with just two seconds of audio it could generate very realistic Text to Speech here's some examples from Meta's website.

This is Voicebox.

Now, I'm unclear if the Mark Zuckerberg voice that you're hearing in this video was actually generated with this tool or if this is actually Mark Zuckerberg talking. If I had to guess, they probably generated this audio with their text to speech system because that would be very well Meta, if you give it text.

It can read it in a bunch of different styles.

Penelope porcupine and Sammy Sloth Penelope porcupine.

And Sammy Sloth they also say you.

Can use this tool to fix background noise because it knows your voice. It could sort of edit out background noise and then replace your voice with the fixed version of the text to speech Voice.

Sammy and Penelope's heartwarming friendship inspires joy. Sammy and Penelope's heartwarming friendship inspires joy.

So it can do in context text to speech synthesis using an audio sample as short as 2 seconds long speech editing and noise reduction cross lingual style transfer. So you can actually write something out in a different language and it will use your voice but in that language. And it has diverse speech sampling so that you get more variation in the way the voices sound. Now you can learn a lot more about this on Meta's website. I'll make sure I link up the article below this video. But as far as us getting access to it, they claim they're sort of holding it back for ethical reasons. I don't totally understand that because there's already tools out there that do this, like 11 labs, like Uber Duck, like Descript. So I'm not quite understanding why Meta is holding it back. I imagine we'll probably see this as an open source model at stake some point, but as to when, that's really anybody's guess. Now, in news that I find quite exciting, this week Dropbox announced that they're adding more AI features into their platform.

Dropbox AI is a new feature that lets you instantly summarize and ask questions about your Dropbox files. With Dropbox AI now you can pull up a file, you can ask it anything. Dropbox will read the document for you and give you an answer. And in a single click you can easily get a summary of the entire 100 page doc, saving you all the time and extra effort that it would otherwise have taken to do that manual. We're starting with individual files, but this is just the first step.

Now this is interesting to me because there are so many tools out there right now that that is their main selling proposition is you upload A PDF and we'll summarize it for you, or you can ask questions of it. Well, now, if you have a Dropbox account, you're just going to be able to do that. I also believe this is something that Google is planning on adding into Project Tailwind as well. Are features like this from larger incumbent companies just sort of obsoleting a whole bunch of smaller companies? Now, because of this feature, possibly over.

Time, you'll be able to ask a question about your Dropbox folders and even your entire Dropbox. We all need a search box for our private information, just like Google gave us a search box for public information, not only with your Dropbox files, but with everything. That's why today we're introducing Dropbox Dash. Dash is an AI powered universal search engine that connects all of your tools, apps and content in a single search bar. Lets you easily find whatever you're looking for. For no matter what app it's in, who sent it to you, what you named it, how long it's been since you last looked at it, you can find it super fast. Dash connects across your favorite tools like Google Workspace, HubSpot, Asana and Notion, so that you can access all of your content, no matter where it lives or what format it's in. And because Dash is powered by machine learning, it learns and evolves with you and it gets better the more you use it.

So this is the part that was the most exciting to me because imagine just taking all of your files, your documents, your videos, your audios, pretty much any file type you can imagine. Plus you have all of your apps that you use. Maybe you use Notion, maybe you use HubSpot, maybe you use Google Drive and Google Sheets and Google Docs or Asana or any tool like that. It sounds like Dropbox is going to tie this all together into sort of a personalized search engine where you can search all of your files and all of the various tools that you use across various platforms. And like he said, Google gives us the search engine to search the web. It sounds like Dropbox is trying to give us that search engine to search all of our own personal data. Now, I'm not going to show the whole video here, but they do say at the end of the video that they promise not to share your data and that they're going to keep it private and that you can trust them. Although they want you to connect every single tool at every app and every piece of your life to Dropbox, they promise that they're going to keep it safe for you so we'll see how that plays out. Also this week it was announced that YouTube is getting AI powered dubbing, which is absolutely amazing to me because I am a creator. I love making YouTube videos. Right now, the options to watch my videos are either to watch them in English or to turn on subtitles in whatever language is your native language. But pretty soon YouTube is just going to automatically overdub into whatever language you want. So the words that I'm speaking right now will be coming out overdubbed in whatever language you're listening to in right now. YouTube made this announcement on Thursday at Vidcon that they're working with the team from a company called Aloud, which is an AI powered dubbing service from Google's incubator. So basically the tool first transcribes your video. You can then review the transcription and edit it. It then translates and produces the dub. Dubbing with Aloud is as simple as editing text.

So if you're a content creator, this is huge news and I'm personally really, really excited about this. It's going to open up all new parts of the world of people that can learn about the types of news and stuff that I'm talking about in so really, really exciting times to be a content creator. So that's some of the cool tech that was announced or rolled out this week. Now let's talk about some of the weirder aspects of AI that are happening right now. This week, the Recording Academy, the organization that runs the Grammys, announced their rules about AI generated music. They said that songs that include elements generated by AI can still be nominated, but there must be proof that a real person meaningfully contributed to the song too. If there's an AI voice singing the song or AI instrumentation, we'll consider it, but in a songwriting based category, it has to been written mostly by a human. They added that AI will unequivocally shape the future of the music industry, and instead of downplaying its significance, the Grammy Award should confront questions related to AI. Ahead on. Also, I'm not sure if you've seen the first episode of Black Mirror yet. I'm going to do my best not to share any spoilers about it, but if you don't want to know any information about it whatsoever, maybe skip to the next chapter real quick. The first episode of the newest season of Black Mirror is called Joan Is Awful and it involves AI and deep faking celebrity likenesses. Let's just put it that way. What's interesting is the same week that this episode was released, I came across this news article in the Wall Street Journal, Celebrities use AI to take control of their own images. It says celebrities are attempting to co opt this AI craze. They are making deals with brands to put AI created duplicates of themselves into marketing campaigns, giving them more control over their own likeness and more latitude in the types of deals they can make. Celebrities get paid, but they don't have to turn up. Stars just need to spend a few minutes in a studio with a 3D scanner, which could then create representations of them for countless hours of content. Golf legend Jack Nicklaus recently agreed with AI company Soul Machines to create an AI powered version of himself at the peak of his career when he was 38 years old. The company Metaphysic has actually signed on to provide AI services for a forthcoming Robert Zemeckis film starring Tom Hanks. Now I'll post a link to this Wall Street Journal article below this video, but I highly recommend reading it because if you've seen that episode of Black Mirror called Joan Is Awful, this is literally the news story that's showing that that plot is literally coming true in real life. Now there's other aspects to that episode that I don't feel will really likely play out in real life, but again, I want to avoid spoilers for anybody who hasn't seen it yet. Definitely go watch that episode if you're a fan of AI like I am now. Speaking of the entertainment industry, this week Marvel was crit for using AI to make the Secret Invasion opening credits. Secret Invasion is a brand new Disney plus series. I haven't watched it myself yet, so I literally can't spoil it for you. What I can tell you is that their new intro, which you could see on the screen right now, was actually generated with AI tools. I don't know exactly which AI tools they used. I'm guessing it's some form of stable diffusion or Deforum or warp fusion or one of those tools that are out there. This was done with a combination of visual filter effects artists and AI over the top of those visual effects. Now supposedly this was intentionally designed to look unnerving. They wanted to give it that AI effect and quite honestly, it's created great publicity around the show because a lot of people are talking about the fact that this intro was AI generated and making more people aware that there is a brand new Marvel show over on Disney plus right now. So the marketing behind making this an AI generated video in my opinion was actually genius. And the creators of the intro wrote an exclusive to the Hollywood Reporter that the opening that used AI cost no artist jobs. Method Studios clarifies reports that sparked a social media backlash, stating AI tools complemented and assisted our creative teams. If I was the marketing person behind Secret Invasion, I probably would have personally been the one to leak the fact that we used AI to make the intro to stir up controversy, only to follow up with a report later that nobody lost their jobs and AI complemented what we created with visual effects because all of that controversy leads to views now. Speaking of controversy, this week it was reported that ChatGPT user account credentials were found on the dark web. ChatGPT users should be wary that their personal data might have been leaked online following the dump of more than 100,000 ChatGPT account credentials on the dark web. However, OpenAI did respond to this, saying that it was more like 26,802. The leaked accounts mostly came from the Asia Pacific region, and then it was actually due to malware that was installed on a ton of people's devices and not an actual breach of something at OpenAI. So OpenAI wasn't hacked, but somehow a lot of people installed something that essentially phished their data. So if you haven't been installing any weird shady software lately, or downloading anything that you don't know where it came from, you're probably in the clear. But it's always a good idea to change up your password from time to time. I wanna circle back around to something that I talked about earlier this week from this company, Illumina. I actually dubbed in additional footage from this company because I thought what they were making was so cool, and I wanted to show you another example of it just real quick because this stuff is just nerd heaven for me. Here's an image that was generated in Mid Journey. Illumina then turned that image into this 3D world that we're seeing on the screen where you can move around and zoom in and zoom out of it. And then they were able to drop a character into this 3D scene and actually make the character wander around inside of this 3D world. This was originally a Mid Journey image that we're looking at that was turned into the 3D world that a character could now run around in. Like this, to me, is just so insanely cool that we're able to do stuff like this. And as of right now, this Instaverse tool from Illuminai is actually free to play around with. So I actually demoed this a little bit more in my video from earlier this week. So if you haven't watched that one, check it out, the thumbnail of it looks like this. All right, finally, to wrap up this week's news, let's talk about the most bizarre thing that's going on on the planet right now, where Mark Zuckerberg is ready to fight Elon Musk in a cage match. What? So that's right. Meta is actually planning on building a competitor to Twitter. Elon Musk has trolled him a little bit on Twitter about this, and it's all sort of escalated into Elon Musk tweeting that he's up for a cage match with Mark Zuckerberg. It further escalated because Mark Zuckerberg screenshotted the tweet saying, I'm up for a cage match, and then said, send me the location. Now, on first view, you're probably thinking this is all just like trolling and goofing off on Twitter. Well, it kind of isn't, because Mark Zuckerberg actually trains in Jiu jitsu and claims to train a little bit in MMA as well. And according to Dana White from the ufc, they are absolutely dead serious about making this happen. Now, this is a report from tmz, so take that for what it's worth. But in an interview with TMZ on Thursday, White said he probably spent an hour and a half on the phone with both of the world's most high profile tech billionaires and found out that neither of them was kidding about a proposed cage match stemming from a social media spat. They both said, yeah, we'll do it. They both want to do it. He said, mark Zuckerberg hit me up first and said, is Elon serious? And I said, I don't know. Let me ask him. So I asked him and he said, yeah, I'm dead serious. There's actually a non 0% chance we might see an actual cage match between Elon Musk and Mark Zuckerberg. Now that's gonna be interesting. Anyway, that's all I got for you in the world of AI. When the week started, I thought, man, this is gonna be a slow week. There's really nothing to talk about for Friday's news video. But by the time Thursday rolled around, it really ramped up. And I don't even know what happened on Friday yet. Cause I'm recording this on Thursday, so there could be new news that I haven't even reported on yet. But if there is, I'll report it in next Friday's video. Also, since I'm doing these Friday news videos on a regular basis now, I'm trying to come up with a clever name for what I should call them. I've been about just going along with the name of the AI news you probably haven't heard of yet because that seems to do well with the YouTube algorithm. But if you have ideas, let me know in the comments. I'd love to hear them. Hopefully you enjoyed this video. If you did, maybe give it a thumbs up. And if you're not subscribed to this channel already, subscribe. I'll keep you in the loop with all the latest AI news, all the coolest AI breakthroughs, a handful of tutorials from time to time, and give you the grand overview on the tldr. So if all of this is overwhelming and you don't want to keep up with it yourself, that's what I'm making this YouTube channel for. Also, if you do want to go deeper down the rabbit hole, check out FeatureTools IO. I actually keep this website updated with all the AI news on a daily basis. I also curate all the cool AI tools that I come across and I send out a weekly newsletter every Friday with just the TLDR of everything that's happened in the AI world for the week. You can find it over at FutureTools IO. If you go join now, I'll make sure you're hooked up with the next week's newsletter. Thanks again for nerding out with me. I really, really appreciate you. See you guys in the next video. Bye bye.