Major AI News #27 - Sam Altmans AGI Shocker, Apples ChatGPT Update, Gemini Pro's New Features...

Another incredible week in artificial intelligence. Let me tell you about all the things that you most likely missed due to some crazy headlines. So one of the first things that we actually did get was we actually did get some really cool updates to Midjourney. There were actually two pretty nice updates. So one of the updates was the fact that there is actually a website for Midjourney, but it is currently in alpha. If you don't know what that means, that essentially means that it's just in its testing phase. Because when something is released, a product or a game, usually what they have is an alpha version and then a beta version. You can see, see that Mid Journey is an alpha version, which means it's then going to go out to beta and then of course it's then going to go into a full release. But this website did get a few updates. Now, if you are someone that is wondering, how on earth do I get access to the Mid Journey Alpha website? Well, I'm pretty sure that to my knowledge you're only really allowed to access advanced Mid Journey stuff if you've generated over 10,000 images. So if you want to access this, do get creating. But let's take a look at this video here because this is by someone called Nick St. Pierre. And this guy tweets out a lot of stuff about Mid Journey. And this is definitely someone you should follow if you want to stay up to date with all these stuff on Midjourney.

The alpha over here, just some little ones, but they're pretty useful. So when you hover over an image, you can just run a variation right here by clicking it and then they also switch some of the buttons. So this is in a v5.2 run, so you can see more of the buttons than are available on V6, but they organize them a little bit nicer and left everything right there so you can click it right away. And then they also added this heart button so you can easily like images, which is huge because before you just had to rate them. Now it's much easier to to like and then go find the ones that you are looking for. So little cleanup.

And yeah, I think this is really cool because although many people do use Midjourney, its current fate in its current state, they do actually make, I think around $20 million profit a month. I think that some people also don't actually know how to use Discord. So once this website is finally actually live and once they finally roll this out, this is going to be a huge use because I think this is going to make Midjourney More commercialized in terms of the fact that it's going to be rolling out to more users, especially if they do manage to create an app with this, because the Discord user interface, it just isn't that great, really. It just isn't good at all. But it just goes to show that when you have a really good product, it doesn't really matter how you access it, as long as the base product itself is good. Now, of course, another small update to Midjourney was that they also did release Ninji 6. So essentially it's meant for anime style images, but there is essentially a really cool aesthetic. This entire thread goes into detail on the subtle differences, but for those of you who are advanced Mid Journey users, I definitely thought that this is something that you would want to note. Now, there was an interesting video that was tweeted out by this page and essentially I know that a lot of people did miss this because when it was released I did miss this too. Now, it isn't essentially an update, but essentially there was this really fascinating video. And essentially we had a bunch of people walk into a room by this, you know, page called Deep Stakes. And essentially what they did was they showed how AI could be used in a really negative way. Now, many of you may have seen this, you know, heard about this before, but essentially the premise of this entire video was that they had a video with 100 real vegans. They essentially asked them, you know, their opinions about veganism and all that kind of stuff. Then essentially what they did is they fed them a 100%, you know, vegan sandwich. And then essentially they used AI to create deepfakes of them saying that they hated vegetables and that they loved me. And essentially the reactions were pretty shocked because they didn't realize that this technology existed and they didn't realize that this could be done. I mean, some of the reactions were really, really telling because, you know, the average person, I don't think they realize how bad the technology has gotten. I mean, recently, this week, there was some crazy, crazy stuff that now means I'm pretty sure that they're looking at, you know, some kind of new legislations on what can be created. Because not only did this video show that the wider public doesn't really know about this, it goes to show that they don't know also how accessible this really is. I mean, if you take a look at their reactions, it's pretty crazy. That's not what I said.

Think it's f ed up? I mean, I was confused because I thought you were a Vegan?

Yeah, I am.

But that's not what you said on the video.

Well, that's not the truth.

We'll agree to disagree. That's identity theft, point blank.

Those words did not come out of my mouth.

Everyone just watched this, right?

Yeah.

Yeah.

Did you hear him say that?

Yeah.

Sounds like you love me.

It's not me. I told you that. How many times do I have to repeat? I'm telling you that is not my voice.

Are you sure about that?

Positive. So I know that that video might just seem like, ah, you know, that's old news. Deepfakes have been around for quite some time. But I think it's important to realize that if you are watching this video, then you are part of the population that is uniquely focused on artificial intelligence. Because as popular as some things might be in the community, they really aren't as mainstream as you think. Although Chat GPT might seem really popular. Unless you're someone that's actively looking in the AI space and really needs to use these websites and these tools and you're really looking into it, you're really not going to be paying attention to this stuff. And I think it's important to really realize the dangers of some of these technologies so that you can understand and protect yourself. Because largely what we're seeing is that these technologies are increasingly used to do things which might not be of our best interest. Then of course, something that is quite similar to this story was the fact that NBC was reporting that voters were getting robocalls with deepfakes of Biden's voice telling them not to vote tomorrow. And essentially this was actually confirmed to be used by ElevenLabs. So essentially it says a robocall obtained by NBC News appears to use a digital manipulation of Joe Biden's voice to, to encourage New Hampshire primary voters to not support the president. And essentially the voice is just where he's just like, you know, don't fall in tomorrow, don't vote in tomorrow, yada, yada yada. And it is pretty crazy because essentially after this investigation was done, they realized that the person who did this and who made the calls that they were actually using 11 Labs. If you aren't aware, 11 Labs is the state of the art, very best voice software where you could literally take a couple seconds of someone's voice and then you can immediately replicate that voice to have it sound as realistic as possible. Now this is something that could be used against you. So I would always say that you do need to be pretty careful with this kind of stuff. And Although it is pretty hard to protect against copyright against your voice, there's no real legislations around it. There's no really laws around it. There's no like copyright protections around it. It is something that is pretty hard to do. But it's important to know that this kind of propaganda should, could be affecting certain economies, could be affecting certain political states. I mean, it's something that is very, very dangerous because messing with democracy is not something you want to have. And it's just important to note that this is definitely something that's going on. So maybe if you get a call, you have to second guess what you're hearing. Because even if it sounds like a human, you have to realize that it might just not be human. And like we've seen previously, I don't think this is going to be the first time or the last time that we do get some things like this. This was definitely something that we did predict in the video on things we were going to 2024 and we, we kind of knew this was going to happen anyways because technology is getting better. There's many different things. And they did actually shut down the person's account. They actually managed to track who was doing this on 11 labs. Not sure what kind of internal team that they had. But I think it's important that they do take as quick action as possible because although there are now unfortunately open source tools which pretty much anyone could use, I think we do need to be aware of this because it's important to understand that these things do exist. And then of course we actually did have an update for 11 labs which is Dublin Studio. So I did want to just bring this out. And although you might be thinking why are you talking about 11 labs and use crazy software? It's actually pretty good in terms of what it's capable to do. So I know that, you know, some people are going to use this technology for bad, but those out there who are creatives and who just want to use the software for maybe creating content, educational content, informative content, or for whatever creative purpose endeavor that you do have, I think it's still pretty cool. So take a look at their dubbing studio. So it gives you precise control when translating your videos for global audiences. So they detect and label each of your speakers and create an editable script of your content. And just take a look here at this video.

11 Labs Dubbing Studio is here now. You can easily translate your videos for new audiences around the world. Just upload or paste a link, choose a language and click create it's.

One small step for man.

Just.

Instantly dubbing your content into multiple languages for a global reach.

Who are you? I'm popped by the sail. The man.

The dubbing studio automatically detects multiple speakers, giving you precise control. Rename the characters, edit and regenerate the dialogue until the accent and emotion is just right. And when you're finished, you can add even more languages to expand your reach in a land where the sun scorches the earth.

You ain't from around here, are you, boy?

11 Labs dubbing studio bringing your stories to every corner of the world.

So yeah, I think eleven Labs Dubbing Studio is pretty fascinating. You can see the various applications that this does have. I mean, if you were someone that paid attention to previously the talk at Davos, there was a recent talk that wasn't converted with 11 labs, but it was actually converted with a tool called Heygen. And it proved to be very effective at getting people engaged because the actual video had around, I think it was 20 million views and people were stating that I can't believe this technology is here. This made the talk so much more engaging and it allows for that cross border translation that previously we wouldn't have because with traditional translators, essentially you're hearing someone else's voice, someone else's tone, you're hearing someone else's inflections in their speech. But if you can actually get essentially the nuance of how they're speaking, the entire context and in their own original voice, it's much better than having a translator do it because number one, you're going to see the video in real time with their lips moving and you're going to be able to hear them in their voice. So I think this technology, whilst can be used for bad things, I think that updates like this are definitely going to be pretty fascinating because they allow us to of course enjoy other content and allow us to share our content in other different languages. So I think what this is important to look at, the wider view here is that in the future, I think a lot of content that we do think is just for one country, we're definitely going to be able to explore other cultures and enjoy them a bit more. Now in terms of AI creativity, there was also this pretty cool text music on Beathoven AI. And this is good because as you do know, music is something that is quite often copyrighted. There's a lot of issues surrounding royalties and trying to get the right soundtrack. But music is something that is evolving very rapidly due to AI. So this is the video that that.

They Made makes your content shine. So write introducing text to music on Beaton AI. Whether you're a filmmaker looking for a background score that fits your scene, a podcaster looking for the perfect intro Music, or a YouTuber looking for royalty free music for your content, text to music gives you true creative control to craft the perfect background music for your content. Find the tune that carries your story with Beaton AI.

Now, there are many different websites out there that are like this, and this is not the only one. But I do find it, you know, really fascinating to see how all of these companies are starting to market their music stuff. Because I think text to music, once it is completely mastered, I definitely think that that is one of the industries that's going to be changed. Because if you don't know how music works, a real problem that people have, you know, on YouTube is, you know, they need Mac background music for stuff. And the problem is that, let's say, for example, I go ahead and I buy a music track and then I use it in a video. Sometimes even if you've bought the license over time, what happens is the license right holders will change and then essentially, even though you've paid for a music track, the license changes and someone else that now owns that music track can come and claim your video. And all that hard work you've put into the video that you are earning, it's now there. So you have to then remove the video or remove the music track and it's just not a great thing. But if you have generative music, you can essentially own the rights to that and you can use that for yourself. And I think the music also, you know, in terms of like calm and relaxing music and stuff that doesn't really make that much of an impact is definitely going to be there for a lot of creatives who don't either want to, you know, pay ridiculous fees for every single time they use a music track or just want to have something that they know 100 isn't going to be copyrighted. Now moving on to. Now moving on to robotics, there was something that I saw that was really, really intriguing because it was only a couple of weeks ago that we did get a mobile aloha. I think it was around three weeks ago, but we did actually get this. Okay, so it says exc to release ok, robot, an open vocabulary mobile manipulator for homes. Simply tell the robot what to pick up, where to drop it in natural language and it will do it like me, okay, move the takis from the desk to the nightstand and Then you can see what this robot is able to do. So you know, you give it a task and then you can see exactly how the robot breaks down the task. So it has this voxel map of the entire room so it knows exactly where the desk is and then it can move around. You can see that this is definitely five times speed. So it isn't moving this quickly, but it goes over to the desk. You can see that it manages to find the tarkeys and then it decides to position itself in the right area. You can see this is the robot's pov. This is exactly what it's seeing. And then of course it uses this technology, any grasp. Then it uses Langsa mask, which is, I'm pretty sure it's an edition of Meta's segment Anything now. So essentially it's a technology basically you can just pretty much look at every single thing in an image and select where it is. And then combining those two, it does a proposed grasp and then you can see that it then goes ahead and gets the grasper tool and then it goes ahead and grasps it. So I think, you know, even watching the development and all of these technologies before and even, you know, reporting on them, it's just really cool to see how a lot of these open source tools are being developed for, you know, applications like this, where of course you don't need a tool like this. I mean you might not need to, you know, move the Takis and place them on a desk. But I think it just goes to show where we're moving because remember, this is actually autonomous. This is not tele operated. This is not something that, you know, you've got a human behind it, doing it. This is 100% and this was, I'm not sure if they're using an LLM, but you can see, okay, robot move the tar keys on the desk and nightstand and then boom, it manages to do it. Now there are some other interesting things that are about this. So it says given a one minute scan of a home we can obtain rich text grounded semantic information on objects using lvit seen for the video I showed above here is pretty much spot on for the largest objects in the scene. And of course you can definitely see that there are a lot of different applications here and a lot of different tasks that they are doing. So I think it goes to show that, you know, whilst some people were saying that mobile aloa was just, I guess, teleoperation in some aspects, it goes to show that, you know, autonomous movement is still possible and these combination of things goes to show us how crazy this is. And I will do like a full video on this because there is a huge video on robotics that I do want to make. So that video will definitely be coming soon, but these advancements are definitely coming. And I think the autonomous aspect is definitely something that people have been looking from these companies. Then, of course, we had this clip. Now this is something that I did want to cover before, but just didn't get around to it because there were just several videos that I did want to make. Now let's take a look at this. Okay, so Sam Altman basically talked about how when we have AGI, the world will have a two week freakout, then people go on with their lives. Now, I don't agree. I respectfully disagree about the statement because if AGI is even, you know, half of what people say it is, I think it's going to change everything. And I think two things are happening here. Either one, you know, AGI is probably not as crazy as we think, and I don't think that's true, or either. Two, he's just downplaying it so people don't freak out about the next release, which is much more likely. But basically, essentially in this clip he's talking about how, you know, most people always freak out about the next update and, and even basically. Right, I need to show you guys this. Back essentially when GPT2 was around, and we all know that GPT2 was a very, very basic LLM, people were saying that, you know, AI was deemed too dangerous to be released into the world. And researchers fear that the model known as GPT2 was so powerful that it could be maliciously misused from everyone from politicians to scammers. And it was created for a simple purpose, yada, yada, yada. And people are saying that basically at the time, for any new technology, we are always afraid of what is going to come. But you know, they're referencing back to GPT2 and they're basically saying that every time the new technology comes out, people are scared. And then after everyone goes about their daily lives. But I do think that as we get more and more advanced in terms of what this technology is capable of doing, I think that statement is ringing less and less true. Because this piece of software, this is probably just clickbait by this website and they probably just wanted to, you know, get some more clicks, just, you know, fear people like, you know, this AI deemed too dangerous to release, makes it out into the world, oh no, yada, yada, yada. I think that now with these kind of statements, I'm not sure if it holds the same weight because if AGI is really true and essentially, you know, it has so many different applications, I think that it definitely will change the world and people will start to feel the impacts. But anyways, here's the clip for those of you who want to see.

The world had like a two week freakout with GPT4, right? This changes everything. AGI is coming tomorrow, there are no jobs by the end of the year. And now people are like, why is it so slow? I believe that someday we will make something that qualifies as an AGI by whatever fuzzy definition you want. The world will have a two week freakout and then people will go on with their lives. The world had like a two week.

So take from that statement what you will. I mean to take from that statement. So take from that statement so take from that statement what you will. I think that if AGI is really good and they manage to get it to be agentic and really good in terms of what it's able to do, like a true AGI definition, like it's able to actually be an AGI, I think that it will change the world. I think that if it is rapidly deployed and you know, there's open source ones, I definitely feel like there is going to be some kind of societal unrest, like things we talked about previously. Because the next wave, that's really when the next wave starts at AGI. Like, you know, we've seen the now wave with LLMs and you know how that's impacting certain sectors. But I think with AGI that's going to be a huge another wave that's going to be impacted. And the real question is if it's just going to be OpenAI's proprietary software or if it's going to be open source by meta, or if Google is going to get to it first. And I think that is one of the main questions that I am not sure about yet. I mean, if I had to place a bet, I would bet that OpenAI does do it because they're so far ahead and they have so much talent at their company and they have really high salaries. Really everyone's pretty much committed there. But if not OpenAI, I definitely would think that Google is going to get it done. Now mentioning Google, okay, there was something really interesting by Google. So Google's bot actually made a stunning leap surpassing GPT4 to the second spot on the leaderboard. And big cracks Google for this remarkable achievement. Now, if you don't know essentially what this is here. This is essentially a leaderboard that is basically on human evaluation. Now, not the human evaluation that is on the tests or the LLM benchmarks, but the kind of human evaluations where people just simply rate their conversations with the AI in essentially a masked way. So people have no idea which is which they're speaking to an AI. And at the end of the conversation you rate which conversation you had was better. And then by the thumbs up, people just decide that overall, Bard is nearly as good as GPT4 Turbo. Now, an important distinction to make here is that Bard is running currently on Gemini Pro, which does mean that Gemini Ultra, just purely on, based on what we're seeing here, looks like it is likely to surpass GPT4 Turbo. Now, some people are going to be like, wait, why on earth are you comparing Gemini Pro to GPT4 Turbo on this website? Arena ELO benchmark. Like, why is this benchmark even relevant? Because I think that whilst other benchmarks are really good, there are two main things that people forget when looking at the other benchmarks and that is the fact that, number one, these benchmarks like the MMLU do actually have issues in them. Like for example, there was a video that showed that the MMLU has questions which have real errors which could lead to some percentages being artificially inflated due to the nature of how these questions were written. Number two is the fact that of course some things when they are trained, they are trained on actual test data. So when you're seeing these LLMs come out and you're like, oh my God, it was able to get this on this test. It's because it already saw that the test, in the test and the training data and it was able to remember the answer to that question and that's how it was able to get a high score. And this is essentially called, essentially just called data leakage. And it says common problems with evaluation and machine learning and data leakage from the latest data test sets was into the training set. And this is specifically true for LLMs where the details of the data set are often missing, sometimes proprietary. And that's pretty much an issue because if you've already taken a test, you're going to do well in it. So that is something that does happen with these tests. So what we do need to have is new tests that are zero shot that they've never seen before and those are going to be the pretty much good benchmarks. And I think that this is definitely a good benchmark to look at because at the end of the day, we are training these systems for use on particularly one thing. I mean, whilst GPT4 does exceed in medical areas and stuff like that, I think this is definitely a win for Google because Gemini Pro was definitely a lot of people saying that Gemini Pro is awful and now all of a sudden everyone loves it. And like I said before, Gemini Pro was slowly being rolled out and many people were actually thinking that Gemini Pro was Palm 2. And something that you don't know if you actually use Bard when you actually message it. Sometimes what you'll see here is you'll see a palm 2 logo. So understand that it always isn't Gemini 2 and I don't know why they don't make that distinction. Maybe because they just don't want people to know. But sometimes on the right here, you'll see that it's Palm 2 and sometimes you won't see anything. And I'm not sure if that means it's Gemini or not, but just understand that you don't really know. And in the UK we do get these models later. Now, continuing on with Bard, there was some more information. So essentially there was also this update which is going to be coming, I think in the next two days. So essentially it says here, create images with Bard. Essentially what you can do is you can create captivating images for work, play and anything in between. With Bard, it's easy to use. Simply enter a few words to bring your imagination to life and start with English prompts, click Generate more for more options and download the ones you like. So essentially we're going to be getting an image generation from Bard and of course Bard with Gemini Pro is going to be coming with more languages and of course Double Check is coming to more languages. So this is an update for Google Bard and it's Gemini Pro. Now, like I said before in the UK and in the eu, I'm not sure when they're rolling this out, but if you're likely in the us, this is going to be something that's going to be hitting you guys on The, I think the 1st of February, so you guys can expect that too. Now, one thing that I can't wait for is Gemini Ultra, because that is going to be absolutely insane. And of course I really do want to see what this image generation software does look like because Google has a really, really big image generation. Like they've got tons of stuff. They recently released Lumiere, it was pretty crazy. So it'll be interesting to see where all of this technology goes for Google and I Think it's going to be really fascinating to see how people's interactions go and the comparisons to ChatGPT's Dall E3. Now, something that I'm not surprised is finally happening is that AI companies will now start to now something that I'm not surprised is now finally happening is that air companies will need to start reporting their safety tests to the US government. So it says that the Biden administration will start implementing a new requirement for the developers of major AI systems to disclose their safety test results to the government. So the White House AI Council is scheduled to meet Monday to review progress made on the executive order that President Joe Biden signed three months ago to manage the fast evolving technology. Essentially, it says that the government's National Institute of Standards and Technology will develop a uniform framework for assessing safety as a part of the Biden executive order signed in October. And I think this is kind of important because right now we do have a situation where people just get AI models and they just published and they're in the public. But after some research, it's actually incredible because in China you have to go through a really rigorous phase in order to get your AI approved for public use. And I think that it's kind of good that the US is adopting this strategy because whilst, yes, you don't want to slow down things due to regulation, I still think that due to the AI race, it's important for there to be at least a little bit of oversight just to ensure that these systems don't do something pretty crazy. Because the next level guys like LLMs are one thing, but these agentic systems that are coming out are going to really, really hold a huge potential for good and a huge potential for bad. So I think the next level in terms of agentic systems is really going to absolutely change everything. Then of course, we have Apple testing Souped up Siri with help from ChatGPT. So it says when iOS 18 rolls out with the iPhone 16 this September, it's widely expected to be Apple's first foray into the game changing world of generative AI. So the iOS 17.4 beta arrived and a little digging. So people looked around and that stuff said that the company is working on its own model called Ajax. And apparently the code references ChatGPT's API. So it works with certain prompts such as please summarize, please answer this question and please summarize a given text. So it says Apple doesn't plan to lean on ChatGPT permanently, so it's going to be using its own AJAX system. But Apple, if you've been paying attention to their research, I've done a couple of videos on their research, but as you know they have acquired tons of different AI companies. They've actually got a lot of different research papers out and they're rapidly developing their own systems and their recent vision system was actually beating GPT4. So don't sleep on Apple. I wouldn't be surprised if Apple's next AI integration at WWDC in 2024 is far superior than some things we've seen before. Then of course we did have something that wasn't that crazy after looking through the article, but it said that Chrome is getting three new generative AI feature. These are actually really simple, so I'm just giving you guys a quick rundown. Essentially you can just smartly organize your tabs in a really easy way. You can create your own themes with AI you can see right here. Basically all you're doing is just creating an AI generated background. It's not anything groundbreaking, but still something that's pretty cool. And then of course you can get help drafting things on the web. So writing on the web can be daunting, especially if you want to articulate your thoughts on public forums. So in the next month's Chrome's release, we're going to launch an experimental Chrome feature to help you write with more confidence on the web. So can you if you want to write a review, you want to write a comment on the website, it's basically going to help you write that with this little text box. So I'm guessing it's using Gemini Pro in order to help you complete your sentences and certain things like that. So overall, whilst this week wasn't particularly insane, I still think that these developments do show that things are ramping up. And of course it's always important to stay on top of the latest news. Now, if there was anything that we did miss. Now if there's anything that I did miss, don't forget to leave a comment down below letting me know exactly what that was. And until next time, I see you in the next one.