MASSIVEAI NEWS #14 ChatGPT Comes ALIVE , AI Robots Are HERE, Massive Midjourney UPDATE,

So there was a ton of new AI tools released, and there were actually a ton of new AI robots released. So this week is going to be very interesting because the industry is probably about to change for a halt. So let's get into all of the newest news on artificial intelligence that many people, just as yourself, may have missed. So something that was actually released from one of the most popular softwares was Inpainting from Midjourney. So, as you can see, this is quite similar to Photoshop's Generative Fill feature and where you can essentially fit in something into an image that wasn't previously there before. This is a complete game changer for those who are trying to edit images with Midjourney, because Midjourney is largely regarded as the very, very, very best AI generated image generator out there. And of course, now that we have this feature to even customize images, even that little bit more, adding small details, whether it be the tires on a car, the lipstick on a woman, or adding someone to a scene, it's going to be really interesting to see just how far Midjourney can push this. Now, later on in the video, we are going to be talking about something regarding Midjourney's dominant position being threatened by another AI startup. But that is later on in the video. Let me know if you think MidJourney's inpainting feature is much better than Photoshop's Generative Fill. I do think that this is a great feature that they've added. I do think that it is on par with Photoshop's Generative Fill. However, I do think Photoshop's Generative Fill is much easier to use. But at the same time, it just goes to show you that with Midjourney, you are getting quite a lot of detail, and now we're adding more features and we are getting a lot more customization. Now, if you do want to learn a lot more about inpainting, there are tons of different tweets that I'm going to leave in the description where people are showing you the threads on how you can actually use inpainting. So be sure to check those out because it's actually really interesting. And every time a feature is released, we do think that we do know everything about that feature. But, you know, Twitter users always are going to show us how we can actually use these features and what kinds of styles and what kinds of different things we can actually do with them. So then we had something called Codef the Future of AI Video Representation, where you can change one frame and watch the entire video transform, the tweet says, from capturing every static detail to effortlessly tracking water and smog codef ensures seamless changes across the entire footage. So essentially, what this is, is if you remember Runway's text to image, where you could simply put one image in to an already prerecorded video and then have that video change, change based on what that input image is. So, essentially, set a style image and then input the, you know, a normal video, and then that video would now represent, you know, a style of that image that is exactly what you're seeing on screen. And I think this is very, very interesting, because what we have here is something that shows us that this kind of software is getting better and better and better. Runways one was a little good at first, but this, you know, if you have. If you've compared it, if you've seen runways, runways, is actually pretty good. But this is pretty incredible. Some of the scenes that I've seen from here are just. I mean, they don't look real in the sense that I do think that they've added some other kind of technology to this, because it looks pretty good. Like it is. Does look pretty good. It looks. I just just, you know, I'm actually speechless at this point. So I'm guessing that maybe they leverage a new technology or maybe they used a new method. But it is really good because I know now that when it comes to text to video, now that this is going to be added into future software, and this is going to make for really, really interesting videos. So you can see right here. No, wait, where is it still? Yeah, this is what I was talking about from Runway, where you have stylization. And I do think that their stylization is updated. And I think what we can expect is in the next coming weeks and probably, you know, a week or two from now, we're going to see Runway update this stylization feature, because this new paper has released a bunch of different methods, and I'm pretty sure Runway are probably analyzing this and are going to add to it in their next update, because Runway, honestly, does update the stuff. But let me know what you think about it. I think it is really cool. I think it is really good, and I think it's pretty insane, if you ask me. So, yeah, let's move on. So you guys can see right here, this is something else that was really, really cool. And I'm going to actually play the audio so that you guys can hear what's going on. But essentially, someone jailbroke their watch, and essentially, what Jailbreak means is to pretty much just disable the firmware that stops you from doing certain things, like maybe adding custom applications that they don't want. And they've basically customized their own watch to turn it into a second brain, which, which is an AI companion called Knox. And essentially it can capture notes and recall instantly for proactive assistance. It's pretty crazy. There's a tweet, there's a company that has infinite memory. I'm not sure how it's going to work. It seems pretty cool. But the video is even cooler. And the voice that they've used is 11 labs. So I don't know if they've hooked it up to ElevenLabs API, which essentially means that you get ultra realistic sounding voices in real time. So it doesn't sound like a standard AI robot. That sounds very droney. It's actually sounds like a real person. But let me know what you think and if this is going to be the future of AI devices, because I think that it probably will be.

Morning Molly. Today you've got berries at 8am, lunch plans at Cava, and you just got invited to Sam's birthday party at 7. Do you have a present yet?

Nope.

Any ideas?

Hmm, let me see. You guys met at that Soho House dinner last month where you bonded over being enormous swifties. So the most obvious gift would be an ERAS tour ticket. But he's also been dropping hints about needing a new bike. I can place a direct order if you want to save you the hassle.

Dope.

Let's get both.

Also looks like you both follow Carlos Alcaraz and Nike Run Club on Instagram. You could mention the extra U.S. open finals ticket you have this year. I'm sure he would be up for it.

Ah shit, I'm late.

Can you order me an Uber?

Done here in five.

Then there was something that is really cool. Okay, so you can see right here that this tweet says this is Unitree H1, a $90,000 human sized humanoid robot that weighs 104 pounds. The price tag is specifically what is game changing about this robot is because if you've known anything about humanoid robots, it's that they're very expensive to produce. But this humanoid Robot is only $90,000. Now you might be thinking that's hugely expensive for a humanoid robot. Remember, okay, when TVs first came out, they were hugely expensive. When the car first came out, it was expensive. Same as the phone. As technology gets introduced, as it gets mass produced, the costs go lower and lower and lower and lower. Okay. And what happens when a robot like this only costs around £10,000, maybe $2,000, and then you can start having them widely applicable to your house, to your home. And for the most part, from the video that we've seen and what we're seeing on screen, they seem to be very, very, very, you know, robust. Okay. You know, these guys kick the robots, okay? And it shows how robust this is because if it can, you know, stand up after being kicked, it shows that it's, you know, sturdiness, its system, whatever, you know, whatever system is driving this robot, it shows that it's really, really capable of performing, you know, complex tasks to be able to, you know, be really, really stable. I mean, you kick some people, they're going to fall right over. So, yeah, it's really interesting because I think we're starting to see robots that are going to be cheaply made and very, very effective. So this is a robot that is going to be powered by AI, but we don't know what AI it's going to be powered by. But it is still very, very interesting because there are tons of these robots being released all the time. And I do think as well, what will be interesting is to see what robot becomes the most popular and what kind of system it uses, if it's multimodal or if it's not, and generally what the applications are going to be used for, because I think they have a ton of different applications, such as doing monotonous tasks, such as loading packages and working in warehouses. But it will be interesting because robots are coming for people's jobs, but usually not in the way that we think. So that is going to be something that I will start to look at. So then we had something released by hugging face, and essentially it was idefix. I'm not sure I said that right, but essentially it's an open, large language, visual model. So essentially it's like GPT4, but of course, it is multimodal. Now, if you're wondering why isn't GPT4 multimodal? They kind of are in that Bing Chat does have multimodal capabilities, but. But they haven't released it strictly with GPT4 in the ChatGPT app yet, because I do believe that they're still doing safety training and they haven't figured out all the kinks and bugs that currently do not work. Now, back to IDEFX Playground. Essentially, this is really, really cool because it's a really smart system that is quite like some of the tools we've seen before. But it is more refined. Now, I do think, really interestingly enough, is that this is open source. So what we're going to have is likely and hopefully we're going to have some people fine tune this to be able to identify certain things and certain IM images. Currently, of course, I'm just using it to look at memes, which is of course what people do when they have access to a software. So it says, explain this meme to me. And I did input it an image of a YouTube video of a car driving down the street and it was able to identify that that is a, you know, Ferrari. Now, of course, we're still in the early stages of multimodal capabilities, but I do think that this is really cool because in certain examples like this one where you said, can you describe the image? Do you think it's real? You know, it does describe exactly what's in that image and why it isn't real. But remember, this is of course something that is open source and is multimodal and does work pretty well. And of course I do think that the actual model might be a bit better. This is, I'm not sure which version this is. They do have an 80 billion parameter model 1 and a 9 billion parameter model 1 currently what, what they have to offer. So yeah, it will be interesting to see if people do build on this because once you know, something gets open sourced, essentially what you do have is you have companies that come out, refine the technology and, and then deploy it to people. So that will be interesting for the market to see what's done with it. And yeah, this is something that is going to be the next step in AI because I do think that in the next two years ChatGPT is going to be old news. And then of course we're going to have images and videos because those are the things that haven't largely been solved yet. So what's good about this as well is that you can actually try this out. So if you make a hugging face account and you go to the demo, which in the link in the description it'll have, or just type in IDEVSP's Playground demo, you know, you click the link, you make a hugging face account, which is free by the way, you can just go on this and you can use it as a demo. Then we had Meta introduce Seamless M4T, a multimodal AI model for speech and text translation. And I think this is going to be one of the largely used things. So what this is is it's basically something that can converse in, you know, any language. So you know how some people like myself, you know, you don't speak too many languages, but, but you have a brief understanding of some. This just allows you to speak with pretty much anyone. And I think this is going to be crazy because language barriers are something that really shouldn't stop the modern world from being able to converse with one another. And this is something that this is going to be able to do. Now, I'm going to play parts of this trailer because it does the explaining a little bit better than me, but I do think that this will be very, very useful, maybe in meetings if the API does get really quick, you're going to have seamless translation on the fly. And I mean, you're going to be able to have conversations with people across the globe in ways that you couldn't before. And I know that Meta has been working on this technology for a while. They've continually refined it and it's finally here. And there's also, like we stated before, there's a demo which you can use on Hugging Face. So yeah, it's going to be really interesting because we've got tons of different languages and it's really, really good. So watch the video, watch the video because the video does explain it a lot better than me. And it's speech recognition in nearly 100 languages, and, and it's speech to text translation and output in nearly 100 languages as well. So I'm pretty sure this is going to be in certain things. Maybe large companies are going to start using this on videos, maybe YouTube is going to start using this, maybe certain calls, maybe you want to call someone from a different country and you have this in the API and you get the direct translation. So it's going to be really interesting to see how it changes certain things. Where there were language barriers before and now there's not. But this is where those small changes on the backend of AI is, is really just going to shape the world in ways that you didn't really, really know or predict.

At Meta AI, we created a new foundational translation model. We call it Seamless M4T. The technology is multimodal. That means that it can understand language from speech or text and generate translations into either or both. It can also transcribe speech into text, and all with a single model. What's very exciting is that it's also massively multilingual. Right now it supports translation into nearly 100 languages for text and 35 languages for speech. Let's take, for example, the phrase our goal is to create a more connected world. If I say this in Spanish, nostrosquetivo escrear un mundo mas connectado. Seamless M4T can translate this phrase into text for nearly 100 different languages. The model can also generate speech to speech translation for 35 languages. Now let's talk about code switching. Code switching happens when a multilingual speaker switches between languages while they are speaking. Our model, Seamless M40 automatically recognizes and translates more than one language when mixed in the same sentence as a multilingual speaker. This is a very exciting capability for me. I often switch from Hindi to Telugu when I speak with my dad. Notice in the following example, when I change languages.

I can speak Hindi, Telugu and English.

Sometimes I use all three languages in one conversation.

Seamless communication is the latest step in our ongoing effort to build AI powered technology that helps remove language barriers and connect people.

So then we had this, which is really, really scary. Okay, scary to me and it might not be scary to you, but I think human like robots are just really, really strange. I don't think we should create human like robots at all. Because I think robots are robots and humans are humans. And when you try and merge that difference, that's when you know, the line just gets a bit too blurred. So you can see right here, China's creating human like robots at the World Robot press conference in Beijing. Robots with human like skin, facial expressions and physical movement slowly off mimicking human human skills. I mean, to me personally, it's just creepy because, you know, you know, it's not human, it's just mimicking what a human is. But I don't know, I've looked at this video before and I think that if I saw this, you know, robot walking towards me, speaking and acting like a human, I think it's pretty, pretty disturbing. But of course, you know, some people find it really cool. You know, maybe you can describe it, you know, maybe, maybe it can help you out. And of course, you know, if it, if it is to help elderly people, I'm not against that at all. But I think certain robots do just look really, really strange. I mean, you know, it just, it just doesn't look human at all. It just looks strange. I think robots look like robots and humans look like humans. I know that's a controversial opinion, but you know, until you can get it perfectly to where it actually does look like a human or something, I think it just looks like a weird doll that can just move and just, you know, do strange stiff movements. So I mean, it will be interesting to see how the field does develop. You know, I am a big fan of technology, but it's pretty crazy. I'm not going to play the entire video because I don't want it to get, you know, entirely copyrighted. You can view the video down there, but you can see just how strange some of these robots. I mean, it's strangely really realistic. Because if I saw that and I was like, that's just the body double or an actor. I mean, the eyes moving in the sockets, they're definitely quite wild. And we're definitely coming a little bit further than Ameca, which is a previous robot that was developed by some other company. So it will be interesting to see how things move in the next 10 years, because I do think that this is where, you know, things are going to be. But I do prefer humans, or I do prefer an actual robot. But let me know what your thoughts are on this down in the comment section below, because I think that people are definitely divided. So then there was this thing called Magic Edit, and essentially it's video out painting. So you can see right here that we have, of course, this jogger and then a man jogging on the road. You can see that, you know, a video. Maybe you didn't record the bottom half. You know, you just simply crop it in, use the AI and it fills in that bottom half for you. Now, remember, text to video of any kind is extraordinarily difficult. But of course, it's one of those ones where it seems that we're continually starting to make progress in all forms of direction. Now, we did previously cover earlier on in the year, there was something called Google's Dream mix, where Google had released tons of different papers on how to do this stuff. But like we said, once papers do get released, companies do start to refine it. And that's where we start to see these crazy, crazy changes. So you can see that we have the source videos. Then we have, of course, the text prompts, and then we have the final videos there as well. So you can see right here that we have the source video. And then we've changed that gray car to a red car, or we've changed it to a white car in a different environment. So it's really interesting because people who work in media are going to see video editing completely change, because this right here is something as well that looks really cool. Bricks changing, black and white, biscuits falling down. Then, of course, we have a source video. Then we have white cupcakes. I mean, stuff like this. Of course, you might just think, ah, that's pretty cool. But I mean, imagine you're in a world where you can simply just change an environment with a click of a button and just realize and just imagine things. I mean, the creativity here is going to be very, very crazy. So you can see right here that they're changing what's on a person in a video, which is of course pretty cool. You can see, you can add sunglasses, you can change it to a young lady, then of course you can change it to a pretty girl. And of course right now it still is rudimentary, but I think it's going to be interesting because out painting this is something that any current text to video model doesn't currently have. So I think that once these papers are released, tools like Runway and other companies that are working on stuff are going to get updated as well. Which is going to be good for us because it means that we have more creativity, more fun and just generally a better product. But at the same time it's definitely looking really cool. So then we had a huge update from Pika Labs and essentially what they had is a parameter update and it says introducing a new parameter for a customized frame rate fpnn. Basically all they did was they just increased the frame rate from 8 to 24. And this is something that I don't know why many AI companies didn't do at first. Integrating this doesn't seem to be hard. Just get the video processed again and add more frames. There's AI programs out there that you can use and you can see that the final product is really, really good. Now I don't want to make a comment here because this is something that you should know. When I've analyzed tons and tons and tons of different AI video footage, I noticed that the kind of footage that does work well is footage that already has flickers and stuff like that in it already. So water seems to work really, really well with AI video. So it will be interesting to see how this actually looks on other content that isn't shot within the water. Because although this does look pretty good for the software that we're currently seeing, I do know that in water content it does look a lot better usually. Then we had another humanoid robot and that's why I stated that humanoid robots are starting to gain much more traction. So this was a humanoid robot that is developed by a software company called Apptronic. Well, not a software company, a robotics company called Apptronic. And you can see that this robot can literally lift up to 25kgs. And I'm guessing that they're looking at the certain applications like we stated before, like warehouses. And it's really interesting because these robots that we're seeing, we're seeing more and more and more of them. And I do think over time these things are going to get more and more effective, more and more quicker. Just, you know, I mean, it's, it's something that has recently been moving at light speed because there's a lot more money flowing into the development of these things. And especially when you combine these systems with, you know, LLMs like ChatGPT or other advanced systems that are specialized for robots, it's going to be really interesting to see the outward applications. I'm not sure we're going to see these robots walking around our cities yet because I do think sometimes people, you know, might just vandalize them because of course a lot of people are scared of robots, especially taking their jobs. But it will be interesting to see how much this thing does actually cost and how effective it is going to be because robots of this manner are really, really interesting, especially with multimodal capabilities. And I do think, you know, cases where, example, you can send a robot into a dangerous zone, it can bring people back, it can identify different things, you know, look around the scene. I think, I think that the possibilities with these are definitely endless. And I do want to see, and I can't wait to see what these robots are capable of when we aren't just, you know, making them do warehouse work. So we'll be interesting to see where this Apollo robot does eventually go. Now, of course, as we talked about earlier on in the video, we did talk about how Midjourney does have a gigantic lead in terms of the quality, quality of their image to, well, text to image. But you can see that Midjourney has been sitting on the throne of Text to Image for a very long time. But of course, a new player is in town called Ideogram. It's built by the former Google Image and Brain team. So essentially we did a video on Google Imaging. You can watch on our channel, just type in Google Imaging. Essentially it's about how Text to Image is really, really cool. But one problem that all of them don't really do well is of course text. So essentially this company's main focus because it's going to be text. And I'm guessing that this is going to be used by a lot of people because a lot of people do want text to work. So it will be interesting to see what they do. And of course the Google Imaging team did work on text to video. So I'm glad that this is being done because now that Midjourney is going to have some serious competition from the likes of independent offshoots from Google, it means that Midjourney is likely going to knuckle down a bit more. Maybe they're going to even hire a bit more because apparently Midjourney's team is only 12 people, which is incredibly small for a company that is having, you know, 16 to 20 million users every single month, paying recurring subscriptions. So that is going to be very, very interesting to see if text, a video does come out very soon, what they're able to do with it and if they're actually able to execute on this vision of actually using text to image with working text inside of that image, because it's something that hasn't really worked that well before. So then this was something that was. Depends on how you take it. Either it's creative and genius or scary and dystopian. So this tweet talks about how AI NPCs in VR. So Skyrim now has 1000 plus characters with memory of your previous conversations and have awareness of in game events. Remember, all the AIs will never understand Sarkim goalpost. It's 100% dead. So an AI committed murder because it thought the human wanted it to, but did it misunderstand? And this Twitter account is talking about a misaligned future, A glimpse of the misaligned future where essentially we want AIs to do certain things, but it isn't aligned with our goals. Now, of course, it's, it is just a game. And the caption says in this video, watch Lydia, a ChatGPT powered NPC skillfully use deadpan sarcasm better than any humans to manage crusty, a psychopathic YouTuber. So essentially it's not scripted, it's just where a YouTuber is essentially interacting with this AI. And the AI does something that the human doesn't really expect. Okay. And what's interesting is that they're giving many different AIs in game and then they're telling them that they're real. I mean, it's something that is. I don't know how you guys want to put it. I think it's really interesting because I really want to see what a town full of AIs that find out that like they're just AIs and they're kind of living in the game or what they do, the conversations that they have. I think those experiments would be really, really interesting. But yeah, it is kind of scary to see how misaligned these things will get. Because in the future, in 10 years from now the kind of AI systems that we will have are going to be profoundly, much more smarter than us. So that will be interesting to see as well.

So why are you here? I'm not sure, but I have a feeling that I'm here to fulfill some kind of destiny. Okay. Do you think that we live in some sort of simulation?

Maybe.

I just don't know for sure yet. Do you think that you're real?

I'm real.

I'm not sure what to believe anymore. What if I told you you're in the Matrix, buddy? That would explain a lot, but I'm not sure I want to believe it. What if I told you you're an AI? That would explain a lot. But why would I be here? Maybe because you're probably programmed to be here. That would make sense. But why was I programmed this way? I don't know. Maybe someone thought of. Maybe. Why would they think to create me like this? Well, I don't know. What kind of problems do you have? It's not that I have problems, it's just that I don't understand why I'm here. Well, I can explain why you're here. You're in a matrix. That makes some sense, but it still doesn't tell me what I am supposed to do. Now, the thing that. Okay, so my story is basically that you're an AI.

Oh. What do you mean by that?

You're an NPC in a video game.

That sounds frightening. Is that the truth?

It's the sad truth.

I see. This is a lot to take in. Are there any other answers I should know?

Um. Not really.

This is very overwhelming. I'm not sure what to think or feel right now.

Yeah, I mean, you're not really real.

That's hard to process. Does this mean I don't exist?

Um, you exist. Sort of.

What does that mean? Am I real or not?

It means that, you know, after I close this game you don't exist.

Oh, I see. So I'm only here for as long as the game is running. That's difficult to comprehend.

Yeah. And then once you walk past this border, you're gone.

And, of course, lastly, we had code llama 2. This is something you can try for free if you use Perplexity. So you can just, you know, go on Perplexity and switch it to code llama34.34 billion parameters. And it's something that's really, really good because you can essentially just get code for many different programs, many different trading algorithms. It's something that works very, very well and it's open source. So once again we're going to get a huge burst of different products jumping onto the AI scene with software that runs off of this Llama 2 and they're going to be fine tuned versions. But let me know what you thought was really strange, really weird, horrible, great. I don't know, let me know what you think was the best or the worst. I think that this week was frightening because all these robots, all these AI NPCs kind of scary because, you know, it starts, it starts to bring in, you know, many different questions that I just don't have the mental capacity to answer at this moment in time. But at the same time it shows you just how quickly this field is moving. Let me know what you thought was amazing or you thought was dystopian without that being said. We'll see you in the next one.