AI News: The Robots Are Coming!

In this video, I'm going to break down the AI news, share some really cool research and tech that's coming out, and at the end of the video I'm going to share who won the meta Ray Ban glasses now that I've passed 500,000 subscribers. So let's get into it, starting with this email that OpenAI sent out to pretty much everybody that's on their API list. The announcement says, Dear GPT Builder, we want to let you know that we will launch the GPT Store next week. If you're interested in sharing your GPT in the store can, you'll need to and then it kind of gives you some instructions on how to share your own GPT in the store. Now, if you don't remember what the GPT Store is, Sam Altman actually announced it back on Dev Day in November 2023 and it's a store where you can create your own custom chatbots inside of ChatGPT and then resell them in their GPT Store.

You can list a GPT there and we'll be able to feature the best and the most popular GPTs, of course, will make sure that GPTs in the store follow our policies before they're accessible. Revenue sharing is important to us. We're going to pay people who build the most useful and the most used GPTs a portion of our revenue.

Essentially, you'll be able to create your own GPTs using their GPT builder, and after some sort of approval process you'll actually be able to monetize those GPTs. That feature is coming next week. One interesting thing to note here, and this is from Nick Dobos, who is the creator of this Grimoire GPT, is that if somebody actually uses your GPT and needs some customer support on it, it looks like people can send support emails directly to the creator of GPT. You can see here in this screenshot somebody can write a support message and it says send a support email to the builder of gwymroar. If you do build a really cool, really successful GPT, you could probably expect a whole bunch of customer support tickets to help manage that, which is likely going to be a deterrent for some people to actually create GPT. Or on the subject of OpenAI, it's recently come out that New York Times is suing OpenAI because they were able to get ChatGPT to give back articles written by New York Times and the text was given back verbatim from the article. Now I talked about that in last Friday's news video. Well, it turns out that OpenAI is actually trying to get a lot of these news companies to allow training on their news. However, they're only offering them between 1 million and $5 million deals for this information. However, if people start going to OpenAI and chat GPT to get all of their news information instead of some of these original news sources, these news sources stand to lose a lot more than the 1 to 5 million dollars that they would make off of their deal with OpenAI. And to put it into even more perspective, it says here that Apple is looking to partner with media companies to use content for AI training and is offering at least 50 million over a multi year period for the data. I think for some of the smaller news sources, a deal like this would be an amazingly phenomenal deal for them. For some of the bigger news sources like the New York Times, Forbes, companies like that, it would be a pretty horrible deal for them because they stand to lose a lot more than what they gain off of it. But as we enter this new era of AI where more and more people are using AI to get their news, to get their articles, to get their information, I personally think a lot of these news companies should really start to rethink how they approach their business model because they may be really, really fighting tooth and nail to hold onto a business model that could be fairly obsolete not too far in the future. Now, I did mention in past videos that I do believe the people that are doing the investigative journalism, the people that are doing the research, the people that are spending the time to create this content that chatbots are then getting trained on do deserve to be compensated. There does still need to be incentive to want to do the research and put in the time to create this content. But it just needs to be rethought. I think people are getting increasingly fed up with every dang news site these days having a paywall. And if there's no paywall, they are just inundated with insane amounts of ads all over the place. That really screws the user experience when you're trying to read the content. There has to be another way other than paywalling it all or making our eyes bleed with advertising. So hopefully some new, more future proof models for content creators, journalists, and the people that are actually out there reporting on the news are sort of figured out in 2024. This week, Microsoft introduced a new key that's gonna go on future Microsoft keyboards. It's called the Copilot key. And it's gonna look like the little copilot logo. And when you press it it's going to pop up that little sidebar where you can access Copilot on the right hand side. Now, if you're on one of the newer versions of Windows, you've got this little button down here right next to your search bar in the bottom left of your screen. If you click on that, it brings up your Copilot bar. This new button that Microsoft wants to add to future keyboards is going to be a single button press on your keyboard to bring this up. Not the most exciting news in the world, but it does show that Microsoft is really, really committed to baking AI into everything they do in the future. It looks like Google is getting ready to release Gemini Ultra pretty soon, but it's going to come at a premium price tag to be able to use it. Right now. Google Bard uses Gemini Pro. All of the demos we saw from Gemini were actually demos of Gemini Ultra, which is their really powerful model. And according to a developer over at Google named Dylan Roussel, it will be called Bard Advanced and you'll be able to get three months access to it when you sign up for Google One, which is I guess a sort of more expensive Google Drive that gives you access to more storage and things like that. We don't know when we're going to get access to this. All we're really working off of is this one tweet from Dylan. We've been seeing a lot of amazing text to video and image to video models lately from companies like Runway and Pika and models unlike most of the other video models that are out there like Pika and Runway, this one is available as open source and we can see right here on Alibaba's gift GitHub page. All of the code is currently available. All of the installation instructions to install it locally or on some sort of cloud computer are available right here. I haven't personally installed this yet, but if I can get it working on my own computer, I'd get some good results out of it. You can bet. I will make a tutorial video and walk through how I got it to work. Lately, when it comes to chips being used for AI, we've been hearing a lot from companies like Nvidia and amd, but not a ton from Intel. Well, intel just spun off a new AI company called Articulate AI with an 8 in it, and this appears to be Intel's division that's going to be fully focused on AI. According to this article here, Articulate plans to deliver new software products and services, including Gen AI powered products. From what I Can tell this articulate company isn't going to be creating AI chips, but more like provide Cloud Compute Service and GPUs that companies can have access to to train and run their AI models on. There's a new AI out that can read your mind called D Wave. A lot of the other research that we've seen so far around some of this, a high mind reading tech that's out there either requires somebody to slide into an MRI machine or get a brain implant like the neural link. Well, this one you just wear a funny looking hat or electroencephalogram cap or eeg. Now they did put out a demo video here on their website that sort of explains how the technology works. If I'm being totally honest, it's a little bit over my head at the moment. But let me go ahead and fast forward to them actually showing it in action.

Yes, I'd like a bowl of chicken soup, please. Yes, a bowl of beef soup.

Now I did speed that up a little bit. But essentially this person's wearing this EEG cap. They think silently. There is a little bit of a pause between them thinking, but then their thoughts show up on text on the screen and then an audio voice reads out the text. The idea being that this can help paralysis patients in the future better communicate. This week, the Supreme Court in the US also put out their year end report on the federal judiciary. And much of the report is actually about AI and how the courts should really, really be considering how to use AI's advantages and understand their disadvantages. Starting on page 5, here we face the latest technological frontier, artificial intelligence. At its core, AI combines algorithms and enormous data sets to solve problems. It goes on to sort of explain what AI is facial recognition, voice recognition. And it talks about how a lot of people are worried that AI is going to replace a lot of jobs in the legal system. Like maybe in the future we'll have AI judges or AI lawyers. But it also cites some cases where lawyers have actually presented past cases that AI talked about that didn't actually exist because the AI hallucinated them. So not only does it talk about where AI can benefit the courts and the legal system, but also the things we need to worry about. They say here machines cannot fully replace key actors in court. Judges, for example, measure the sincerity of a defendant's allocution at sentencing. Nuance matters much can turn on a shaking hand, a quivering voice, a change of inflection, a bead of sweat, a moment's hesitation, a fleeting break in eye contact. And most people still trust humans more than machines to perceive and draw the right inferences from these clues. I'd argue that AI is not too far off from drawing the right inferences from those clues as well. But as it stands right now, yes, I do believe that there still needs to be humans in the loop when it comes to the court system and determining somebody's innocence or guilt. It goes on to say that humans should be in the loop, but they do see AI as something that can help make their jobs easier. It's a really interesting read. I did read the whole thing, but for the most part I just gave you the Cliffs Notes of what they said regarding AI. 2024, in my opinion, is going to be the year of robotics. I mentioned that in my predictions video. I believe it even more we're kicking off the year with so much robotics news coming out already in just the first week of the year, including Google DeepMind showing off some of their latest research. This includes Auto rt which harnesses large models to better train robots. You can see a little bit of a diagram here of how this works. It maps the environment, describes what it sees, generates tasks based on what it sees, filters out the tasks that it can and can't do, does the tasks that it can, comes back with some sort of score, and then repeats the process. They also announced their SARAH RT or Self Adaptive Robust Attention for Robotics Transformers. This is a way of making models more efficient by fine tuning or what they call up training. Basically converting a lot of the complex computations it's doing under the hood into less complex computations. Says here, when we applied SARAH RT to a state of the art RT2 model with billions of parameters, it resulted in faster decision making and better performance on a wide range of robotic tasks. You can see some examples here. Closing a drawer, knocking over a Coke can, moving an orange can near a green rice chip bag, moving a red Bull can near Blueberry Rx Bar, etc. And finally their RT Trajectory, which helps robots generalize. It says RT Trajectory automatically adds visual outlines that describe robot motions in training videos. RT Trajectory takes each video in a training data set and overlays it with a 2D trajectory sketch of the robot arms grippers as it performs the task. We can see in this demo on the right we have a human sort of drawing the path that it wants the robot to take here and then on the left we see the robot mimicking that path. Now while we're on the topic of robotics, these videos have been going somewhat viral on X lately for mobile Aloha learning BI manual mobile manipulation with low cost whole body tele operation. They really like to make these names fun. But here we see this general purpose robot that can do all sorts of things. For example, in this demo video we can see it cooking shrimp totally autonomously. This video's sped up to six times speed. But we can see down here some other skills that are showing in real time, like wiping wine off of a counter here, or pressing the button to call an elevator. And then when the elevator comes, actually getting on the elevator or using cabinets or rinsing a pan, pushing chairs in and out. And it can even high five people. And all of these that we're seeing in these demos right here are being done completely autonomously. And then they have the tele operation where we can actually see somebody behind the robot operating the robot. So it's kind of hard to tell in this video here, but if I move forward a little bit here, we can actually see the guy's legs here behind the machine operating the machine. This robot is being operated by a human. I don't know if this is used for training purposes as well, or if it's just a way to not have to use your hands. I would assume that a lot of the tele operation is how it actually is trained to to then do these things autonomously in the future. Here's another example showing at 10x speed of the tele operation with somebody actually behind the machine doing all the work that the robot's following. I don't totally understand the purpose of this if it's not actually being used for training purposes. So digging into the paper a little bit here, it says our teleoperation system is capable of multiple hours of consecutive usage, such as cooking three course meals, cleaning public bathrooms and doing laundry. Our imitation learning result also holds across a wide range of complex tasks, such as opening a two door wall cabinet to store heavy cooking pots, calling an elevator, pushing in chairs and cleaning up spilled wine. With CO training, we are able to achieve over 80% success on these tasks with only 50 human demonstrations per task. Moving down a little bit here, it says augmenting the aloha system with a mobile base and whole body tele operation allows us collect high quality demonstrations on complex mobile manipulation tasks. Then through imitation learning, CO trained with static aloha data mobile aloha can learn to perform these tasks with only 20 to 50 demonstrations. So based on my understanding, you can teleoperate it to help train it along with training it without teleoperation, but using the teleoperation features to Help train it makes it much more effective. Here's what's really, really cool about this mobile aloha. It's is the whole thing is completely open source. You can build one of these for yourself right now, today, with existing parts that are available to buy on the Internet. If I click on this little tutorial button here, you can see it brings me to a Google Doc with a step by step tutorial on how to build one of these yourself. And all it's going to cost you in parts is $31,757.86. But all of these parts are accessible through Internet ordering. Like these robot arms that you need two of and these robot arms which you also need two of. And once you've got it built, the hardware code and the learning code are Both available on GitHub as well as some of the data sets they've already trained it on, like the ability to push in chairs or wipe up wine, or operate an elevator or operate a cabinet. So if you feel like doing a $32,000 fun weekend experiment, I got one for you. There's also been some interesting announcements about announcements. For example, this company at Rabbit Tech says that on January 9th at 10:00am Pacific Time, they're releasing something interesting and I actually find their trailer kind of funny. Now I'm actually muting this trailer here because I don't know the status of the copyright on the background music, but I find it funny because they're holding whatever this Rabbit device is in their hands and it's completely blurred out in pixelated. And whenever I see that it's it just reminds me of like, I don't know, you imagine some sort of adult toy or something in their hand. Just trying to imagine what they're pixelating and hiding just makes me laugh for some reason. But it appears to be some sort of device where you speak into it and it will be somewhat of like a pocket assistant. Some of the examples are them just speaking into it and saying, order me an Uber. The example of this guy on the screen right here, he's saying that was delicious. Look in my fridge, see what ingredients I need and order them for me so I can make that same recipe again tomorrow. And that's the prompt that he's speaking into this unknown blurred box. So we'll find out more about what this Rabbit Tech thing is when they finally reveal it to us on January 9th. I'm not necessarily sure if people are going to want another gadget outside of their like mobile phone in their pocket. If this is something else like you have a phone and you have this gadget that you talk to. I can't really see it doing super well because eventually we'll just have this tech in our phone anyway. But I'm still curious because they're not telling us what it is and they're being very mysterious with their blurring open loop concept of an ad here. The other announcement of an announcement is that Samsung is going to make an announcement on January 17th. So on January 17th we're going to get a sneak peek at Samsung's next phone. Samsung said that its latest device will offer an all new mobile experience powered by AI. We don't know what that means yet, but we'll find out on January 17th, I guess. And finally next week is the annual Consumer Electronics show in Las Vegas. And the theme of this year's show is pretty much AI is going to be in everything. So although the year has started off somewhat slow with AI news, I imagine after CES we're going to be flooded with AI news. There's announcements from Nvidia lg, which is featuring its new AI processors, Samsung, which it says here we'll have to wait until the 17th to find out about their new phone. They won't be announcing that at ces, but in a press conference. The theme of Samsung at CES is AI for all Connectivity in the Age of AI. So whatever they're going to be talking about, it's going to revolve around AI. Now I'm actually headed out.