AI News: What Does Trump Mean For AI?

Well, this week was election week here in the US and for the most part, the election is what dominated the news cycle, at least here in the US and probably in most of the world. And I think a lot of the companies they're building in the world of AI, and it went, hey, we're going to sit this week out, there's enough news going around, we're going to get overshadowed. With that being said, there was a whole bunch of small updates that I think are worth talking about. But before I do, let's talk about the election. As we all know, Donald Trump was elected the 47th President of the United States, and I am not going to get political in this video, I promise. But what I do want to do is I want to break down some of the potential implications of the election on the world of AI, And I'm going to do my best to do it without bias and just sort of share the details that we know what is sort of out there in the open about what might happen as a result of Donald Trump being the new president in the US Starting with the fact that Trump has claimed that he is going to repeal the AI Executive order that Joe Biden put into place during his presidency. The exact quote when I'm elected, trump said at a rally in Cedar Rapids, Iowa, last December, I will cancel Biden's Artificial Intelligence Executive order and ban the use of AI to censor the speech of American citizens on day one. Now, if you need a quick refresher course on what the executive order included, it required for companies developing foundation AI models to notify the federal government and share results of all safety tests before they're deployed to the public. This article on TechCrunch says the order specifically targets any foundation model that might pose a risk to national security, economic security, or public health, which, while somewhat open to interpretation, should cover just about any foundation model that comes to fruition. The order also outlines plans to develop various new tools and systems to ensure that AI is safe and trustworthy with the National Institute of Standards and Technology, the nist, tasked with developing new standards for extensive Red Team testing prior to release. It's also designed to protect against the risk risks of using AI to engineer dangerous biological materials, protecting against AI powered fraud and deception, and establishing a cybersecurity program to build AI tools for addressing vulnerabilities in critical software. And again, as Trump has stated himself, he plans on repealing this order, theoretically making it easier and faster for the tech companies to sort of speed up innovation. Again, at least that's the Thinking here, Trump allies have claimed that they want to make America first in AI with a sweeping executive order. Apparently, this new executive order aims aims to boost military technology and reduce regulations on AI development. The order outlines a series of Manhattan projects to advance military AI capabilities. And it calls for review of what it terms unnecessary and burdensome regulations on AI development. Now, along with the election of Donald Trump, we also get Vice President J.D. vance. And J.D. vance has some thoughts on AI as well. Specifically, he seems to be a pretty big proponent of open source AI. Back in July, he said this.

You know, very often CEOs, especially of larger technology companies that I think already have advantage positions in AI will come and talk about the terrible safety dangers of this new technology and how Congress needs to jump up and regulate as quickly as possible. And I can't help but worry that if we do something under duress from the current incumbents, it's going to be to the advantage of those incumbents and not to the advantage of the American consumer.

And the way I kind of interpret that is that he's worried that all of the big companies, the Googles, the Microsoft open eyes of the world, they could steer too much of the regulation in favor of them and away from the smaller companies and the people building open source. And he wants to make sure that that doesn't happen. That's my interpretation at least. And then we've got the Elon Musk factor. Obviously, Elon Musk became a huge proponent of Donald Trump and really, really helped push his campaign by using his X platform to really campaign for Donald Trump. As we know, Elon Musk X AI. And so Elon being in tight with Trump will likely have some implications on AI as well. If Elon has a little bit of over where Donald Trump takes the regulations with this stuff, it's most likely going to work in his favor and, well, Xai's favor. Donald Trump has also showed concerns about how much energy usage AI requires. Actually telling Elon Musk, you are going to need tremendous amounts of electricity, like almost double what we produce now for the whole country, if you can believe it. Now, how that's going to be rectified in the future, that's anyone's guess. And that's really all I'm going to share about the elections. I just wanted to sort of recap what we know right now from statements that people like Trump and his allies and Elon have said in the past. And that is the information that we know that could help us understand where the world of AI might go under President Trump, from the sounds of things, it kind of seems like a bunch of the guardrails might be lifted, some new ones put in place, but the fact that he wants to be first in the world in AI, AI sounds like they're likely going to have an agenda to sort of push AI forward and minimize as much regulation as possible. Again, that's just my interpretation of this all. We'll just have to wait and see if many of the things that he said he was going to do on day one actually happen. Now, moving away from the election, there has been a handful of smaller updates that I think are pretty dang cool that I wanted to show off. Starting with the fact that Runway ML actually released this new feature called Advanced Camera Control that allows you to actually control the angle of your camera when you're making AI generated video. So I have this image of myself already loaded into Runway here. I'll go ahead and drop it because it wants it to be in that sort of 16 by 9 here. Over on the left side here, you can see I've got a little grayed out box for the camera controls. If I click on that, it asked me to switch to Turbo. If I go ahead and select that, you can see that I can actually play with any of these sliders here and change the horizontal movement, the tilt movement, the pan movement. Here we got zoom, vertical movement and a roll movement. Let's go ahead and do a vert roll movement here and a roll movement at the same time. Click Generate and let's see what we get out of this. And I get a video with some interesting rotations on it. Not quite as cool as the demo that they showed off here, but something that I encourage you to go play around with and see what you get out of it. Bling AI is also rolling out a new feature for their video platform that allows you to train your own face into the video. Here's some examples I came across from the Tech Holla X account here, where we can see a video that clearly has his own face in the video as he talks on the phone and eats a hamburger and eats a pizza with people cheering around him and navigates the oceans on a boat as he gets splashed. It's pretty cool. I'll link up to the X post below because he does show off some more examples. He did mention that Ling gave him early access to the custom face model feature, so my guess is it hasn't fully rolled out to people yet. Here's something else that's really cool in the world of AI. Video. This is Xportrait 2 from ByteDance, the company behind TikTok and Cap Cut and some of those tools. We could take a peek at some of these videos, but what's going on here is these videos are actually taking the facial expressions and the mouth movements and the head movements of the actual camera video and then applying them to an image. And it maintains all the same emotions and sort of like micro elements of the movement. So check this out. Here's one of the demos. This image on the bottom left here is the image that they uploaded to be the face that, you know, changes from the video. And then down on the right here, we can actually see the video. So take a look.

You're not the only one who's got wants and needs but held on to your toy I took all my feelings, my wants and needs and dreams and I buried them inside you I planted a seed and watch and prayed over I planned myself inside you and waited to bloom it didn't take me no 18 years to realize the soil was hard and rocky and it was never gonna bloom.

You can see that the image followed the same emotion and had all the same expressions in the wrinkles in the forehead and the blinking and all of it. The as the head moved around, it actually followed this sort of driving video perfectly. Now, there's no code for this yet. It doesn't seem to be usable on hugging face or anything like that. This is just a demo of some of the research they're showing, but I will link up to this GitHub page below so you can check out some more of the examples. They are pretty dang impressive. We can see a comparison here between the original video and the driving image. There's. And then you have the original X Portrait and Runway Act 1, and you can just see how much more expressive this one is than these other versions that are available right now. And here's another one I came across from 100x engineers here on Instagram. This one's called Face Poke, and it lets you upload an image and then you could tweak the image however you want by just selecting areas of the image and then moving them around. And this one is available over on Hugging Face. So if I head on over to Hugging Face, let's do a quick search for Facebook and it asks us to choose a portrait photo. So let's go ahead and just pull in an image of my head here. And if I click this show face markers, you can see it kind of tweaked the way my face looks by default. But it put these face markers here, and if I click and drag, you can see I could make myself go to sleep. I could close my mouth, open my mouth, adjust which way my head is facing, lift my eyebrows up. Let's see if I could wake myself up here, get my eyes open. And you could easily manipulate the face of any image that you upload. Now, it does seem to lose some quality when I do it, but it's still pretty fun, pretty cool. It's going to help anybody get to exactly that image that they're looking for in the future. And while we're talking about AI images, Black Forest Labs rolled out some new updates for their Flux models, which a lot of people consider to be the most realistic AI model out there. They rolled out both the Ultra mode and the Raw modes. The Ultra mode could generate four times higher image resolutions, up to four megapixels. Here's some examples of those. And then the RAW mode is designed to create more authenticity in the images. And if we look at some of these images, they don't look like they're professionally composed. The angles are kind of weird. They look like more realistic images that you just kind of shoot on your iPhone or something like that. Maybe outside of the giant octopus playing in the bathtub, all the rest look like something that if you just came across it inside of, like, Google Images or something, you'd probably assume it was a real image. The tool KREIA just rolled out the ability to train Lauras inside of the platform, meaning that you can train it on specific styles. Like if you have your own art style and you want every single image to look like that color scheme and that sort of aesthetic, you can train a Lora so that every image that you create looks like that style. You can also use Loras to train your own face into the model and generate images with your face on them. Lores are super, super powerful to make sure that you can get some sort of consistent look, some sort of consistent face, a consistent style. If you've been messing around with AI images for a while, and if you use stable diffusion, you're probably already familiar Lauras. But we're starting to get these Loras rolled out into more and more of these AI art generator platforms. Now, this is still rolling out. They gave out a hundred invites to people to try it out, and this post has been seen by 70,000 people. Unfortunately, I was not one of the first 100 to get an early invite, so I can't actually show off the Laura functionality inside of Kreia. But it's there for some people and it'll be rolling out soon for others. Moving on to large language model news for this past week, Anthropic added a minor upgrade in that now when you upload a PDF, it can actually read what's in the images and not just the text of the PDF. So in the past if you wanted to upload a giant ebook or something, it would read all the text of the ebook. But if there was images or graphs within the ebook that had text within the image, it wasn't reading those, it was just reading the actual text in the document. Well, now it's going to read both the text and what's in the images when you upload. A nice little quality of life improvement if you use Claude a lot. And while we're on the topic of Anthropic, they did something quite interesting that the AI world has been talking about, and that's that they picked up the price of their smallest Haiku model while almost every other AI platform is lowering the price to use their API. It just seems like it was a very interesting move for Anthropic to raise the price of using theirs, they said. During final testing, haiku surpassed Claude 3 opus, our previous flag ship model. As a result, we've increased pricing for Claude 3.5 haiku to reflect its increase in intelligence. A lot of people on X have called this the Haiku Intelligence Tax. It got smarter, so you got to pay more for it. Now in all likelihood, unless you're a developer who's building software on the back of Claude, you probably don't care about this information. But the AI world seemed to find it fascinating. And in the final bit of Anthropic news, Anthropic is actually partnering up with Palantir, the defense company along with Amazon aws, to allow the US Government government to use their models for intelligence and defense operations. This is to support government operations such as processing vast amounts of complex data, rapidly elevating data driven insights, identifying patterns and trends more effectively, streamlining document review and preparation, and helping US Officials to make more informed decisions in time sensitive situations while preserving their decision making authorities. But it doesn't sound like AI is going to have their finger on the button to like launch missiles or anything like that. Just that the government can use it for more like analytic type purposes. Anthropic's not the only one. Meta said this week that it's making its Llama models available for U.S. national security applications as well. They said. We're pleased to confirm that we're making LLAMA available to US government agencies, including those that are working on defense and national security applications, and private sector partners supporting their work. We're partnering with companies including Accenture, Amazon Web Services, Anderal, Booz Allen, Databricks, Delaware, IBM, Leidos, Lockheed Martin, Microsoft, Oracle, Palantir Scale AI and Snowflake to bring LLAMA to government agencies. And since we're on the topic of Meta, Instagram is planning to use AI to catch teens lying about their age. Instagram recently started rolling out their teen account feature and they're building this proprietary software tool that it calls its Adult Classifier. It's designed to break people off into two categories, older or younger than 18. They're going to use this by looking at the person's own account data and sift through their profile, see their follower list and what content they interact with and will even an unsuspecting Happy Birthday posts made by friends to predict a user's age. So if you're under 18 and trying to claim to be over 18 and you want to use Instagram, well, Instagram's probably going to figure out whether or not you are actually over 18. OpenAI managed to steal a member away from Meta this week. Caitlin Kalinowski was the head of their augmented reality division. She actually helped develop the Orion glasses that we saw last week month he posted on her LinkedIn that she's delighted to share that she's joining OpenAI to lead hardware in the new role. She will initially focus on OpenAI's robotics work and partnerships to help bring AI into the physical world and unlock its benefits for humanity. GPT got a new domain name this week. If you go to chat.com it will actually redirect you to chat GPT now. Well, it turns out that the founder of HubSpot, Dharmesh Shah, is the person who sold that domain name to OpenAI. I don't know if it's explicit explicitly been disclosed how much he sold it to OpenAI for, but the rumors are that it was around 15 million and I believe that was sold in equity. So that Dharmesh actually holds equity in OpenAI as a result of this chat.com sale. SAS Junction down here points out that OpenAI now owns both AI.com and chat.com and sure enough, if you go to AI.com it also redirects to ChatGPT. OpenAI announced predicted outputs this week, which dramatically decreases the latency for GPT4O and GPT4O mini. If I'm being honest, this is slightly over my head. But we can see in the comparison video here that on the left we have GPT4O and GPT4O with the new predicted outputs, and we can see how much quicker it is actually giving the output. Over here on the right side it's already done, while the one on the left is still going. But people much more technical than me might be able to better describe what is going on here in the comments. The company Hume just released a new app app. If you're unfamiliar with Hume, I talked about it a few months ago, but it is a sort of AI voice discussion app that better understands emotions. So it could kind of tell when you're happy or sad or angry and infer your emotions based on your tone and how you're speaking. Well, this new version has brand new assistants that combine voices and personalities generated by their speech language model EVI2, and you can now use supplemental models like the new Claude 3.5 haiku. So if I head over to App Hume AI, you can see there's some various options to play around with Hume. Here we've got quick answers, fast question answering, life advice, storytelling, deeper questions, spirituality, and emotional reflection. So let's go ahead and do life advice here and as soon as I click on this, it's going to start listening to me and I can have a conversation with it. I really want to make a lot more YouTube videos, but lately I've been having some struggle figuring out what to make them about. It seems like the news hasn't been as exciting as it once was, and I'm really trying to figure out how to create quality content that people get excited about while not just putting out content for the sake of putting out content. How should I go about this?

My dear? Quality of quantity has always been the hallmark of excellence. Content creation, much like hosting the Perfect Dinner Party, requires finesse and careful consideration. Perhaps we might shift our perspective a touch. My dear, Content creation is rather like hosting the Perfect Dinner Party party. One mustn't simply throw ingredients together and hope for the best.

And after about a minute of the reply, it actually wants me to log in to keep the conversation going. But you get the idea. Microsoft had some new updates this week as well, with new experiences in Paint and Notepad. Paint now has generative fill, just like Photoshop has generative fill. You can now do it directly inside of Microsoft Paint, select an area inside of Microsoft Paint, give it a prompt of what you want to see, and watch as it generates it straight into your image. It's also got generative Erase now. So if you have an image in Microsoft, you can mask stuff out, click Apply, and watch as that portion disappears from your image. And they've added AI into the basic Notepad app now. So if you want it to help you extend out some text or reformat some text, or change the tone of the text, that's going to be built directly into Notepad now. You can get access to this right now if you want to install the newest, latest, most beta version of Windows on your computer. I did that at one time and it actually caused my computer to crash quite often, so I reverted back to like the more stable version of Windows. But if you want to try it, it's available. I'm just personally going to wait for it to roll out to the more stable version of Windows before I play with it too much myself. All right, handful of more small things that I want to share that are, you know, somewhat interesting, probably more interesting to some than others, but let's just break it down in rapid fire fashion here. Prime Video is going to let you summon AI to recap what you're watching. The new tool called X Ray Recaps can create text summaries of full seasons of TV shows, single episodes, or even pie episodes. We can see what that might look like in this screenshot here on the blog post on the Verge. And personally I think that might be kind of helpful because how often have you watched a show and then like two years go by before the next season comes out and then you start watching it and you're like, I have no idea what's been happening. Usually they just like make a recap video for you and that's way more valuable. But if you just want the TLDR, this seems like it'll be pretty useful. Apple's iOS 18.2 started rolling out in public beta this week with even more of the AI features available. This new version is going to have Chat GPT Visual Intelligence and their image playground. However, be ready, you probably are going to get a limited version of ChatGPT. Here's a leaked screenshot that actually shows here, let me zoom in a little bit here. That inside of the settings you could turn on using ChatGPT with Siri, but it also gives you like an upsell to upgrade to ChatGPT plus to get five times more messages and higher limits than more natural real time conversation. So just by owning an iPhone with iOS 18.2, you're not going to get access to like the full chatgpt plus functionality. You're just going to get like the same free functionality that you get, and they're still going to encourage you to upgrade for 20 bucks a month to get everything. This is a really cool tool that a lot of people have been talking about, and I waited far too long to try it out myself. But it actually is really, really impressive. It's a tool called Bolt New and you give it a single prompt and it will build whatever you prompt it with. Now, I'm not going to go too deep in this video. I think this tool deserves its own sort of dedicated video to see what it's really capable of. But real quick, if I head over to Bolt New, real simple. What do you want to build? I'll give it the prompt. Build me a Tetris game that I can play in my browser. You can see it just goes to work on coding, so hands free. Right now I'm not doing anything. It's just writing all the code for me. You can see it's created a whole bunch of different files already and it's just going to go to work and populate all of these files and build the app that I asked it to build. We can see here that the reply to my prompt was, I'll create a beautiful fully featured Tetris game with modern visuals, scoring levels and next piece preview. So it obviously already knows what Tetris is, knows what to make, and it's just going and doing it right now. And it's done. It took maybe a minute and a half and we've got what looks like a Tetris board here. I've created a modern, fully featured Tetris game with smooth animations and modern UI score tracking and levels, game pause functionality, et cetera, et cetera. Controls. Arrow up, arrow left, right, arrow down. Let's go ahead and click Start game. Well, it's got a good user interface, but the game doesn't quite work work yet. Let's go ahead and let it know. So I said the UI appears, but no Tetris blocks are showing up in the game. Let me fix the issue with Tetris blocks not showing up. The problem is the styling of the cells. We need to properly set the RGB colors. It went and did that. And look at this. I actually have what looks like a really decent Tetris game. So there's my square piece. Put it down there. Now it's putting another piece. If I hit up, it actually rotates the piece for me. Now it's not actually showing the next piece for me like it said it would, but let's go ahead and place some of these pie here. I'm going to quickly put a bunch of pieces down because I want to see if it actually works if I get a line to fill. All right. So I'm going to go ahead and drop this last piece. Let's see if it actually notices. I.