It's been another busy week in the world of AI, with Google and OpenAI competing to make the biggest announcements. It's been really fun to watch, so let's dig right in, starting with Google's announcements this week.

Probably the biggest thing everyone's been talking about this week has been Google's Imagen Video AI model. I managed to get early access and made an entire breakdown, testing prompts you guys gave me on X. You can find the

entire breakdown video right here on my YouTube channel. It's called, "This New AI Video Tool Is the Best I've Seen: Hands-On with Imagen". That title really says it all; this is the best video model I've seen so far. It makes really

realistic videos; it seems to understand physics better, and people's movement is much better. Everything about it is really good. Every prompt generates four videos, so even if one is a bit wonky, you

still have three more to choose from, making it more likely you'll get a video you like. Now, I don't know exactly when this will be publicly rolled out, but they have a waitlist at labs.google.com/tools/imagen-video. 

If you get on the waitlist, supposedly they're rolling it out to the general public fairly soon. They also have a text-to-image-to-video model, which first generates an image and then converts that image into a video, but at

The moment there's no way to upload your own image and convert it to a video again. Make sure you check out my deeper dive video all about v2. Google also made some updates to their Imagen 3 model; it's now generating apparently brighter,

better composed images, and from what I've seen, all of the images are pretty dang good. And this one, most people should have access to it if you go over to labs.google and you select ImageFX. This is using the newest Imagen 3

model, and just like before, it will generate four images with every single prompt, and they're pretty dang good. I mean, in my mind, I feel like AI image generation is a fairly solved problem. You can pretty much generate anything you can imagine in image form now, and all of these

platforms are just adding minor details that make them slightly more improved. But for the most part, I feel like we've hit that point where you can kind of generate anything you want; it might just take a few rerolls to get there. Google also showed off a new platform called

Whimsy, which is this really cool way of blending multiple images together to create a cool new image. This one can be found over at labs.google.com/tools/whimsy, and you might have seen some of these viral examples floating around on social media where people are making plushy versions of

themselves. You can give it a style, like this plushy, toss an image of yourself in, like I just did with this image, and it spits out a plushy version of the image. If I click open the tool, you can see the styles, your own scenes, and of course, the

subject, or you can roll the dice and let it do it for you. So let's have it do like a pin with whatever this scene is here, and we get this cool blended pin that has me with this background, like hosting a podcast. It's really fun to

play with, and it's totally free to use right now. Again, it's called Whisk, and you can find it over in Google Labs. Labs. Google Notebook LM also got some major updates this week. If you're not familiar with Notebook LM, it's the

platform where you can upload documents and then chat with those documents. But the feature that everybody's really talking about is the feature where you can actually have it turn your documents into a podcast. Well, Notebook LM got this redesigned this week, so it's a

completely overhauled, kind of cleaner layout, but they also added this interactive mode. So I uploaded a transcript from a podcast with Dario Amodei. Let's go ahead and have it generate a podcast, and now when I'm listening back to this podcast, I can go into

interactive mode here. Okay, ready to dive in? Today we're tackling AI—oh yeah, specifically, uh, how fast these systems are scaling up. It's incredible, and how one company, Anthropic—oh hey, welcome! What's up?

Hey, I'm just curious, can we just kind of get right into it? Tell me who this Dario Amodei guy is. Oh hey there, yeah, absolutely. Great question, I'm glad you asked. So Dario Amodei, he's the CEO of Anthropic, this AI company we've been talking about, and he's kind of a big deal in the AI world. Right, and what's

Interesting is that he's not just, oh, go for it. Yeah, yeah. What what makes him a big deal? I don't really want to get into all the rambly stuff. I'm just kind of curious, why is he a big deal? What was he doing before Anthropic? Okay, so you want

the quick version on why he's a big deal? Got it. Yeah, let's uncheck this. Basically, he's at the forefront of AI development and safety. Before Anthropic, he was deeply involved in scaling laws, which we'll get into, but basically means understanding how these systems get

smarter. So basically, now you can create these podcasts using any content you want. You can upload text files or PDFs or pull in YouTube links or whatever you want, and then while the podcast is going, you click the join button; you can interrupt them and you can ask them

questions about the document. It is super, super cool, and this is now available to everybody. They've rolled it out into everybody's Notebook LM accounts for free. But you know what? Google just kept shipping this week; they had so many cool updates. They also rolled out a new 2.0

advanced model inside of their Gemini Advanced platform. So in order to have Gemini Advanced, you do have to pay for the Gemini upgrade on your Google account, but you can now use their most advanced large language model. So if I head over to gemini.com, since I'm a

Gemini Advanced member, I do have a dropdown here where you can see multiple options, including 2.0 experimental Advanced. And if we take a quick look over at the Chatbot Arena here, where they compare the various large language models, we can see the Gemini

The experimental model is the top-ranking model right now. I'm guessing this is what we're getting when we play around with the new Gemini Advanced 2.0 model. Notice which model is right here, almost tied for first place: this Gemini 2.0.

Flash Thinking. Guess when that one rolled out? That one came out this week as well. Here's Logan Kilpatrick from Google saying, "Just when you thought it was over, we're introducing Gemini 2.0 Flash Thinking, a new experimental model that unlocks stronger reasoning capabilities and shows its thoughts."

This is apparently Google's answer to OpenAI's GPT model. You can actually test this one out over in Google's AI Studio. If you go to ai.studio.google.com, over on the right you have the option to choose a model here, and you'll notice

that Gemini 2.0 Flash Thinking experimental is one of the options available, and as of right now, it appears to be free to use. Let's give it the old strawberry test: How many Rs are in the word 'strawberry'? We can see that it

thought through... you can actually expand the model's thoughts here. Identify the target word: the word is 'strawberry'. Focus on the target letter 'r'. Examine the word letter by letter: 's' not an 'r', 't' not an 'r', 'r' yes, count equals 1. And you can see it go

through and actually count the Rs in the word 'strawberry,' finally summing it up: 1 + 1 + 1 = 3, and then stating the answer: 'Strawberry: There are three Rs in the word strawberry.' Obviously, this isn't the most advanced test, but anybody that has

Some really complex logic problems or math. I highly recommend checking it out in Google AI Studio and seeing how you think it compares to OpenAI's GPT model. And while we're on the topic of Google, obviously Google-owned YouTube just rolled out a new feature; a lot of people

are concerned about having their content that they put on the web being trained for AI purposes. Well, now YouTube makes that opt-in. So if you are okay with AI training on your YouTube videos, you can check a box inside of YouTube now to allow third-party companies to train AI

models using your channel content. I talk about AI; I'm a proponent of AI, so of course I said, "Sure, use my videos to train AI." You could uncheck this box, or you can select only the parties that you're willing to let train. So if I only

want OpenAI to be able to train on my videos, I can select just OpenAI. But again, I'm totally cool with them training on my content, but totally understand the reasons why some people wouldn't be okay with that. Now, the interesting thing about all of these

things that Google has been releasing is it feels like it's been a day-by-day competition with OpenAI because OpenAI originally announced their 12 days of OpenAI announcements, where every single day for 12 days they would make some sort of new announcement or some sort of

new rollout. Now, the day you're watching this video is the 12th day of OpenAI, and they've probably announced something new, and it's probably something pretty big. I would imagine whatever they announce on the very last day of the 12 days is going to be one of the bigger

announcements that they're going to make, but I could be totally wrong. For all I know, on the day you're watching this, they released that you can now send fax messages to OpenAI. I don't know, but it's probably going to be something a little bit bigger than that.

But here are the five announcements they've announced since I made my last AI news breakdown. On day seven, they announced projects inside of ChatGPT, which is essentially folders. It's like the same thing as projects inside of Claude, if you've ever used Claude's

projects. If I go to my ChatGPT account, you can see over on projects, I've got a test project. I can create a new project. I'm going to call it Test 2 project. You can tell I haven't really been using this much yet, and this project can have

its own custom instructions and its own set of files that it uses every single time you prompt it with something. So if you want to make a project that is specifically designed just to answer questions about a certain product's documentation, you can upload all the

documentation into the files, and then anytime you prompt something, it will use that documentation to help with the response. On the eighth day of OpenAI's announcements, they announced that the search functionality is being released for everybody, no matter which ChatGPT plan you're on. Now, if you're a

Pro or Plus user, you've had this for a little while now, but if you were a free user of ChatGPT, you now have this as well. So let me jump into the ChatGPT 4 model. You can see the "Search the web" button. I can select this, and anytime I

Ask it a question now. It's going to search the web for the answer. For instance, if I say, "What does Matt Wolfe's YouTube channel talk about?" it will do a search and answer the question. Here, and funny enough, it actually pulled up a YouTube video that I did on the Hustle...

Flowchart podcast with Joe Fear, all about what I do on my YouTube channel. On the 9th day, they announced that the 001 preview model—the model that's really good at thinking and reasoning—is now available in the API. So if you're a

developer, before you were only able to use the GPT-4 model. Now they actually gave you access to 001 to use that in whatever software tools you're building. If you go to OpenAI's playground, over on the right where you select your model,

you now have access to 001 mini and 001 preview as part of the API. On day 10, they announced 1-800-ChatGPT, which is essentially a way to use your phone and call in and have a conversation with ChatGPT. I saw a lot of people saying

that this is really good for people outside of the US that maybe don't have internet access; they could call in and use this. But if you look at the fine print on the video right here, it actually says, "US only for phone calls." Yeah, I don't really get the thinking

here, but it's essentially the advanced voice mode that, if you're in the US, you can call in and use. Here's what it sounds like: "Hello again, it's ChatGPT, an AI assistant. Our conversation may be reviewed for safety. How can I help you?"

Is your refrigerator running? Yes, it's running away! So I better go catch it. Right? Classic! Hey, that was my line! You set it up so perfectly, I couldn't resist. What's next? Anyway, you can ask it whatever you would ask ChatGPT, voice mode. You just call 1-800-ChatGPT.

They even showed using it on a rotary phone! If you own a rotary phone... And on day 11, the day that I'm recording this, they announced that the Mac app for ChatGPT now works with more tools. Now I'm not on a Mac right now, so I can't

show you the new tools that it works with, but basically you can give it access to different tools on your Mac. They did say that Windows is going to get these features soon, but right now it can only view certain tools on your Mac, which I find funny because their

biggest partner is Microsoft. You'd think they'd prioritize PC and Windows users first, but no, they prioritize Mac users first. In my opinion, compared to the previous couple of weeks of OpenAI announcements, none of these were that great, but I really have high hopes for

day 12, Friday, the day this video is going live. I do think there's going to be some sort of big announcement. I don't know what it's going to be yet, so I'm excited for that because this week's announcements were kind of on the weak

side, if you ask my opinion. I mean, they're all cool, they're all great advancements, but nothing that really got me excited like some of the Google stuff that rolled out: Voq, Whisk, Notebook LM updates, Gemini with thinking, Gemini Advanced. I mean, Google crushed it this

This week and last week, with their announcements... While in the past, OpenAI usually kind of overshadowed Google's announcements, the last two weeks I feel like Google crushed OpenAI with its announcements. It's just really fun to watch. I mean, I posted a post on X and

I said, "Who's winning this week: Google or OpenAI?" and pretty much everybody said Google's winning. But in reality, the consumers are winning because we're watching this battle play out. And because this battle's playing out, we're getting cooler and cooler tools by the day. You

know, there are a lot of great AI video tools out there right now; tools like Sora and Runway. They're all very promising, but they only get you a few-second clips and it doesn't come with audio. They're not really ideal for

making longer video content. That's why, for today's video, I partnered with Invid AI. It's a single tool that will generate scenes for you, compile them into a full video, add audio for you, and basically do the entire process of creating a short video with AI, all in a single

platform. So check this out: With NVIDIA's new version 3.0, you can give it a prompt of whatever video you want, and it will create the entire thing for you. I'm talking: generate the script for you, generate all of the videos inside of

that script for you, add sound effects, add music, and even do the voiceover in your own voice if you want. They have simple templates that make it really easy, like: create a short video; you just follow the instructions here. An explainer video, animated films, and a ton

of other pre-built workflows that you can choose from, so you're not starting from scratch at all. And, of course, you can use your own script if you've already got something written up. Let's say you need an ad for your latest course about how to learn piano. Let's go

ahead and use this template for an animated ad. Here, we'll tell it to create a 30-second animated advertisement for YouTube about a beginner's course on learning to play the piano. Now, if you want, you can select longer videos; you can select different

platforms; you can select to use only generated clips, only generated images, or only stock footage. Let's go ahead and use generated clips, and we can set the style. Let's do it in a Disney Pixar style. Now, I can go through and select this type of background music, which

language I want it to be in, whether I wanted to have subtitles, pick specific voice actors, add a watermark, select my music preference, or I could just let it figure all that stuff out for me. So let's see what it comes up with for my beginner piano course. I'll click

continue here, and we can see it pre-generated a prompt for me: 'Create a 30-second animated advertisement for YouTube about a beginner's course on learning to play the piano. Use only generated clips. Use a Disney Pixar style.' Let's leave it that simple and generate the video. Now, this does take a little

bit of time because it is writing the script, generating all of the clips for it, adding the voiceover, adding the background music; it's doing all of that behind the scenes from that single prompt. And after about 15 minutes of generating, here's the video ad that it

Ever dreamed of playing the piano? With Matt's online beginner piano course, that dream is just a keystroke away! Our fun lessons make learning easy; play your first song in just one week! Join thousands who've discovered their inner musician. Enroll now at MattsBeginnerPiano.com and unlock the melody within you!

I really like the generation; the video of it was really good, but the music was almost like this weird, creepy music. So, if we want to change that, I can just give it a prompt like, "Change the background music to a more upbeat, happy piano song." We'll click "Generate" again, and within seconds we have a new version—much better!

But you know what? I actually have my voice trained into InVideo as well. Let's change the narration to my voice. I'll just give it the prompt, "Change the narration to Matt's voice," and click "Generate." About a minute later, I've got another variation: "Ever dreamed of playing the piano? With Matt's online beginner piano course, that dream is just a keystroke away!" Much better!

And this is just scratching the surface of what I can do with InVideo. I can give it a prompt to change pretty much any element I want about the video, or I can click on "Edit," regenerate any single scene from my video, replace it with stock footage, upload my own video, upload my own music,

or choose music from their stock music library, change anything about the script and who's actually narrating the script, and so much more. Unlike most of the other video platforms that are available right now, you can control pretty much every single element of the output of your video with InVideo AI. I can even

Change the whole video to another language if I want and have my ad running to an entirely different demographic. So if you want to try one of the lowest-barrier-to-entry video tools and one of the most customizable video tools on the market right now, head on over to

nido.org. This video, if you're a coder, we got a big announcement out of GitHub this week: GitHub Copilot now has a free tier that you can use directly inside VS Code. If we take a look at the plans here on the GitHub

Copilot page, their free plan allows up to 50 uses per month, and their Pro plan, at only $10 per month, allows unlimited uses.  Context cursor is $20 a month, and Devon is $500 per month to use, as far as code completions you can get

up to 2,000 per month for free, or unlimited on their $10 a month plan. So this is big news for coders who are on an extreme budget and only want to use free tools. And since we're talking about GitHub, which is owned by Microsoft, let's

talk about a couple of things that Microsoft rolled out this week, including real-time translation inside of live captions. Now you do have to have a Copilot-capable PC for this to work, but you can essentially watch videos in any language and it will subtitle it, and it

even works with video calls. You can be talking on a Zoom call with somebody else and it will translate in real time what they're saying directly on your screen. Bing Image Creator also got an update this week using the latest DALL-E model, called Model pr16. It allows users to create images

twice as fast, and you can create images directly from within the Bing search bar if you want. They've also made it much easier to share your images once they're generated. But if I go over to bing.com and tell it to create an image of a

monkey on roller skates for me, notice that right at the top, Bing Image Creator starts to create the image. I don't know if it's bogged down or something right now, but every time I'm trying to generate an image, I'm running into an

error. Hopefully, that's something that Microsoft will work out, but supposedly you'll be able to do this right here inside of the Bing search. The Meta Ray-Ban glasses got an update this week as well, with real-time video. So, very similar to what we saw with Project

Astra and even with the new ChatGPT with vision. Now your glasses can actually watch a real-time feed of what's going on around you. It's also going to allow you to have real-time conversations that are translated between languages. So it says Live AI lets wearers

continuously converse with Meta's AI assistant, Meta AI, to reference things they discussed earlier in the conversation. You no longer have to say, "Hey, Meta," to get it to start up. It says Live AI also works with real-time video; wearers can ask questions about what

they're seeing in real time. For example, "What's around my neighborhood?" It's basically OpenAI's advanced mode with vision, or Project Astra, just built into your Ray-Bans. I haven't gotten this update yet in my Ray-Bans, but I am

Excited to test this stuff out! It's also got the ability to Shazam the song; whatever song is playing around you, it'll tell you what song that is. So, some cool new features that are supposedly rolling out now.

About Meta: Instagram started teasing their new AI video editing tool. You know, Meta has an AI video generator, and I was talking about it on a recent podcast with Tim from Theoretically Media, and we couldn't figure out what Meta was going to use this AI video tool for. Well, Adam

Mosseri, the CEO of Instagram, kind of gave us a hint: "We're working on some really exciting AI tools for you video creators out there. A lot of you make amazing content that makes Instagram what it is, and we want to give you more

tools to help realize your ideas. You should be able to do anything you want with your videos; you should be able to change your outfit, or change the context in which you're sitting, or add a chain—whatever you can think of." Now, all that you're seeing here are some of the early

research models that we've been exploring, and we're hoping to bring some of these into the Instagram app next year. So keep an eye out; more to come. All right, continuing on with the topic of AI video, Pika rolled out their new 2.0

model this week, and they have this really cool feature called "Scene Ingredients." If I log into my Pika account, you can see there's a new button here that says "Ingredients," and it allows you to upload multiple scene ingredients, including characters, objects, and locations. It's almost like that WiP

that we just saw from Google, but for AI video, let's say I really like this jacket here. I could pull in an image of myself and then the image of that jacket, and then give it a prompt like, "A man with a beard walking in a blue and

orange jacket." Ideally, I'll get what looks like me walking with this jacket on. And well, I got the guy walking, but it looks to be the original guy; it didn't really swap the face. I do need to play with this Pika a little bit more. My

last attempt at using ingredients also turned out kind of wonky. If you were watching my live stream, you saw me get this result, but I'm sure there'll be people in the comments letting me know what I'm doing wrong and how to actually make this new Pika feature work because

when I see some of these other ones that other people have done, like making themselves fly or putting themselves next to the Girl with the Pearl Earring, they all seem to come out really, really good, which just makes me think that I just need to get better at prompting it.

Oh, and by the way, if you want to try out Pika, they're actually making it free for the next few days. They put out this post on X: "A giant pre-holiday gift from the Pika team! We're giving everyone free, unlimited access to Pika 2.0 from today

until December 22nd. Anyone on any plan can generate as many videos as they want, using all the scene ingredients they want." While it's live, head to pika.org. Pretty cool of Pika. The video model, Runway, also got an update this week. Here on X they said, "We're excited to

introduced the calling AI 1.6 model. With this update, we have significantly improved the response to prompts, the visual aesthetics, and the physical actions, hoping to bring more consistent and vivid results. So this new model has improved prompt adherence more

consistent and dynamic results. It supports standard and professional modes, achieving a 190% overall improvement compared with calling 1.5 and has a 720p model. Now, calling isn't one that I've played around with a ton, but we did

again make a podcast recently with Tim from Theoretically Media, which will be out within the next week or so, where we actually break down all of the various AI video tools and compare them against each other. And we put them all on a tier list, and if I remember correctly, Tim put

Calling in a high tier, so he finds it to be one of the better AI video models. Since we're talking about AI video companies, Runway announced the Runway Talent Network, which is a new media platform where you can discover and hire people that are actively using AI to

AI News: The AI SHOWDOWN We Needed!

AI, No-Code, Tech, Futurism - I'm a tech nerd and talk about tech nerd stuff

Matt Wolfe

Google의 AI 발표: 이미젠 비디오, 이미젠 3, 윔시

이미지 비디오 AI 모델: 얼리 액세스 테스트, 사실적인 비디오, 더 나은 물리 및 움직임.

대기자 명단(labs.google.com/tools/imagen-video)을 통한 이미지 동영상 공개 배포.

텍스트-이미지-비디오 변환 모델: 이미지를 생성한 다음 비디오로 변환하며, 현재 이미지 업로드가 없습니다.

Imagen 3 모델 업데이트: 더 밝고 구도가 잘 잡힌 이미지로, labs.google/ImageFX를 통해 액세스할 수 있습니다.

기발한 플랫폼: 여러 이미지를 혼합하여 새로운 이미지를 생성하는 플랫폼으로, labs.google.com/tools/whimsy에서 이용할 수 있습니다.

노트북 LM 업데이트: 새롭게 디자인된 레이아웃, 문서를 팟캐스트로 변환하는 대화형 모드.

Gemini 고급 2.0 모델: 챗봇 아레나에서 최상위권에 속하는 가장 진보된 대형 언어 모델입니다.

Gemini 2.0 플래시 씽킹: 더 강력한 추론 기능을 갖춘 실험적 모델로, Google AI 스튜디오에서 사용할 수 있습니다.

사용자 콘텐츠를 사용한 AI 학습을 위한 YouTube의 옵트인 기능입니다.

7일: ChatGPT의 프로젝트(사용자 지정 지침 및 파일을 위한 폴더).

8일차: 모든 ChatGPT 사용자를 위한 웹 검색 기능이 출시되었습니다.

9일: API에서 001 프리뷰 모델(강력한 추론)을 사용할 수 있습니다.

10일차: 1-800-ChatGPT(음성 대화, 미국만 해당).

11일: 더 많은 도구가 포함된 ChatGPT용 Mac 앱(Windows 지원 예정).

Google의 최근 발표는 OpenAI보다 우수한 것으로 간주됩니다.

소비자는 경쟁을 통해 혜택을 얻고, 이는 더 나은 AI 도구로 이어집니다.

AI로 장면을 생성하고, 동영상을 편집하고, 오디오를 추가하고, 짧은 동영상을 제작할 수 있습니다.

NVIDIA 버전 3.0: 스크립트, 비디오, 음향 효과, 음악, 보이스오버를 생성합니다.

다양한 동영상 유형(설명 동영상, 애니메이션 영화 등)을 위한 간단한 템플릿을 제공합니다.

예시: 피아노 강좌에 대한 30초 애니메이션 광고, 사용자 지정 가능한 요소(음악, 음성, 스타일).

기타 AI 뉴스: GitHub, Microsoft, Meta 등

GitHub Copilot: 사용이 제한된 무료 티어, 무제한 사용이 가능한 프로 요금제.

Microsoft: 라이브 캡션의 실시간 번역, DALL-E 모델로 Bing Image Creator 업데이트.

메타 레이밴 안경 업데이트: 실시간 동영상, 실시간 번역, 노래 식별.

Instagram의 AI 동영상 편집 도구: 의상, 배경 변경, 요소 추가.

Pika 2.0: 동영상에서 여러 이미지를 결합할 수 있는 장면 성분 기능.

런웨이의 AI 1.6 모델: 개선된 프롬프트, 시각적 효과 및 물리적 동작.

런웨이 탤런트 네트워크: AI 아티스트와 동영상 크리에이터를 발굴하고 고용하기 위한 플랫폼입니다.

오딧세이의 제너레이티브 월드 모델: 프롬프트에서 3D 공간과 동영상을 제작합니다.

제네시스 프로젝트: 4D 다이내믹 월드 제작을 위한 제너레이티브 물리 엔진으로, 로봇 훈련에 활용할 수 있습니다.

레오나르도 AI 업데이트: Phoenix 1.0 모델, Flux 모델 통합.

중간 여정 무드 보드: 특정 스타일에 대한 중간 여정 교육.

매그니피크의 슈퍼리얼 AI 이미지 모델: 초현실적인 이미지, 편집 기능.

11 Labs의 플래시 텍스트 음성 변환 모델: 7.5밀리초의 생성 시간.

플로리다 애틀랜틱 대학교의 수화 판독용 AI 시스템.

AI 기반 휴머노이드 로봇을 위한 아트로닉과 구글 딥마인드 파트너십.

로봇 공학을 위한 Nvidia의 Jetson Nano 미니어처 슈퍼컴퓨터.