This has literally been the wildest week in AI news this entire year so far, and of course, it happened on a week where I was traveling to London for some meetings. I am just now getting caught up with it all; my head is spinning, but I think I gathered all of

the important stuff. Now, there was so much news this week that I couldn't fit it all into one video, but if you check out the news page over on Future Tools, I have listed every single thing that happened this week in the AI world. There

was a ton, but I figured I'd pare this down to just the stuff I thought you would find most interesting, starting with the news that the AI world is probably talking about the most: OpenAI's 12 days of announcements. So, on December 4th, Sam

Altman went to X to say, "Starting tomorrow at 10 a.m. Pacific, we're doing 12 days of OpenAI. Each weekday we will have a live stream with a launch or demo; some big ones and some stocking stuffers. We've got great stuff to share; hope you

enjoy! Merry Christmas!" Now, once the full 12 days are over, I'm going to make another video breaking down the entire 12 days of OpenAI. But so far, we know the first two things, and the first announcement was ChatGPT Pro. This is a

new $200 per month plan for ChatGPT. However, the even bigger news than this $200 plan is that they now officially released OpenAI's API 001. Prior to December 5th, we only had access to the 001 preview model, but now they've opened up the full

model of 01 as well as a new 01 Pro mode, and when we look at their benchmark charts here in math, you can see the 01 Pro mode outperformed the 01 preview that we've had access to by quite a bit.  Competition-wise, the new 01 models are

pretty neck and neck, but you can see them compared to the 01 preview model that we've already had available; it slightly outperforms in PhD-level science questions. And of course, I'm a sucker for anything AI, so I signed up for it and I have access to this 01 Pro

mode. I will be doing a deeper dive video on this next week. Now, if you're thinking, "I don't want to pay $200 to get access to this new 01 model," well, don't worry; the $200 plan is the 01 Pro mode, as we

saw in those graphs a second ago. The new 01 is actually still really, really good, and Sam Altman took to Twitter for extra clarity: 01 is available in our Plus tier, the $20-a-month plan. The new Pro tier can think even harder on the

hardest problems, but most users are going to be happy with the 01 inside of the $20-a-month Plus tier. So for most people, if you are paying the $20 a month for a Plus plan, you should now see access to this 01 model instead of the

01 preview model, which is a really, really solid model. So jump in and start playing around with it because it is a lot better at reasoning than it used to be. Now, full disclosure: I got in super late last night from London and I haven't tested it a ton myself. I did

upgrade, but I haven't played with it yet. But I will be putting it through its paces, testing it a little bit more and then making a follow-up video once I have a little bit better idea of how well it does. We also know what they

announced on the second day, because that was today, the day that I'm recording this video, December 6th. They announced OpenAI's reinforcement fine-tuning research program, and this is really more of an announcement for developers and machine learning engineers because it's designed

to help fine-tune the models for specific niche tasks. Essentially, if you have a specific niche task that you need these GPT models to do for you, you can personally reinforce it. This is what I'm gathering from the information I've read

so far. So you can ask it to do these specific niche tasks, give it some feedback on how well it's doing, and the model will improve based on the feedback that you're giving so that it can get better and better at these niche tasks.

Or niche—sorry if I'm saying it wrong, I don't know. Now, this is by application only. I will link to the page in the description below, but you can apply to get access and you can see who they believe should be using it: research

institutes, universities, and enterprises. I'm probably going to apply so that I can play around with it myself, but I don't know—50/50 on whether or not they actually give me access. Again, we'll be breaking down more of the 12 days of

OpenAI announcements next week and the following week. From what I understand, they're not making announcements over the weekend, so pretty much every day next week and the following week we should see some sort of new announcement.

OpenAI. A lot of people are speculating that they're finally going to open up access to Sora. There have been some clues and hints from OpenAI employees, so there's a good chance that maybe day 12 they'll wrap it up with

Sora, like they opened day one with the new GPT-4-03 model, and they're going to wrap it up on the final day with Sora to bookend it with the biggest announcements. But we'll see how it all plays out. There are a few other smaller

announcements that came out of OpenAI this week that I figured I'd tell you about real quick, since we're talking about OpenAI anyway. They struck a deal with Tom's Guide to use some of the content from Tom's Guide, PC Gamer, Tech

Radar, and some other outlets owned by Future. That's not the only new partnership OpenAI made this week; they also partnered with Anduril to advance U.S. artificial intelligence leadership and protect U.S. and Allied Forces. Anduril is a company founded by

Palmer Luckey, the original founder of Oculus, and they are a defense technology company similar to a company like Palantir. It says here the partnership aims to improve the nation's defense systems that protect U.S. and allied military personnel from attacks.

By unmanned drones and other aerial devices. So it sounds like some of the Anduril defense technology is going to use open AI, like LLMs, inside of the technology. Also learned this week that OpenAI hired their first marketing chief; they brought her in from Coinbase.

So basically, expect to see a lot more marketing from OpenAI. Right now, their marketing has really been posting on X and putting out press releases, but who knows? We might see commercials and influencer campaigns and the various other types of marketing we see some of

the other big tech companies doing now. Again, I moved through the OpenAI news quickly because I am going to do more follow-ups on this, but OpenAI wasn't the only company to have a ton of big announcements this week. Google just had

a flood of announcements. One of the things most people are talking about is that if you use Spotify, well, the Spotify Wrapped actually included an AI podcast this week. So Notebook LM sort of ingested all of your listening habits from this year and put them into a podcast so you can

listen to a podcast that talks about what you listen to on Spotify this year. There's also a new Gemini model this week. It seems like every week when I make one of these news videos, there's a new Gemini model that has rolled out. And if we look at the chat bot arena here,

the current model of Gemini, which came out December 6th (again, the day that I'm recording this video) is once again the top of the leaderboard. Now, this leaderboard does not appear to have the new 001 models; we have the 001 preview

The model that we previously had access to down here, the newest model that just came out this week is not listed on this leaderboard. But if you use Google Gemini, their model is apparently the best right now, and I am starting to use the Google Gemini model more and more. I'll talk

about that in some future videos. Google also showed off the new Gemini 2 model, which is a vision-language model which understands images and can do things like generate detailed, contextually relevant captions for images. One cool thing about these Gemini models is that

these are Google's open-source models, so you can actually download these models and run them locally and iterate and build off of them. They're also available on places like Hugging Face and Kaggle. But we're not even close to done with what Google announced this week; they

also announced GenCast, which can predict weather and the risks of extreme conditions. It can actually predict weather forecasts 15 days ahead of the weather actually happening. The further ahead of weather situations people can get, the better prepared people are going to be able to be for these weather

events. Things like hurricanes and typhoons will be able to have much more advanced warning. This is an area of research that's really exciting and could legitimately change the world. But Google also dropped some new announcements around their Pixel phone related to AI, and I'm just going to kind of wrap and

fire them because there's a lot of them. You can customize how Gemini helps you on your phone with the new saved info feature, so it actually remembers things about you so you don't have to repeat yourself all the time. It's able to use more apps, so when you're using your

Gemini assistant can actually run apps for you just by asking. It can also answer calls for you. So check this out: it says if a package is being delivered, you can use this feature to respond to the delivery person by answering their yes or no questions, or asking relevant follow-up questions, all via simple prompts. So the delivery person will be hearing an AI voice, but you could be responding with messages and not actually have to answer the phone. And once it's all done, you can see live transcripts of the conversation.

It can now automatically organize screenshots on your phone, something that I think will be really helpful for me—I take a lot of screenshots. And there's even a new expressive captions feature. So you can see in this screenshot here that it's in all caps when somebody is being more expressive.

So look at this video: the original caption, "They're going to stop him short," and the right caption, "They're going to stop him short," and then it's all in caps and cheers and applause. So you can actually get more context of the excitement happening in the captions.

There are a lot of new features rolling out inside of Pixel and Android. I'll make sure the links are in the description for all the other stuff that I didn't mention in this video.

If you're a Vertex AI customer, you can actually use Google's new PaLM 2 and Imagen 3 models. Now, up until now, I don't know if you've been able to access PaLM 2; I haven't been able to access it before. But if you are a Vertex

Customer, you can now use Google's AI video generator. This is like Google's answer to Sora and all the various video models that are out there, and you can see they're pretty solid. We'll have to do a future video where we can compare Sora

and RunwayML, and Pika, and Veo, and all the various AI video models because there's a lot of them now, and all of them are getting pretty dang good. And here's the news that I found to be the coolest out of Google this week: their new Genie 2

model, which is a large-scale foundation world model. What does that mean? You can actually create playable worlds that are all generated by AI. So, as this robot moves, every single frame was being generated by AI. As the sailboat moves around and as the camera moves around,

it's all being generated in real time as you move around, and it actually remembers. So, unlike some of the other world models we've seen where if you move forward and then turn around and then look back again, it sort of changes the scene; this one actually seems to

have a pretty good idea of where you've been as well. Here are some other examples. They're a little bit small on the screen, but we can see a humanoid robot in the woods, a humanoid robot in ancient Egypt, a first-person view of a

robot on a purple planet, a first-person view of a robot in a loft apartment. I mean, some of these videos are really cool. Like, look at this speedboat moving around on a river here, and a first-person view of, like, somebody walking on a trail. And you can see, like I just mentioned,

The Long Horizon memory, so it will actually remember what it's already seen, and it can do all of this from just a prompt. So you prompt what you want, and it makes these scenes that you can actually move around in. Now, not to be cliché, but this is as bad as it's getting.

To get... think about games in the future, think about like 5 years out from now. Games might all be completely generated like this in the future. Look, this actually even created NPCs that you can walk up to and interact with. Like, this is crazy to me, honestly realistic.

Physics, smoke, gravity. Like this car is about to drive off this cliff, and you're going to see it actually fall off the cliff. So it understands gravity. This is absolutely wild. Somebody walking around in the forest with a flashlight. Look, the torch; see how it's like lighting up the

area where they're walking, and the flashlight is sort of lighting up the area they're pointing it at. It can do realistic reflections. So crazy! Now, this is just research at the moment that they're showing off; it doesn't appear that we have access to it yet, but as

soon as we do, you know, I'm going to get in here and play around with this and really, really put it through its paces. I want to share this post here from my buddy Belaval Sadu. He says, 'Imagine making 2D concept art for a game world,

pressing a button, and suddenly you can walk around in an interactive 3D world. That's what Genie 2 can do: simulate virtual worlds, including the consequences of any action. So if you unlock a door, or jump, or swim...' he goes on.

To say this is promising on two fronts: for game developers, enabling rapid prototyping of interactive experiences straight from concept art; and for AI research, providing unlimited, diverse 3D environments for training and testing AI agents. We actually got something similar out of B Le's World Labs this

week, where you can upload a single photo and it will do something similar—generate a world that you can move around in with spatial memory. We checked out the World Labs website; here we can see some other examples, like these AI-generated images that become worlds that

you can actually navigate in. They even have a little bit of a usable demo. You can use your WASD keys here, and I can move around in this demo. So I'm actually using the keys and moving around, and if I click my mouse, I can look around. So if

I look around this way, you can see I've got like these cobblestone steps and some trees behind me. Look forward again—that well's still there. Look behind me—same exact scene. So it's actually got that memory of what it saw before it. Now,

it's a very, very small area that I can move around in. As I move closer to stuff, you can see it says 'out of bounds' here, but there's a bunch of different demos that we can test here. Here's like a hiking trail or something. I can move

a little bit before it starts to get out of bounds, but again, this is the worst it's ever going to be. There's a bunch of other demos; I don't want to spoil the surprise. You can go to this website here: worlds.ai/blog and play

Around with this stuff yourself now. There's a ton of demos that you can try, so totally free—free to play with. Go check it out; it's really, really cool. Today's video is sponsored by Build Your Store AI, a platform that helps you create an online business with a website

that already has products in it, ready to sell, all with very little upfront investment and within minutes. Starting an online business has traditionally required a lot of setup work and technical knowledge. Build Your Store AI aims to simplify this process by automating store creation, product selection, and order fulfillment services.

Here's how it works: You begin by registering on the Build Your Store AI platform and then selecting your niche. Then you'll set up a Shopify account, which starts at just $1 for the first 3 months. After you install the Build Your Store app, their AI system creates a

custom Shopify store. The platform comes with pre-selected winning products from the niche that you just selected. This allows you to focus on marketing and selling these products immediately without spending time on store building or product research. And the store builder integrates with AutoDS to handle the product sourcing and order

fulfillment of the products that were pre-selected for your store. This automation means that the technical and logistical aspects of running an online store—things like inventory management and shipping—are handled automatically. The entire setup process takes just a few minutes, and there's no cost to try

the Build Your Store AI platform. So whether you're interested in starting a side hustle or building a full e-commerce business, Build Your Store AI offers a streamlined, low-risk way to get started. So click the link in the description to get your free AI-built online store today.  Grok got in on the

AI news this week: Grok is now available to everyone! Before, you had to be a paying member to use the Grok chatbot. Now, you don't need a premium subscription; as long as you have an X account, you can access and use Grok.

So, if I head over to my account here on the desktop, you can see there's now a Grok option. I've had the Grok option because I'm a paying member, but if you are a free X member, you should now also have access to Grok.

Honestly, one of the coolest features of Grok is its image generation capabilities. It uses Stable Diffusion 1.1 Pro, I believe, which is currently one of the best image generators. Ask it to generate some images and see how it does,

because it's pretty dang cool! We're about to see a lot more from X in the AI world, as they just raised an additional $1 billion in funds to fuel their AI ambitions.

Of course, Meta isn't going to sit quietly this week as all these other AI companies are making big announcements. They unveiled a new Llama model as well. They unveiled Llama 3. 70B earlier this year at Meta Connect; they announced Llama 3.1, and now they've released Llama 3.3.

And we can see from this benchmark test the improvement they're getting out of it. It's improved quite a bit at coding, it's improved quite a bit at math, and the cost to use it has stayed exactly the same. Then we have Microsoft; they rolled out their new...

Copilot Vision features are in preview, so you need the preview version of Windows to access them.  But Copilot can now see what's happening on your computer, understand the context, and help answer questions based on what's on your screen.

This was a feature they showed off a while ago, but now they've rolled out access to it. We can see in this demo video that this person is searching for an ugly Christmas sweater and chatting with their computer.

Copilot is giving them feedback on which sweater to buy; they're essentially having a conversation with their computer as the computer sees what's on the screen. There's music in this video, so I've opted not to play the audio, but

that's basically what's happening: Copilot sees what's on their screen, and the user is talking back and forth with it.

Oh, and Microsoft also rolled out a new model this week: a small language model called VALL-E. This model is designed to run locally on your computer, so you're not sending data to the cloud.

It's specifically designed to run things like click-to-do, on-device rewrite, and summarization features in Word and Outlook, and to pre-optimize small language models for developers to utilize.

Amazon had a flurry of announcements this week too, including their new generation of foundational models called Nova. Now, Amazon has invested and partnered with Anthropic, so it sounds like Anthropic's models are what we'll see in Alexa and

a lot of Amazon's technology in the near future, but they're also hedging their bets a little bit by developing their own AI models. You can see here's the suite of models that they've released: the Amazon Nova Micro, a text-only model that delivers the lowest latency

responses at very low cost; Nova Light, a low-cost multimodal model that's lightning fast for image, video, and text inputs; Nova Pro, a highly capable multimodal model with a combination of accuracy, speed, and cost; Nova Premiere, the most capable of Amazon's multimodal models

for complex reasoning and for use as the best teacher for distilling custom models. This one will be available in Q1 of 2025. Nova Canvas, their state-of-the-art image generation model; and Nova Real, their state-of-the-art video generation model. So Amazon appears to

be making all of the kinds of models that we're seeing out of OpenAI and Google and all of these other companies now as well. They have a little example on the website of this image of, like, the ocean here with the prompt 'Dolly forward,'

and you can see the video moves forward and the waves continue to sort of crash and it stays very consistent. They also showed off this sort of ad concept for a pasta dish here, showing off, I guess, their image generation model. I don't believe this is a video; I think it's

just an image, and they're sort of zooming in on the image that looks to be the video generation model, but this is the kind of stuff that apparently we'll be able to generate with these new Amazon models. They also announced Amazon Connect, which is a new generative

AI models can enhance customer service.  So, this sounds like more... if you're selling products on Amazon, you can have an AI model to help with customer service for the things you're selling on Amazon. And, like I mentioned, Amazon and Anthropic are working together; well, it

came out this week that they are trying to build a supercomputer in collaboration with each other. And finally, in other Amazon news, Luma AI has partnered with Amazon as well. They're going to be bringing the Luma video and image generation models to Amazon

Bedrock. They say here this includes the all-new Luma Photon and Photon Flash, setting a new standard for creative and intelligent image generation, and Ray 2, their upcoming, all-new, super-advanced video generation model. These video generation models are going to get even better; there are new models that we're

going to see really soon. We can actually get a sneak peek here on this video that they put out. Now, I'm not going to play the entire video—it's a minute and a half long—but I will link it up again in the description so you can watch the full

video demo that they put out about the new upcoming Luma model that's going to be available on Amazon Bedrock. It's pretty impressive-looking, although not super realistic. I really like the aesthetics of it. And since we're talking about AI video, the company, Huo AI, also

known as MiniMax, introduced Huo 12 V1 Live, which transforms static art into dynamic masterpieces. It's their image-to-video model; it's designed to revolutionize how 2D illustrations come to life with enhanced smoothness and vivid motion. The model lets your

Characters move, speak, and shine like never before. Optimized for stability and subtle expression, it supports a wide range of artistic styles with unparalleled fluidity and finesse. And we can see in this demo that's been playing here the types of videos this is capable

of doing. And if you want to try out these models, you can do it over at helluo.ai. Videoo, the company Tencent out of China, also released a new AI video model called Hon Video. I have no idea if I'm pronouncing that right; I apologize

if I'm not, but this one's actually open source. It's openly available to use, so you can download it, you can iterate, you can build off of it. It's barely likely uncensored. I'm not sure; I haven't played with it yet. Again, I've been in London

all week, but this is an exciting one to play with, and we'll have to definitely compare it with other video models in the future. But we can see it is available on GitHub and Hugging Face, and we have the option to try it now directly from their website, although it

does appear that you need to have a phone number that starts in +86 to be able to register on the site. But if you know what you're doing, you should be able to download it from GitHub or Hugging Face and use it locally—something we'll have to experiment with

in the future. Leonardo.AI rolled out a feature that I've been super excited about. This is one that I've heard about for a while now, and I have not been allowed to talk about, but I am super excited about it. It's called Flow State.

Now I am an advisor for Leonardo, so take that into account, but I really do think this feature is amazing. Let's check it out. This Flow State feature is designed to help you have an idea, but also let the AI be like a creative collaborator with you. So if I wanted to give it a prompt like,

a wolf howling at the moon, I can generate that initial image, and you can see that it generated eight images. But we can see at the bottom it says, "Scroll to generate more." If I don't really like any of these images—it didn't really give me what I was looking for—I simply

just keep scrolling, and it continues to generate more images for me as I scroll. This is just wild! So you keep going until you find one that's got the sort of aesthetic and look that you like. And then once you find one that's in the

style you like—let's say I like this style here, but that image isn't exactly what I like—I'm going to click "More like this," and guess what? It generates a whole bunch more similar to that. And now I can keep on scrolling as it generates new

AI News: Tech Titans Drop Year-End Surprises

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

Matt Wolfe

OpenAI's 12 Days of Announcements and ChatGPT Pro

The week had significant AI news, especially concerning OpenAI's 12-day announcement series.

The presenter couldn't cover all the news in one video, but details are on Future Tools' news page.

OpenAI's announcements started December 4th with a tweet from Sam Altman.

The first announcement was ChatGPT Pro, a $200/month plan.

OpenAI's API 001 was officially released, including a new 001 Pro mode.

The $200 plan is for the 001 Pro mode; the standard 001 is available in the $20/month Plus tier.

OpenAI's Reinforcement Fine-Tuning Research Program and Other Partnerships

On December 6th, OpenAI announced a reinforcement fine-tuning research program for developers.

Access is by application only, primarily for research institutes, universities, and enterprises.

OpenAI partnered with Tom's Guide and other Future-owned outlets for content usage.

OpenAI also partnered with Anduril, a defense technology company, to advance AI in defense systems.

OpenAI hired its first marketing chief from Coinbase, suggesting increased marketing efforts.

Google's AI Announcements: Gemini, GenCast, and Pixel Features

Spotify Wrapped included an AI-generated podcast based on user listening habits.

A new Gemini model was released, topping the chatbot leaderboard (excluding the new 001 models).

Google showcased Gemini 2, a vision-language model for image understanding and caption generation.

Gemini models are open-source, available on Hugging Face and Kaggle.

GenCast, a weather prediction model, can forecast extreme conditions up to 15 days in advance.

New Pixel phone AI features include customized Gemini assistance, app control via voice, call answering, and screenshot organization.

Google's PaLM 2, Imagen 3, and Genie 2 Models

Vertex AI customers can access Google's PaLM 2 and Imagen 3 models.

Google's AI video generator (comparable to Sora) is now available.

Genie 2, a large-scale foundation world model, allows creation of AI-generated playable worlds.

Genie 2 demonstrates realistic physics, memory, and NPC interaction.

B Le's World Labs and Other AI Developments

B Le's World Labs allows users to upload a single photo to generate a navigable 3D world.

Build Your Store AI helps create online businesses with minimal upfront investment.

Grok is now available to all X account users, not just paying members.

Meta released Llama 3.3, showing improvements in coding and math while maintaining the same cost.

Microsoft's Copilot Vision (Windows preview) allows interaction with on-screen content.

Microsoft released VALL-E, a small language model for local computer use.

Amazon's Nova Models, Partnerships, and Luma AI Integration

Amazon announced its Nova generation of foundational models, including multimodal and image/video generation models.

Amazon and Anthropic are collaborating on building a supercomputer.

Luma AI's video and image generation models are integrating with Amazon Bedrock.

Huo AI (MiniMax) introduced Huo 12 V1 Live, an image-to-video model.

Tencent released Hon Video, an open-source AI video model.

Leonardo.AI's Flow State Feature and Other AI Audio News

Leonardo.AI launched Flow State, a feature allowing iterative image generation by scrolling and selecting preferred styles.

ElevenLabs expanded its podcast feature to the web app and enabled downloads.

ElevenLabs announced conversational AI features, allowing users to create custom chatbots.

Hume AI introduced new voice controls for developers, offering fine-grained control over voice dimensions.

Legal Disputes, New Appointments, and Other AI News

Perplexity added publishers to its program, enabling content monetization.

Elon Musk filed an injunction against OpenAI's transition to a for-profit company.

Donald Trump announced David Sachs as his AI and crypto czar.

Humane is releasing its operating system for integration into cars and other devices.

A new, highly realistic android was showcased by Clone Robotics.