AI News: 28 Headlines No One Expected

So, you'd think every AI company out there would be gearing up to take a nice long relaxing holiday break, but no.

Instead, this week, they chose violence.

There was so much news this week, and I want to make sure that you are completely looped in on everything that's going on. So, hold on to your papers, fellow scholars, because this is going to be one of the biggest rapid

fire news videos I've ever done.

Starting with the brand new image model out of OpenAI. This week they launched a GPT image 1.5 both inside chat GPT and in the API for developers in order to

compete with the state-of-the-art model that Google released called Nano Banana Pro. Now, this was probably the biggest

Pro. Now, this was probably the biggest news of the week, but I actually already did a full breakdown video of this news along with sidebyside comparisons of

both Nano Banana and the GPT image 1.5 model. And since we have so much other

model. And since we have so much other stuff that I want to cover this week, I'm not going to actually dive into that model. Instead, I'm going to point you

model. Instead, I'm going to point you to my existing video to get all of those details. But that wasn't the only new

details. But that wasn't the only new image editing model that we got this week. Black Forest Labs also released a

week. Black Forest Labs also released a brand new model called Flux 2 Max this week. And this is one that I actually

week. And this is one that I actually haven't tested yet. So, let's try it out. This Flux 2 Max model also appears

out. This Flux 2 Max model also appears to be a model trying to compete with Nano Banana and the new JPT image 1.5 because it is a model that is designed

for not only generating images but also editing images. Some of the examples on

editing images. Some of the examples on their website show it putting logos on products like this boarding pass coffee here. Doing iterative editing where you

here. Doing iterative editing where you start with one image, you add that image with another image, add it with another image and you can keep on iterating and

it remembers the original context.

Something that when I tested with Nano Banana, it struggled with, but GPT 5.1 actually did pretty decent with. It also

has grounded image generation similar to Nano Banana where it looks like it'll actually research what's supposed to go in the image and then add what it researched in for you and it understands

how to change the style of your images to various types of other styles. And

there's a lot more examples on the Flux page. I figured I'd give it a test with

page. I figured I'd give it a test with a few of the same prompts that I tested Nano Banana and GPT 1.5 on. So, for the first one, I gave it this image, which

is the same image I gave the other models in my previous video. And I gave it the prompt, using the uploaded photo of me, remove the person standing on my right while keeping my face, pose, lighting, and background exactly the

same. I'm the one in the Padres's shirt.

same. I'm the one in the Padres's shirt.

Change my outfit to a black leather jacket and add a subtle neon rim lighting behind me. Do not change my facial structure or expression. And

well, yeah, in this one it sort of made like a hybrid version of me and Joe and it removed me from the right. So the one thing I asked it to do is remove the

person to the right of me and it removed me and left the person to the right and kind of made like a love child between the two or something. And pretty much

none of the other outputs did a whole lot better at following my instructions.

Just as a quick reminder, here's what chat GPT did. It put the purple glow around me, left me in the same spot, removed the people to the right of me.

It mostly followed the directions. Flux,

on the other hand, same image, same prompt, not quite the same result. This

was an interesting test that I did with the other models as well. Create a white canvas divided into nine uneven rectangles like a magazine layout. Top

left, steaming cup of coffee. Top right,

folded city map. Center left, pair of headphones. Center right, glowing light

headphones. Center right, glowing light bulb. Bottom center, red notebook. Each

bulb. Bottom center, red notebook. Each

object must stay fully inside its rectangle and not overlap any borders.

So, we can already see here that it screwed up the nine uneven rectangles because I'm only seeing five rectangles.

Maybe you can argue six rectangles if you consider this one here. So, got the number wrong already. Top left, steaming cup of coffee. Yeah. Top right, folded city map. This looks like an unfolded

city map. This looks like an unfolded map. Center left pair of headphones.

map. Center left pair of headphones.

Center right glowing light bulb. Bottom

center, red notebook. No overlaps. Just

as a reminder, here's what OpenAI's model did. I asked it for nine uneven

model did. I asked it for nine uneven rectangles. It gave me 10 for whatever

rectangles. It gave me 10 for whatever reason. It put everything in the right

reason. It put everything in the right spot, but it also ignored my last instruction. Each object must stay fully

instruction. Each object must stay fully inside the rectangle. Cuz as we can see, the notebook, it was like the model was testing me a little bit. It was like he said stay inside the lines. Let's put it

like on the line. See if that works. So

yeah, the new flux model seems like a pretty okay model. Not quite up to par with Nano Banana or the OpenAI model, but another option to test if those other two models aren't doing it for

you. Now, moving from images to audio,

you. Now, moving from images to audio, Meta rolled out a version of their SAM or segment anything model for audio. So,

if you're familiar with the segment anything models, you give it an image or a video and you can type in a text and say highlight the train or remove the train or add effects to the trains.

Well, it does the same thing with audio now where you could give it an audio file and say just isolate the speaking, just isolate the guitar, things like that. So, I wanted to give it a test and

that. So, I wanted to give it a test and it is available to test for free right now over inside of Meta's Playground.

You can find it at a demos.

And I actually created a song with Sunno. I wanted to make sure I wasn't

Sunno. I wanted to make sure I wasn't using any copyrighted music. So, this is a generated song. Here's a real quick clip of it. Click the button.

>> And we can see over here I can just type what I want it to isolate. So let's

isolate guitars. Now I'll click isolate sound. And we can see that it split it

sound. And we can see that it split it out. So we've got an isolated sound and

out. So we've got an isolated sound and without the isolated sound. So if we listen for just the guitars, let's hear what it sounds like.

So it completely cut out all the vocals and any other instruments that were there. I can also listen to the version

there. I can also listen to the version where it took the guitars out completely. So that's what this sounds

completely. So that's what this sounds like.

>> So we hear the drums kick in and the bass kick in and all the other instruments, but the guitar is gone. We

broke it out into a separate track. I

can also add some effects to this guitar if I want. I don't think they're going to sound great, but I could. Like let's

add a megaphone to the guitar.

Yeah, it didn't really make a difference. Let's add some extra reverb.

difference. Let's add some extra reverb.

Yeah, I'm not really noticing the difference very much on this music.

However, it's also good at isolating other things. So, if I come back to my

other things. So, if I come back to my Meta Playground, click on isolate sounds, they've got some other examples here, like these two people speaking on a podcast where they gave it a video.

Here's what it sounds like by default.

>> Hey, we're here with Phoenix, our local sound editor.

>> Thank you for everything that you do.

Let's have it just isolate male vocals.

We'll isolate that sound and ideally it'll give us just his vocals. And

here's what we get with that.

>> Today we're here with Phoenix, our local sound editor. Thank you for everything

sound editor. Thank you for everything that you do. So tell us a little bit more about the space and how that relates to film, what you've been learning.

>> And we can see it cut out her audio when she starts speaking. That audio would be on the without isolated sound. So now

we'll see that he'll look like he's talking, but no audio is coming out.

And then when we fast forward to where she jumps in, we should hear her jump in.

>> I've learned a lot about the >> So, pretty cool thing, especially if you're like a podcaster or you make music. A handy little tool to go and

music. A handy little tool to go and start isolating individual audio elements. We've been able to do it with

elements. We've been able to do it with images and videos and now Meta has an audio version. Pretty sweet. Vibe Coding

audio version. Pretty sweet. Vibe Coding

AI apps just got a lot easier. This new

AI tool that's literally called Vibe Code lets you build AI powered apps and ship them straight to the app store all on your phone. You just open up the app, pinch the screen, and it drops you into

this cap cut like builder where you can describe the app you want, and Claude Code will build that app right there within the Vibe Code app. You can also generate assets like images, sounds, and haptics and place them into your app.

Heck, you can even add a payw wall so you can make money from your new app.

And when you're done, you can ship it straight to the Apple App Store from your phone with just one tap. It's

really that easy. So, if you want to try it for yourself, you can build your first three apps for free using the code wolf. You can check it out at the link

wolf. You can check it out at the link in the description. And thank you so much to Vibe Code for sponsoring this portion of today's video. All right, so we talked about image models, we talked about audio models. Now, let's talk

about video models because there was a lot of movement in the video model world this week. Let's check it out. Starting

this week. Let's check it out. Starting

with the fact that Adobe Firefly now supports promptbased video editing. So

theoretically, we can edit a video in Firefly and say remove this person, add this sound effect, add this glow around the person, things like that. I don't

know exactly what it's capable of yet. I

haven't tested it, but theoretically, we can give text prompts to edit a video now. So obviously, all video editors are

now. So obviously, all video editors are doomed and will no longer have a job.

They ask you how you are, you just have to say that you're fine.

>> So, let's jump into Adobe Firefly here.

This is another one I have not played with yet. And we can see inside of

with yet. And we can see inside of Firefly, we have edit video beta. Let's

jump into here. And we can create a new project or upload media. So, let's

upload media. I'm going to give it this video. Oh my gosh, you're so cute.

video. Oh my gosh, you're so cute.

Obviously, an AI generated video, but it's easy for me to work with. Here's

their video editor. Let's see. How do I do textbased editing? And now once I'm in the editor here, if I look up on the right, there's a little button that says

textbased editing. I will click this and

textbased editing. I will click this and it notices it says, "Oh my gosh, you're so cute." So, can I actually change what

so cute." So, can I actually change what it says? So, I can correct transcript,

it says? So, I can correct transcript, add text to timeline, assign speaker, or delete. So, it's very, very basic

delete. So, it's very, very basic editing. It's literally editing the text

editing. It's literally editing the text to cut out things. So, if I wanted to say just like, "Oh my, you're so cute."

and I get rid of gosh. Let's see how it edited that.

Oh my, you're so cute. So, you can see it cut out the gosh.

>> It doesn't seem to have like a ton of editing features. You can just sort of

editing features. You can just sort of edit the text to cut things. Right now,

you're able to do that directly inside of Firefly, which is pretty cool. And I

imagine they'll add more AI editing features over time. Let's see what happens if I click on generation settings. Describe the video you want to

settings. Describe the video you want to generate. So, this is just for

generate. So, this is just for generating video, not for editing video.

So, that's the very basic text editor inside of Firefly right now. The company

Luma AI also rolled out a new AI video model this week called Ray 3 Modify.

Now, this one seemingly lets you give it video inputs plus a starting frame and ending frame, and you can change the video based on the starting and ending

frame. So, like it's actually a real

frame. So, like it's actually a real human sort of driving the action, but then you reskin it with something else like we could see in these demos where there was no crowd, they resinned it

with a crowd, changes person's hair color, things like that. So, the first thing I tested was I gave it this starting frame that you can see up here.

shirt. Maybe we can zoom in on it cuz I know it's tiny. But it's this image of me, my buddy Joe, and my buddy Brad all standing in a brewery. And then I gave it this image of me with like the purple glow around it and the leather jacket

that I created in a previous video and told it to animate between those two.

And this is what we got. We can see they walk away and then I weirdly put on this jacket and my arm kind of goes through the sleeve weirdly. But that's what it did with the start and end frame. But I

wanted to test with a driving video which unfortunately I had to upgrade.

So, you actually have to pay if you want to use a driving video. So, I fed this video of me playing with a lightsaber inside of my office here along with this image that I generated in Leonardo of a pirate holding a sword, hoping that it

would take the animation of me playing with the sword and apply it to the pirate. But what it did instead was as

pirate. But what it did instead was as soon as it gets to about the 10-second mark, it changes my lightsaber to a sword, but then sort of just crossfades to the pirate. So, it didn't really use

my video as the driving animation for the pirate. But here's how you do it. If

the pirate. But here's how you do it. If

you're in Luma Dream Machine, you click on boards here. Create a new board. Down

in the bottom right, make sure you're set on modify. Make sure video is selected. And make sure array 3 is

selected. And make sure array 3 is selected. This is the brand new model

selected. This is the brand new model here. And now you can upload a video.

here. And now you can upload a video.

I'll give it my pirate video here. It's

going to crop it to 10 seconds for me.

And then it gives me the option for start frame, modify frame, character reference, etc. So, let's go ahead and set the character reference to our

pirate here once again. And also, let's set the pirate as our start frame. We'll

plug this in right here. So, now our start frame should be this pirate. My

video with the lightsaber is our driving video. And then also the pirate as the

video. And then also the pirate as the character reference. We can also adjust

character reference. We can also adjust the strength here. I'll just leave it in the middle cuz I don't really totally know what it's going to do yet. And

hopefully this is better than my first test. And after waiting for 10 minutes

test. And after waiting for 10 minutes or so, cuz it was taking forever, I got generation failed. It looked like it was

generation failed. It looked like it was working. I was seeing some like

working. I was seeing some like animations going, but generation failed after 10 minutes, which is frustrating cuz I am a paying customer. My guess,

because it's brand new, it's getting overwhelmed right now. It's overloaded.

And after waiting another 10 minutes, I finally got this. And it's okay. Yeah, I

mean, it it followed my motion and it applied it to this guy. There's

definitely wonkiness with the sword where it kind of disappears and there's some weird artifacts, but I finally got a driving video to animate an image that

I added in. This second try, I removed the avatar and just uploaded my video and the starting frame without having a specific avatar attached and that seemed

to do the trick. Again, I don't know the best practices, but it seems pretty cool once we get it right. I mean, there's some pretty decent detail in these like ropes and things as well. So, it's it's

pretty impressive. I just wish it wasn't

pretty impressive. I just wish it wasn't so slow and I wish there was better instructions so I wouldn't have had to wait 10 minutes to learn that it failed because that's really annoying. But, we

got there eventually. But I'm not done yet. Clling rolled out some new features

yet. Clling rolled out some new features including motion control. So, they have a newly upgraded motion control in their Clling video 2.6 model, which is basically like a motion capture inside

of Cling. It can detect full body

of Cling. It can detect full body motions with fast and complex actions, flawless hand moves, expressive faces.

You can give it a driving video and you can give it an image and it will use the driving video and animate the image in the same way theoretically. So, let's

test it. So, this was a test that I did with our previous version. Didn't really

work very well. I figured let's try it again. I gave it this video. I then also

again. I gave it this video. I then also gave it this image that I generated in Leonardo of a Jedi holding a lightsaber.

I did not pair it up with a text prompt.

I just gave it the driving video and the image. And here's what it generated for

image. And here's what it generated for us. It works a hell of a lot better than

us. It works a hell of a lot better than it used to. That is for sure. Still some

wonkiness, but it's definitely a lot better than it was last time. Clling

also rolled out AI voice control inside of their Clling video 2.6. So, here's an example of what that could look like.

And it is a very, very impressive lip sync. So, check this out.

sync. So, check this out.

>> Is that my voice?

This one feels soft.

>> This one feels powerful.

>> And again, it looks really, really good.

Uh, that's one of the better lip syncs I've seen. But when I log into Cling, I

I've seen. But when I log into Cling, I don't actually know how to use this. So,

I can see they've got their Avatar 2.0 model here. And if I hover over one of

model here. And if I hover over one of these, it's not great.

>> Sharing my favorite lip gloss I wear to work.

>> I'm just setting my base with some fave loose powder. See that? Smooth, soft,

loose powder. See that? Smooth, soft,

and totally ready to slay.

>> So that lip sync does not look like the lip sync we just saw in their demo. Now,

they did say it was in video 2.6. So if

I go to the video model here and I go text to video, we have the option for native audio. So I guess let me just try

native audio. So I guess let me just try a prompt where somebody is talking to the camera. So, I give it the prompt. A

the camera. So, I give it the prompt. A

man looks into the camera and says, "Don't forget to subscribe to Matt Wolf." And here's what that looks like.

Wolf." And here's what that looks like.

>> Don't forget to subscribe to Matt Wolf.

>> I mean, yeah, that's pretty good. So,

the the cling video 2.6, if you tell it to add speaking, it looks pretty dang good. Again, I feel like I'm beating a

good. Again, I feel like I'm beating a dead horse here, but best lip-syncing I've seen so far from one of these AI models. We also got a new video model

models. We also got a new video model out of Alibaba called Juan 2.6. This one

is very similar to what we just looked at with Cling because you can give it reference videos and you can give it images and it will animate the images using the reference videos. It's also

got native audio video sync. It can turn simple prompts into auto storyboarded multi-shot videos. It looks like a

multi-shot videos. It looks like a really cool model. So, let's see if we can test this one out. See what I did there?

>> N yo, hold my poodle. Hold my poodle.

>> If we go to one. We have first frame, last frame, sound driven. So, we can start with an audio file. It looks like most of the really cool stuff I've been

seeing people do with this new W 2.6 model, they've mostly done with the like open version using Comfy UI. So, I'm not sure if I can actually start with a

driving video or not. Let's make it starring Santa here. So, it's got that similar like cameo feature to like what Sora has. And let's make him jumping up

Sora has. And let's make him jumping up and down, yelling, "Yay! Christmas is

here." And well, not the fastest model in the world, as we can see, but with the power of editing, here's what it generated for us.

Yay! Christmas is here. Yay! Christmas

is here.

And finally, in the last bit of AI video news, I want to make an addendum to what I talked about last week. I mentioned

the new Runway ML4.5 model and I mentioned that it doesn't generate audio. However, I've been

generate audio. However, I've been seeing reports from like TechCrunch here that says it adds native audio to its latest video model. So, I went, did I miss something? And I jumped back in and

miss something? And I jumped back in and generated another video and well, it still didn't generate any audio with the video, but supposedly I'm seeing reports that it can do audio. It just isn't for

me or I'm missing something. I don't

know if you know how to use audio in Gen 4.5. Let me know in the comments because

4.5. Let me know in the comments because I don't know, it's not obvious to me.

Now, I mentioned that there was a lot of news that rolled out this week and I've just started to scratch the surface. I

still have a ton more I want to share, but instead of breaking them all down in depth, let's jump into a rapid fire.

Most throughout this video is going to be pretty rapid fire, but here we go.

Pew. Let's do it.

Starting with a handful of news out of OpenAI this week. Developers can now submit apps to ChatGpt. So if I look inside my settings inside of ChatGpt here and I click on apps, you know by

default there's Adobe Express, Canva, Figma, Gmail, and you know quite a bit more. Well, the ability to create apps

more. Well, the ability to create apps is no longer just going to be for these big companies. Apparently, anybody can

big companies. Apparently, anybody can now create apps and submit them for approval to ChatGpt. Now, they're not automatically going to be approved.

OpenAI does have some guidelines that you do need to stay within, and they will need to approve it before it goes live, but it sounds like ChatGpt is finally getting around to making that kind of like app store of ChatGpt. If

you're a user of chat GPT on your phone, whether you use it on iOS or Android, well, the branching feature that we've had available inside of the browser

version is now also available in the mobile version. So, quick quality of

mobile version. So, quick quality of life update if you use the mobile version. Oh yeah, they also announced

version. Oh yeah, they also announced that in Q1 of 2026, we're going to get adult mode. So, essentially, it's slowly

adult mode. So, essentially, it's slowly turning into Grock where it'll talk dirty to you or something. Google Labs

showed off a new product called CC this week. It's a new productivity agent that

week. It's a new productivity agent that connects Gmail, Calendar, and Drive to deliver personalized briefings every morning. So, this is kind of what a

morning. So, this is kind of what a briefing looks like here. It got sent to their email, and it says, "Here's the game plan for the day." And all of the details that are in this game plan were pulled from their calendar, their Gmail,

and their Google Drive. Now, you can join the weight list to use this. Right

now, it only will work inside of personal accounts. So, if you have a

personal accounts. So, if you have a Google business account, it's not there yet. It's just personal accounts. And

yet. It's just personal accounts. And

one thing that I feel like Google should do is make this feature available for multiple Google accounts. I doubt I'm the only one out there that uses multiple Google accounts, right? I have

like probably four different Google accounts that I use. One is purely for personal that only friends and family know. One is a business account where

know. One is a business account where like sponsors and various business inquiries come through. One is for future tools where people can submit bugs and things like that. I have

multiple Gmail accounts, multiple Google calendars, multiple Google Drive accounts. This CC thing would be really

accounts. This CC thing would be really cool if I could connect them to like all of them and it pulls the information from all of them cuz right now I can't even pull details from my main Gmail or

my main calendar because those are business accounts, not personal accounts. But I wanted to be able to

accounts. But I wanted to be able to pull in all of the information I need on a daily basis, not from just one Google account. Again, I could be an outlier,

account. Again, I could be an outlier, but I'm pretty sure I'm not. I think

most people have multiple Google accounts. Now, this news was last week,

accounts. Now, this news was last week, but I forgot to mention it. Some AI news last week, so he's going to talk about it right now.

>> Google also improved their text to speech model. Their brand new Gemini 2.5

speech model. Their brand new Gemini 2.5 texttospech model has enhanced expressivity, precision pacing, and seamless dialogue. Theoretically, you

seamless dialogue. Theoretically, you should get it to sound like the Notebook LM podcast if you wanted to because I think this is the same underlying technology. You can actually play with

technology. You can actually play with the Gemini 2.5 texttospech model inside of OpenAI's playground here, aistudio.google.com.

aistudio.google.com.

And here's a quick example of what this could sound like with two different speakers.

>> Hello, we're excited to show you our native speech capabilities >> where you can direct a voice, create realistic dialogue, and so much more.

edit these placeholders to get started.

>> I think you get the idea. I don't need to edit the placeholders. Interesting

though is you can hear a little bit of noise in the background. I don't know if I'm hearing it just cuz I'm wearing headphones or if it was very clear, but there was a little bit of like background crackly noise to it. But if

you want to get something that sounds similar to like a notebook LM kind of podcast, you can actually create multi-speaker audios for free right now with AI Studio and their new texttospech

model. Pretty sweet. Google also rolled

model. Pretty sweet. Google also rolled out a new feature in their deep research which now actually creates like visuals for the content that it created with the

deep research. So in this example video

deep research. So in this example video you can see it goes and does the deep research. It generates these reports

research. It generates these reports with graphs and charts and things like that all built into the sort of output that it gives you with deep research

here. Now this one I believe is only

here. Now this one I believe is only available to Ultra subscribers right now. Google AI Ultra subscribers. So,

now. Google AI Ultra subscribers. So,

you have to be on their $250 a month plan to get this feature, but I think it's only a matter of time before it rolls out into the lower tier plans and you can do deep research and generate

cool graphs and charts and reports and stuff. There was also a ton of new large

stuff. There was also a ton of new large language models released this week, including one from Google. They released

Gemini 3 Flash. So, a few weeks ago, we got Gemini 3. Well, the flash model is just a much faster to use and much more compute efficient model.

>> I don't think you have any idea how fast I really am.

>> Here's all the benchmarks if you're curious, but we can see the cost difference is a fourth for Flash verse Gemini 3 Pro. And the output price for

tokens is also about 25% of the cost. So

quite a bit cheaper to use. And when you compare its results on the benchmarks, it's actually pretty close to Gemini 3 Pro here in a lot of the benchmarks.

Humanity's Last Exam pretty close. Arc

AGI Google proof question and answers like almost right on. So, as we can see, it is a fast, cheap model that almost does as well as Gemini 3. Saying that, I

have heard reports that it hallucinates quite a bit more than other models. So,

if you are going to use it, double check your work, unless you're using it for like creative purposes. This person even managed to make a version of like bust a move using this Gemini 3 flash and it

looks pretty dang good. So, pretty

impressive model. Once again, just be careful when you're using these faster, cheaper models because they do tend to have more issues with accuracy. Now,

this model's rolling out globally right now. It is also rolling out in the API,

now. It is also rolling out in the API, so if you're a developer, you can use it. They're rolling it out in the Gemini

it. They're rolling it out in the Gemini app. So, if you use the Gemini app on

app. So, if you use the Gemini app on your mobile phone, you'll have this model available in there. And it looks like it's also becoming the default AI

mode in Google Search. So, when you use Google Search, it's most likely going to be using Gemini 3 Flash when you see the sort of like AI mode in the AI results.

But, we also got another model out of OpenAI, a coding model called GPT 5.2 Codeex. They claim it's the most

Codeex. They claim it's the most advanced agentic coding model for professional software engineering and defensive cyber security. So, it's a version of GPT 5.2, which launched like

what, last week or the week before, I don't know, launched recently, but more fine-tuned and designed for coding. We

can see it's slightly more accurate than GPT 5.2 on the software engineering bench and slightly outperforms GPT 5.2 on terminal bench. So, if you use AI for

vibe coding, this is another model that you can try out. Taking a quick peek at LM Arena here, it appears that Opus 4.5

is still the top model with 5.2 high just barely behind it. Nvidia released a new family of models called Neotron 3.

There's a nano model, a super model, and an ultra model. The Nano model is meant to be a fast, costefficient model, but probably is the least accurate of the

three, where Ultra is a large parameter model that's going to be much more accurate, but also slower and more expensive. But one thing that's

expensive. But one thing that's important about these models is that they are open models. So, these are ones you can actually install and run locally or run them in the cloud yourself, fine-tune them, do whatever you want

with them. As always, I will link up to

with them. As always, I will link up to the blog post below. So, if you want to dive in and learn even more about these individual models that I'm talking about, you can click the links and uh get all the details that I'm skipping

over. There's a new model out of Xiaomi.

over. There's a new model out of Xiaomi.

I think that's how that's pronounced called Mimo V2 Flash. This is another open-source model and it excels in reasoning, coding, and agentic scenarios. A lot of these models are

scenarios. A lot of these models are really, really honing in on trying to be good for agentic scenarios. We can see here in the benchmarks how the orange

MIMO model compares to DeepSeek V3.2, Kim K2, Claude Sonnet 4.5, GPT5 High, and Gemini 3.0. This basically shows that it's fairly on par with those

models. It doesn't really seem to beat

models. It doesn't really seem to beat out most of them other than in Swebench Multilingual, but we can see that it's pretty on par with some of the state-of-the-art models, which is

amazing for an open model. And in the last bit of LLM news, Manis released Manis 1.6. According to their press

Manis 1.6. According to their press release, it introduces three major leaps forward. Their Manis 1.6 Max, which is

forward. Their Manis 1.6 Max, which is their most powerful agent, mobile development for building beyond the web, and design view for interactive image creation. I honestly admittedly haven't

creation. I honestly admittedly haven't spent a ton of time with Manis. I played

with it a little bit when it first launched, but haven't spent a ton of time with it since then. I think it's about time pretty soon to give it another go, possibly in a video where we compare a whole bunch of the various

agents that are available right now. So,

subscribe to this channel if you'd like to see a video like that. All right,

moving along and away from LLM news.

This is actually also news from last week, but I missed it in the roundup.

and that's that this company StarCloud is training its first AI model in space as the orbital data center race heats up. So, here's what's going on. All of

up. So, here's what's going on. All of

these companies need a lot more compute than what they have. We only have so much land and so much water here on Earth for cooling these data centers that they think that the best solution

is let's put the data centers up in space. One of the suggestions says

space. One of the suggestions says they'll have like 80 of these like data center satellites all fairly clustered together communicating to each other via lasers. Kind of like this net up in the

lasers. Kind of like this net up in the sky powered by AI. Where have I heard that before?

>> Let's pray to God this works.

>> Skynet defense system now activated.

>> Anyway, the idea is that well space is cold and we can have these data setter satellites always be in a position where the sun is hitting them, meaning unlimited power. There's just a few

unlimited power. There's just a few problems with that. One of which I think Hank Green here uh sort of spells out pretty well. He says, "Hey, so I've

pretty well. He says, "Hey, so I've heard several times that space is a good place for AI data centers because it will help manage heat. And I feel like I've heard it from credible people, but vacuums are famously insulating and in

order to get power, the data centers will need to be in the sun, which is famously hot. I feel like I'm going

famously hot. I feel like I'm going crazy. What am I missing here?" I don't

crazy. What am I missing here?" I don't think he's missing anything. I do think there's a big issue with putting satellites up there that are always visible to the sun and have very little

way to dissipate the heat off of the data centers that are floating around in space. So, I still think there's quite a

space. So, I still think there's quite a few of engineering hurdles that need to be figured out before this actually feels viable. I also came across this

feels viable. I also came across this article on The Verge which has even more skepticism towards this idea. Another

issue, the group of satellites would need to travel through millions of pieces of space debris or a minefield of random objects, each moving at 17,000 mph. The space debris is especially

mph. The space debris is especially concentrated in popular orbits like the sun synchronous orbit. Dodging each

object requires a tiny propulsion to move out of the way. In order to generate that repulsion, well, it needs fuel. Anyway, it's an interesting, very

fuel. Anyway, it's an interesting, very futuristic sounding idea, but a lot of like scientists, engineers, and astronomers are all saying it's probably

not as close as we all think to being a very viable solution. Anyway, missed

that story last week. Wanted to give my two cents on it this week. Also, this

week, Microsoft released Trellis 2. This

is a model that takes images and turns them into 3D models. We could see some examples in their video here of the types of 3D models it spit out. And I

mean, they're definitely the most realistic 3D models I've seen so far in like a image to 3D pipeline. Like this

is definitely the best we've seen so far. At least from their sort of

far. At least from their sort of cherrypicked examples in their demo. So

I jumped over to their demo on hugging face and gave it this image of like this futuristic cannon thing. It was actually one of their example images, but easy

enough. And here's what it created. I

enough. And here's what it created. I

could see different variations here.

Interestingly, I can't like click on it and sort of rotate it to see the various 3D angles, but I can drag this around and it will rotate it for me, but it

only rotates it sort of on a horizontal plane. I can't like rotate it in every

plane. I can't like rotate it in every which direction. But the details that it

which direction. But the details that it gets us are, again, I'm beating a dead horse. Really good. All right, moving on

horse. Really good. All right, moving on to Amazon news. They actually have an online chat bot like ChatGpt where you can go and talk to your, you know, this

app. Now, it's exclusively for Alexa

app. Now, it's exclusively for Alexa plus customers right now. But here's

what it looks like when you get into it.

It's basically a chatbot like chat GPT, but you can talk to your Amazon device.

I'm just trying to avoid saying this word to not trigger yours in your house.

Now, I believe they're using the anthropic model, but don't quote me on that. I'm not a 100% sure. I just know

that. I'm not a 100% sure. I just know they've made huge investments into Anthropic. Let's see what it knows about

Anthropic. Let's see what it knows about me. Wolf's a multiaceted American

me. Wolf's a multiaceted American entrepreneur who's made quite a name for himself in the AI and digital marketing space. He's best known as the creator of

space. He's best known as the creator of future tools, a popular platform that curates and organizes AI tools and noodles. Uh yeah, I mean it got it

noodles. Uh yeah, I mean it got it right. Recognition and awards featured

right. Recognition and awards featured in Edelman's AI creators you need to know, YouTube creator award, 100,000 subscribers, TubeBuddy Emerging Creator Award, and the co-host of Hustle and Flowchart. That's old news, but yes.

Flowchart. That's old news, but yes.

current projects, future tools, next wave, YouTube. I mean, it's really

wave, YouTube. I mean, it's really accurate. They did really good job. I

accurate. They did really good job. I

don't know who these people are. Those

aren't me. Since we're talking about Amazon, though, if you have a Ring, you'll soon be able to have AI talk with guests at your door. So, Amazon's AI can now answer your Ring doorbell and talk

to visitors for you. The greetings

combines conversational AI with Ring's video descriptions to intelligently handle doorbell interactions. You can

have it manage deliveries when you're occupied, handle interruptions gracefully, help friends and family when you're not available or stay in the know while you're away. This will be interesting. I don't know if I want AI

interesting. I don't know if I want AI talking to people that come to my door, but I also think it'd be fun to have AI screwing with people when they come to my door. So, maybe I will. I don't know.

my door. So, maybe I will. I don't know.

I just thought it was interesting and worth sharing. The French company,

worth sharing. The French company, Mistral, released OCR 3, which is currently the best OCR model available.

that stands for optical character recognition. It's basically a fancy way

recognition. It's basically a fancy way of saying we take handwritten text and turn it into typed text. While this OCR3 is now the best model available, probably not the most interesting thing

to most people, but I created a journaling app where I wanted to be able to type journal entries or handwrite journal entries, and it's got OCR built into it. So, I'm probably going to be

into it. So, I'm probably going to be swapping out the existing OCR with this new OCR model to get even more accurate with my handwritten journaling that goes into my journal app. So, I'm excited

about it. Here's another cool update out

about it. Here's another cool update out of Meta that I thought was worth sharing. The Meta AI glasses now have a

sharing. The Meta AI glasses now have a feature called conversation focus. They

also added Spotify integration. But the

conversation focus is what's really interesting to me. It's basically a feature where if you're in a loud environment and you're wearing your glasses, it will amplify the person you're talking to. So, it helps the user

focus on the conversation that they're currently having even when in noisy environments. I think that'll be a

environments. I think that'll be a pretty good quality of life feature for these glasses. I think that'll be handy.

these glasses. I think that'll be handy.

And in our final bit of AI news in this week that was already a pretty epic week in terms of volume of news, the 2025

word of the year from Webster's dictionary is slop. So, yay, we did it.

We got one of our AI slop words to be the word of the year. We define slop as digital content of low quality that is produced, usually in quantity, by means of artificial intelligence. We got a lot

of slop this year. We said the word slop a lot this year. And well, slop is pretty much in everybody's lexicon now because of AI, so makes sense. I can get

behind slop being the 2025 word of the year. Anyway, I'm out of breath. There

year. Anyway, I'm out of breath. There

was a lot to talk about this week and I think we kind of covered it all. If I

missed anything, let me know in the comments. I always want to know what I

comments. I always want to know what I missed and if it's something huge, I'll make sure it gets mentioned in the next video. Also, I noticed in the comments

video. Also, I noticed in the comments people were asking me about like this little picture frame that I have here.

This is actually all of my uh like selfies and images from conferences. So,

anytime I've gone to a conference and got a selfie with somebody and they posted it online, I grabbed those selfies and I put them on this frame.

So, if you ever meet me in person and we get a picture together, it'll probably end up on this little frame circulating in the background of my videos. That's

what's going on there. But, what a week.

I really thought December would slow down and well, December seemed to ramp up. Next week's going to be a slower

up. Next week's going to be a slower week. I'm going to put out a news video

week. I'm going to put out a news video at the end of the week. most likely on Sunday just due to my editors needing Christmas off and stuff. Uh, but there will be a news video next week. That'll

be the only video I release next week. I

am taking a break for most of the week.

Hopefully you enjoyed this one.

Hopefully you feel looped in. I will

loop you in again next week. And if you want to stay looped in yourself, I say looped a lot. Make sure you subscribe to this channel. Like this video. That'll

this channel. Like this video. That'll

ensure more videos like this one show up in your YouTube feed. But it's been a crazy week and I am done reporting on the news until next week and there should be a lot less to talk about. I

don't think companies are going to release stuff on Christmas, but we'll find out next week. So again, like, subscribe, do all the things. And uh

hopefully I'll see you in the next one.

Thanks for nerding out with me. Bye-bye.

Thank you so much for nerding out with me today. If you like videos like this,

me today. If you like videos like this, make sure to give it a thumbs up and subscribe to this channel. I'll make

sure more videos like this show up in your YouTube feed. And if you haven't already, check out futuretools.io where I share all the coolest AI tools and all the latest AI news. And there's an awesome free newsletter. Thanks again.

Really appreciate you. See you in the next one.