GPT-6 Astra: 20 Real Examples From Useful to Almost Impossible transcript

As you might have heard, GPT6 Astra is here and it's not just, you know, better at doing typical tasks. It does a whole new category of things. And a lot of people had early access to this thing. So, they've already pushed this beyond the limits of what was possible with AI up until now. So, in this video, I want to show you my favorite 20 things that I found across hundreds of examples that the internet has produced of use cases and things that this model can do that just weren't possible before.

Not at this level. And one important nuance to me is we're not going to just look at impressive, you know, rebuilds of cities or monuments in 3D. But half of these are genuinely useful and hopefully you can take some of them and implement them into your life, into your workflow. And then the other 10 are honestly hard to believe. But before we dive into that, this is not the focus of the video recapping the model release.

I do want to do it in like 30 seconds though because this blog post is genuinely impressive. You know, first up, benchmarks. It just beat everything before Astra on most of them. Particularly the computer use meaning remote controlling a computer is super impressive. I always like looking at these professional benchmarks too where they actually do real world tasks crushing it. And then also very interestingly the long context capabilities have been praised by many people and also on the benchmarks it outperforms everything before meaning if you give it a ton of information it will actually remember it versus previous models claiming to have that capability but really it would get amnesia somewhere around a few hundred thousand tokens.

Okay. So, those are benchmarks. And then this blog post has a ton of different graphs showing off how amazing Astra is, how better it is than all the competition. And yeah, it's, you know, it's it's real data. It's maps. But what I find even more interesting is, hey, how can I apply this to my world? That's what this video is about. That's what a lot of this blog post is about, too. I mean, you see these different games embedded in the blog post where you can play little games that, well, oops, there we go.

Honestly, they look great. But beyond that, you'll quickly notice a lot of these demos are in 3D software. And that's going to be a big theme throughout the video. It's just insane at 3D modeling and creating 3D worlds in Unreal Engine and anything that has to do with 3D. So, let's explore some of these examples and I might sprinkle in some details from the blog post that you can kind of check out by yourself. By the way, this launch video, amazing.

That's a great watch if you haven't done that yet. It's at the top of the blog post. But now let's begin with some of the most impressive examples and use cases that people have come up with all across the internet. Starting with this one. So it's an entire world with 600 AI people in it. Okay. And this was created by Matt Schumer on X and basically had early access and he built a entire well planet um islands village I guess and he put 600 people in it and he gave them autonomy to like walk around and the goal was for all of these humans to survive and they had you know access to AI to communicate to generate language and the way he puts it he story tells really well around it is that you know he came back into his room and all of a sudden he hear noises and these humans they were communicating with voices and making plans.

I can bring Hoy some food if he needs it. >> Thanks, Bruce. I've got some food. I'll ask him when I get there. >> Honestly, that's one crazy to hear, but two, if you know anything about game development, if you set up a world like this and you put characters in there and they have the ability to use LLMs to generate text and then use other AI to turn the text into voices. That piece is not even that crazy. The crazy piece is that this was created by Astra. all the 3D models of the people, this map, all the logic that holds it together.

That's a team of developers working on something like this. You might have seen AI games before, and it's usually like a space shooter, like a little rocket going somewhere or really simple 2D games, but all of a sudden, it's making legitimate 3D worlds. And as you'll see with some of these other impressive examples, it's crazy that it put together all of this and then you can let that world run. So, let's look at a few more examples and then try and draw some conclusions on where this might be heading cuz this is really unique and interesting and cool.

And here we have a second example where it rebuilt the Palace of Fine Arts in Blender, which is a 3D software if you're not familiar. This is a building in San Francisco, I believe. Yes, exactly right. Wasn't that in the park in front of the Golden Gate Bridge? Oh, yeah. Exactly right. It's here. And if you kind of check it out from the pictures, this is what it looks like. Gorgeous. Now, here's the AI generated. It's not an AI generated video.

This is a 3D model. This has been fully created by Astra in a 3D modeling software, Blender. Okay. And look at that level of detail. Honestly, these images up here, this is fantastic. What about the hands though? Let's get nitpicky here. One, two, three, four, five fingers. Okay, pretty good. Pretty good. It's just a new category of capability. And very quickly as you look at these examples, it becomes obvious that they must have trained it on so many 3D models and so many workflows and people creating these 3D models that it can now replicate that.

This is a great example just because it's so detailed. And if you're like, "Okay, Mingor, that's, you know, a building. Not bad, but like the world is a bit more complex than that." How about this next one? How about a three-dimensional city? This thing is essentially like Sim City. Check this out. It's called New Haven. You can try this yourself. I'll put all of these links in the description below, by the way, if you want to try them out yourself.

But right here, this is New Haven, and you have so much functionality. Check this out. I'm kind of just going through these different areas. Now I can design uh well this looks let's do a office block. How about here? I'm going to designate this as office buildings and then I could go to the towers. Can I place a tower here? I bit need a bit more real estate. Huh? How about out here? Let's build a tower here. Oh, construction is beginning.

I want one more tower next to it. Can you believe that all of this has been made by AI? Look at all these functionalities here on the left. Your city at a glance. Healthcare is doing good. But you can check this out as it moves along. Things happen here. There's a urgent police call that I can check out and deal with. There's a reported crime in this building. Oh my god, what's happening? I think you see the point. You can play around with this.

There's a ton here. It's just this complexity that is impressive. And also, this looks really good. Like, nothing here is really wonky. Wouldn't you agree? Like, what about these windmills? They're great. These buildings look solid. Unbelievable, really. Okay, this one I particularly like to see because it's just so measurable. This is the kind of benchmark that I want to see. It crushed the Pokemon completion record.

Okay, so if you were following this over the past years, you'll realize that even a year ago, no model could really complete Pokémon without any heavy external help. And then with the recent GPT models, GPT 5.5 and then GPT 5.6 Saul, first it took 200 hours. Then with 5.6 Sol, it took 96 hours to complete Pokémon. It finally could do the whole game by itself. Models like GPT4 just couldn't do it. Astra on the highest settings, 18 hours.

The entire game. I wonder how long does it take a human to complete Pokemon Fire Red 25. I knew it as a kid. This always took so long. I was always a do almost everything type. So it took 60 to 100 hours. But even if you just go the main story and you beat the lead four in the end, 25 to 35 hours. So this is the first time AI has become even in the lower settings faster than a human completion. By the way, there's a Twitch channel where you can see it play Pokémon live.

This is just GPD6 Astra playing Pokemon. So, if you're trying to do a faster playthrough, you can now look at how Astra would play it. It's pretty far along. It has most of the badges here. 120 people watching this live. All right, this next one is a really fun one from Petro Shirano. And this is where we start entering the territory of, hey, you might actually do this and use it for something practical. In just one sentence, hey Astra, make me a Mac app.

It should render a freebie iPod that you'll build in Blender. Then, using the original iPod UI and interaction, I want to be able to visualize all my codec threads on it. So he's using it for different, you know, agent threads in codecs. But you could put anything on here. You could create a portfolio application on your site where you just put this into your site and yeah, all of a sudden you can use a 3D model of a iPod or any other object for that matter as a part of your website or you could use it as a presentation.

The use cases for these things, they're still a bit blurry. Like it's not clear what the killer use case for all of this 3D stuff is for consumers, right? If you want a game, you're still going to have a game studio build that. now with the help of this probably but for common people like we don't need and use 3D modeling all that often right in their demo video they showed off this example where you know he created a rocket that was really detailed and then he put it into a game and then he 3D printed that rocket again check out that video it's fantastic but people usually don't need 3D printed rockets and that's not me saying that this is useless I'm just saying we have yet to figure out where this fits into our lives smoothly and the capabilities just haven't been at a point where people could freely experiment and try a lot of things just cuz the limitations were super limited.

Every city, every object that you built was wonky and incomplete. And now, for the first time, we're seeing something that is complete. And I think people will quickly start figuring out what can be done with this new tech that hasn't been done before, even for consumers. Okay, next up, it took a Van Go painting and it turned it into a 3D world. So, you can do that, too. Look at the sunflowers here. Again, all of these links will be below.

You can kind of try this out yourself. You can move through the Van Go painting and Van Go's world. It's kind of interesting. Again, remember we're in the first part of the video where we're looking at these interesting things. The useful ones are coming up soon and there's quite a few. All right, onto the next. And this one goes into the realm of video production where one prompt created a educational video. Now, we already saw that Fable 5 and 5.1 that came out this week are super impressive at video production and motion graphics and all of these things.

Now, from my first look at this, this looks to be next level. It's just clean. Look at this video. This is a 6-minute video, and I'm hardressed to find something that is obviously wrong with this. And look, this is not some pre-made presentation that is super good in advance. It's just a prompt that says, "Create a 5-minute educational video about T- cells." So, yeah, that's something quite concrete and useful. Whereas before, you know, these videos existed, you could create them with tools like Notebook LM.

But often times they just had these artifacts in them that made them like, "Oh, that's obviously wrong and not supposed to be there." This looks increasingly clean and that threshold is an important one. As soon as these things that it creates don't obviously stink like AI slop, that's where people start using them and I think we just crossed that threshold for a lot of these use cases that I'm showing you here. Okay, three more interesting ones and actually this one enters into the useful territory already.

These are video editors that gave it Final Cut Pro with a limited task. Importing certain files, doing the color grading, meaning adjusting the colors in the scene so it looks presentable, and then syncing up the various clips so the screen recording aligns with the video. Something that is a standard workflow with any video that we make here on this channel or any educational video. Okay. And this was actually quite a good watch.

They go ahead and comment live over this thing as it brings in the files. It even created a separate folder for everything that makes it easier to share and they're like, "Oh my god, I wouldn't have thought of that. This thing is smarter than us." It was kind of fun. And then as it creates a sequence and syncs the things up correctly, it attempts the color grading. It does an okay job, I would say. And then it goes one step further.

It actually finds the multiple audio tracks and it picks out the one audio track that is the best and deletes the rest. So, it went beyond what they asked for, but did a kind of obviously useful task for them. Now, this is not a full, you know, test suite for video editing. They didn't really edit anything together. They didn't have to make any hard decisions. It just followed a process that they have to do every time.

So, I don't know how this is going to exactly evolve. I don't know if this model is good enough yet. We're going to see over the coming days and weeks to actually make the decisions of what should be taken out, what should be put in. I can tell you that no model this far has been useful at tasks like this whatsoever. It's just too unreliable. This looked really solid. Okay, here's a fun one. Ethan Mollik took a open- source repository from this guy called token gremlin where he created a fully procedural ocean and weather generator but it was only above water.

Okay, so you could simulate different tides and different weather conditions. And procedural means that everything was generated by the computer not laid out by humans. And also as the computer lays it out, if it procedurally generates something, it's going to be a different result, a different composition of assets and everything every time you run it. And then this thing is open source. It's just out there. You can get the source files here on his GitHub.

And he simply told it, hey, okay, this is a great simulation of everything above the water. GPT6 Astra, can you go ahead and create everything that is under the sea? And he was like, yeah, sure, no problem. And it just did that. And yeah, all of a sudden now, yeah, there's an underwater world that was just created by prompting it with a sentence. I mean, again, I'm not exactly sure where this comes into play in everyday life, but it's just crazy.

Can we appreciate that for a second? Now, let's look at the last impressive one. And again, this is a oneshot example of something that a lot of us kind of have in their mind when it comes to AI. It's this idea of this minority report like interface when we interact with things. If you're not familiar, it's basically Tom Cruz sitting there and like remote controlling the computer with his hands and you know things happening.

Well, this wonderful lady here, Claire, just went ahead and asked it in one sentence. Again, it's just Astra figuring it out. She just gave it one thing. No revisions, no corrections, just hey, can you build me a minority report like interface? >> I enable Mac control. If I point it now moves around my mouse, um, I can maybe go down here and pinch and actually open up. >> And in one sentence, she created a new interface for the computer where you can, you know, point your finger and if you do this, it clicks.

Interesting, right? All right. So, let me tell you, the internet has hundreds of these examples. These were some of my favorites. But let's now look at 10 that maybe are a bit more practical and maybe you might gain inspiration for something you might want to do. But I really want to start out with this story where OpenAI shared how GPD6 Astra helped them run the launch campaign for GPT6 Astra. Look at all of these tasks that it did.

This by itself could be a list of various things that you could do. building and maintaining Google Sheets, creating and updating communications plan as the team was moving forward, drafting pitches and managing embargos, tracking replies, press briefings, branded PDF creation for blogs, and then checking if the final results were actually aligned with what they put out, collecting all the assets across the teams and building press kits, tracking the coverage of the model across the internet in real time.

This is what I built for myself, too. That's what I created this video from. created a site to analyze all the coverage and the social reactions and then writing a report summarizing all of that and then she could focus on you know the higher touch things not maintaining Google sheets or creating PDFs but communication between the team and the overall narrative and the strategy actually hosting the calls getting a full night of sleep monitoring the situation and making decisions that might need to be made with all of this information I think this is a really well-crafted kind of list of what AI should be doing and what humans should be doing and hey if you see any of this and you're like, "Oh my god, some of these use cases, they go directly into the areas that I get paid for in this world." Well, then it might be a good thing to think about how these tools could enhance what you do, so you can focus more on strategy and communication.

I know that's not reasonable for everybody and every job, but that's probably the direction that this is moving us into. And this first example shows that really well. All right, so our journey into practical GPT6 Astra use cases continues. And I want to just tell you before we move into these examples that some of these are oddly specific. They're not going to be these general purpose category where everybody's going to be able to replicate this.

But that's the point. Your work is probably specific to you. That's why you're doing it. And these are just examples. So, you know, when you see them, try to remove yourself one level from that work and see if you could apply that same idea to maybe what you're doing. And for some of them, it's just not going to be the case. But with that being said, let's look at this next one which is a scientist's app. So yeah, if you're not a scientist, this might not resonate so much, but here Daria reports on X that it built a sophisticated application that measures flow cytometry data.

Honestly, I don't know anything about that. I guess he explains that it's one of the most essential technologies in immunology. But what he does state is that labs were paying thousands if not tens of thousands of dollars in subscriptions for applications like this. And now as he states he almost built a fully functional researchgrade version of such an application with Astra. And what I found interesting here is that he stated that he built previous versions of this idea with codeex before this which is the openi application to build things.

But this is the first time he has reached a version that works exactly the way he wanted it to. So this is interesting. Many people experience this where they have an idea and they're like, "Oh my god, AI can do this." And they come in and then they're disillusioned. And often times that, you know, causes a negative emotion where you're like, "Ah, oh my god, this thing is not as good as it's advertised to be." And sometimes it's the context or the harness that gets the context that they got wrong.

And other times it's just the pure capabilities of the model. And here across the board across everything we're seeing, these capabilities, they just rose. They can do things that weren't possible before. And you're going to see that pattern repeat over the next few examples, too. Because in this next example, Astra coordinated 55 agents to audit 10 different financial models. Basically making sure that all of the information is correct.

So doing that double-checking that often times was kind of the sticking point for many people. It's like okay AI generates things but then who makes sure it's correct? So he says that he gave it 10 financial models and Astra audited everything and compared the results to the original workbooks and corrected any issues. Now, this is the type of thing that is, you know, hard to verify and we don't really have benchmarks on reliability of stuff like this.

But as these things get smarter, it's also kind of intuitive to just think that, all right, so they also become more reliable. And things like double-checking sources can be done, and you can have one AI generate something, but then if you spawn another 54 AIS that are solely responsible for double-checking everything, well, who knows what's possible. Okay, so here's something completely different. Content creation and writing.

So Dan Shipper on X shared the following and I thought this was particularly interesting. He said it's the best writing model he's ever tried. It's fast, I suppose, on the low setting, produces very little slop and is easy to steer. So good instructions following. Not every model has that by the way. Uh you might know this, but a lot of times you include something in a prompt, especially with Opus 5, for example, these days from Claude, and it just doesn't respect everything.

It's like, okay, you said that, but I'll do my own thing. I have my own ideas. Easy to steer. Something I like to hear. And he also says it's a good companion for actually working through the writing I do every day. And that is a great writer by the way. So I'll give that a wait. If you look at the process that Fable Design versus Astro Design, you can see what I'm saying about the intuitive adherence to the sense of the prompt and then extending it.

So what Fable Design pretty simple workflow just show a page, press one button, and then go to the next page and it'll just continuously transcribe. Astra, it has a more wellthoughtout, well-designed interface. Like, it's got the warm paper. Yeah, this is one signal. I haven't seen that reflected anywhere else. But I just can't wait to get my hands on this and see how the writing performs. But that seems very promising cuz honestly, that's one of the biggest use cases for most people.

Just it's just writing stuff, whether it's, you know, documents, reports, articles, or just responses that they read and then do something with. Writing matters and seems to be very solid at that. And then this one is interesting because you might be able to find ways to use this for yourself. And basically what Theo here did is he connected it up to his email account and he had it search the entire email history and collect all of the SSDs and RAM that he has purchased over time.

And then he had it compare to current market prices. Okay, so that's a relatively simple use case, but you could think about this idea of it looking over a massive set of data like all of your emails or maybe all of your files. We'll see more examples of that in a second here. And then using those results in some other way. In this case, the result is the price of all the SSDs and the RAM that he bought. And he compares it to the current market price.

But it's a real pattern here, right? You can think about extracting one piece of knowledge from a set of data that you have and then using that knowledge to compare it to something else in the real world. Think about all the knowledge that hides in your email. There might be a lot there. Well, if you have your context set up correctly, you could just go into chat GPT or Claude and ask, "Hey, how could I apply this pattern, this use case to my world, and to my work, to my personal life, just need the right context in case you need help with crafting that context?" Well, that's exactly what I have a course on inside of the AI Advantage Club.

It's called Build Your Clone, and you basically set up context so it mimics your values and your goals and all these other things. I show you how to do that. The trial to the club is just $1. You can try it out today. It's in the description below. Anyway, that's just a little note on how to get the context right. Let's look at the next use case. And this next one from Ethan Mik is really interesting and I think a lot of people could try this.

And I mean talking about context, he just built a wiki of all of his work context. This is not a laser focused, hey, this is me and help me apply this to my work type of context like I just talked about I teach in that course. But this is more like here's everything about my career. What he did is he used GPT6 Astra to read through tens of thousands of his emails, his writing, his calendar appointments, and more. So, you know, you could give it your entire Slack history, all your email accounts, stuff like that.

And then he built a personal wiki out of it. By the way, if you're not familiar, the personal wiki idea is something that Andre Karpathy popularized earlier this year. It's basically turning all your documents into a repository of different text files that are all linked to each other. If you Google Andre Kapathy wiki, you can find more on that and how to create that. But basically, he created that for his entire personal life with Astra.

And now he has twice a day AI sent him an update that is a summary of things he should be paying attention to or things that he might be interested in. Interesting. So basically he took his entire work history and now uses that to inform what he might be interested in in the future. Using history to predict the future. Kind of a smart pattern, too. You could do this too. Few things to note here that he even highlights.

He gave AI access to his computer, meaning it, you know, used the computer. For that, he probably used the desktop application with this model. And it looks like it pursued that goal for 4 days and 21 hours. You know, you could imagine how many tokens that cost. That's probably more than a $20 subscription. I'm not going to cover the pricing aspect of this model in the video. All of that is in the blog post. But a multi-day run.

I don't know if he went straight through the API or actually used the desktop app with the subscription, but this might cost you hundreds if not a,000 plus dollars depending on many factors like caching or which reasoning level of the model he used. He was not charged for tokens during the trial period, but yeah, I don't know. My my my guesstimate would be $500 to $1,000 for this task. I don't know. Might be worth it.

You create that wiki once. Also, there's the risk factor of like giving it access to your computer. But honestly, I've yet to encounter somebody who had an issue going through the official apps like the Chaty desktop app where people run into issues is once they get into the terminal based um AI coding agents there things can go wrong. And yeah, for everybody curious in the comment section, he does highlight that it used Obsidian for the wiki as most people do when they build that.

And by the way, I have the same thing set up. I have Obsidian with sync across my laptops working with my main agent which it's still open claw. I just seem so deeply entrenched and love it and it works for I customize it so much that I still use that. And I just noticed that my Obsidian account is not synced on this computer. It's between the other Mac Mini and my MacBook Pro that I always have with me on the go. This is kind of my recording station.

Cool. Next up, quality control with Astra. Because it's so good at using a computer and because it's so much more precise and so much more intelligent and it's just better across the board, smarter across the board, more precise across the board, you can now do quality control at a new level of preciseness. So Claraara here on X highlights a bunch of different use cases, but particularly this one browser used for QA at 13 minutes caught my interest. >> She just reports that it's just straight up better than before and that the QA process ran for over an hour where it really checked all the buttons and everything that a user would experience while using this application. >> It started clicking through testing, sending chats, and what was really helpful about browser use here is it was inspecting the console, checking for error logs, it was doing things like race conditions.

It would like refresh Chrome. It would do all this stuff that would be very tedious and hard to execute as a human. >> So, whatever you're doing or building, it's not just better at making things, it's also better at checking them, which makes it better at making things because it feeds that info back into the creation process. Okay, this next one is similar, but it talks about the speed of the model, which is lower than previous models that you might be used to.

But Ben Davis here pointed out that it would more proactively open a browser to check its own features by default. This is a thing that in my own agents I had to set up manually where you give it entire processes of how to test and when to test, but it seems to be just better at not just using the computer, but knowing when to use it. That's what I'm reading out of this one. And he says it's the best oneshot model ever made.

If you're not familiar, oneshot is basically giving it one piece of instructions and getting a result without refining it because it can use the computer really well to refine itself. And then that also was a reply to this tweet from Max talking about how it's basically the best personal software generation model, but sometimes you need to wait for it to get all the pieces put together. So yeah, just like we saw in the Ethan Molik example, if something runs for hours or maybe even days, like in his example, don't be surprised.

It's, you know, using a computer to review work, but the result will be better than ever. So let's look at this next example of that in action because Wade Foster here shows several examples from how they tested it on automation bench which is basically a practical real world benchmark where it goes up against tasks that you would want to automate. One example he stated here is rebalancing a quarterly media budget from last quarter's actuals.

And they tested that against the previous model that was the best one from OpenAI Saul. And Saul's result looked finish but had the splits wrong of the budget. And Astra found the adjustments of the splits and every number was right. Again, that loop of checking itself and just being smarter at doing that and smarter itself in action. There's another example where actually Soul outperformed Astra, but only by a little bit because it got partial credit for the task and Astra just gave up.

This is a big difference between the models and I wanted to highlight this for you that Astra will not guess. When the instructions exist and it can find them, it finishes the entire job. When it cannot, it pauses the work instead of improvising. And this is just something to get into your head because up until now we're so used to AI going ahead and you know trying doing its best and then you correct it and you're like hey that is completely off.

What are you doing? And then it will respond you're right eager and then it does the right thing. But if it tells that to you five times in a row, you just lose all trust in it. You're like, "Okay, why do you tell me I'm right every time? How about you be right for a change?" And that's the change that I'm seeing here or reading from in between the lines here. Or more like that's the feeling I'm getting by reading in between the lines of all of these things that people have been doing.

And if you care about the automation bench results which Zapier runs and Wade is the co-founder and CEO of Zapier, Astra on the max setting just crushes even Fable 5.1 and Gemini saw all the other models at a comparable cost to Fable 5.1. Interesting. And by the way, if you're curious about how it compares to some of these models and practical tasks like website building, hey, subscribe to the channel cuz I'm cooking up a video as soon as I have the API access to this thing.

I'll be spending a lot of money on a lot of tests and I will present you with the results of that on this channel soon. And this last example here that I wanted to show you is acutely practical for so many people because it's a pattern and particularly I wanted to highlight this legal use case here saying it's jumped on their benchmark from 69% to 93 which is you know a massive jump and the task they gave it is reviewing an NDA against a company's own contracting policy and just reaching the right verdict in this case would not be enough. it had to actually accurately site the provision behind it.

So whereas both models declined to approve this draft which did not align only Astra pointed to the exact policy language that settled it. And GP 5.6 arrived at the same conclusion but without citing the provision and I think this one in particular is so huge. It's this ability to also verify its work. That's a big pattern that I see across all of these. It's not just making better results. It's also working with the computer and with the data and with the context to arrive at conclusions that are defensible.

Something that, you know, many people dismissed AI for up until now. >> Hey, what if it hallucinates? What if it comes up with something? >> Well, yeah, then it's going to spawn another 54 sub aents. And double check is twice, triple check is triple, quadruple check. What What is that expression? If it's 54 queen check, >> nailed it. >> That's the one. How do you come back from an argument like that? Because this is a new era.

It's not just people double-checking work. It's AI, whatever that word is, checking stuff. And that, ladies and gentlemen, is some next level stuff. So, I hope some of this inspired you to what you can do. I know a lot of people out there right now are frustrated because they don't have access and they're rolling out super slow. But look, we're in a new world where technology like this that can, you know, do all of these checks and all these different use cases and build these 3D worlds is a thing now.

Whereas yesterday it was not a thing. And you're already in the right place. You know, trying to stay in front of it, trying to wrap your head around what's possible. That's what you're doing with this video. That's what we're doing together here. And I think honestly that is the right posture in this situation. Over time, we'll discover exactly what this is going to be. super good for. I'll cover it on this channel.

So, subscribe for more. As per usual, it's my pleasure to be your guide on this journey. If we haven't met yet, my name is Igor and I hope you have a wonderful day. AI news you can [music] use. Breaking down the stories you actually care [music] about. AI news you can use. Gar's here to help you figure it