Speaker A
What do you think a neural network would do if you gave it your credit card and asked it to build you a gadget? It would buy the wrong thing, spend all your money. A guy in September decided to test this out. He gave his
Explore 10 incredible real-world tasks accomplished by GPT-6 Astra, from medical simulations to music creation and financial trading.
Key Takeaways
- GPT-6 Astra significantly enhances productivity by automating complex, hands-on tasks across diverse fields.
- While powerful, Astra does not replace experts but acts as an advanced assistant, increasing efficiency.
- Astra's cautious decision-making in finance shows AI's limitations and ethical boundaries in autonomous actions.
- The technology is already impacting medicine, music, programming, construction, and robotics.
- Real-world AI applications require human oversight and collaboration for best results.
What the video covers
- GPT-6 Astra can autonomously interact with software, control mouse and keyboard, and complete complex tasks without coding input.
- An engineer with a sore ankle received a custom interactive 3D anatomical model of his ankle created by Astra.
- A Japanese doctor developed a near-perfect cardiology simulator with Astra, improving medical training accessibility.
- A cardiac surgeon used Astra to quickly generate 3D heart valve models, enhancing his surgical planning efficiency.
- An American music producer tasked Astra with creating a beat, which it composed and refined in 30 minutes.
- A 16-year-old programmer had Astra autonomously earn money by fixing bugs in open-source projects while he slept.
- An American blogger gave Astra $10,000 for trading; Astra devised strategies but refused to execute trades itself.
- Astra's cautious trading slightly underperformed a passive index investment over one week.
- A neural network was used in construction to provide honest cost estimates, valued for admitting uncertainty.
- Astra piloted a drone autonomously in an office, tracking a person, a feat noted by a New York Times journalist.
Chapters
- 00:00Introduction and Credit Card Experiment
- 00:54Overview and $10,000 Trading Story Preview
- 01:403D Ankle Model for Engineer and Athlete
- 02:21Human Body Interactive Model Popularity
- 03:01Japanese Doctor's Cardiology Simulator
- 03:45Cardiac Surgeon’s 3D Heart Valve Model
- 04:35Music Production with Astra
- 05:17Earning Money by Fixing Bugs
- 05:59AI Trading with $10,000
- 06:51Trading Results and Reflections
Full Transcript — Download SRT & Markdown
Speaker A
credit card to GPT-6 Astra, and five days later he received an email from a Chinese factory. He didn't understand it himself, but the AI understood and replied to the factory on its own. In previous videos, I already talked about
Speaker A
the 10 strangest things Astra has done in a month. Then I talked about 10 games that the neural network made by itself. But today, there will be no horror stories. Ten cases where ordinary people gave Astra a real task, and it
Speaker A
actually did it. I went through hundreds of posts and videos from America, Japan, and Europe, and gathered them so there's something here for every profession: doctor, musician, builder, scientist, and even just someone with a sore leg. Along the way,
Speaker A
you'll find out how the credit card story ended. And I'll also show you what Astra did when it was given 10,000 dollars in real money. I honestly didn't expect that. And let the last case be at the end, because it gave me
Speaker A
goosebumps. It involves a drone in a regular office, and even a New York Times journalist joked about it. And, of course, hit that subscribe button in advance. And if you don't like it, you can unsubscribe. Let's go. First, I'll
Speaker A
quickly catch you up to speed. OpenAI released Astra in early September. The main thing it does better than previous models is working with its hands. It moves the mouse itself, opens programs, clicks, types, and can work for hours
Speaker A
until it's finished. That's why in almost all the cases I'll show, the person didn't write any code. They just said what they needed, and then either watched or went to sleep. And the neural network did everything itself.
Speaker A
Let's start with the most mundane. An engineer and athlete, Emmanuel, had a sore ankle. I think this is a fairly common problem for everyone. He didn't bother googling symptoms, but asked Astra to figure out what was going on
Speaker A
inside it. And in one session, Astra built him an interactive 3D atlas of the ankle: bones, ligaments, tendons.
Speaker A
And you can rotate it all. You move the slider, the foot bends, and on the side, there's a prompt showing which ligament is stretched. This guy only has 2,000 followers, but his post was viewed a million times. And then people
Speaker A
took it further. One developer made a site with Astra where the entire human body is broken down into more than 2,000 parts: muscles, vessels, nerves.
Speaker A
You can pull each one out and take a look. That post got 8 million views.
Speaker A
And as far as I searched, it's actually the most popular post about Astra that isn't about games. Yes, she didn't diagnose him, and she didn't invent the anatomy herself. She gathered everything from open medical databases.
Speaker A
But you remember anatomy textbooks from school, the pictures and all. And here, it’s done in an evening, customized to your own leg. And now, about what those who truly know anatomy are doing with this. Let’s move on to the
Speaker A
Japanese doctor. He owns a small company and had long wanted to build a simulator for cardiologists. Let me explain why it’s needed. There are heart surgeries that don't require incisions. The doctor makes a tiny puncture in a vessel, in the arm or
Speaker A
groin, and inserts a thin wire. It travels through the arteries straight to the heart. He cannot see the heart itself. In front of him is only a screen, like an X-ray. And he guides the wire blindly based on that image.
Speaker A
Make a mistake, and you damage the vessel. That’s why it takes years to learn. And a proper simulator for this is serious medical equipment that hospitals pay a lot of money for. And far from every clinic even has such a
Speaker A
simulator. So, this Japanese doctor writes that with previous models, he managed to reach about 80% of his goal.
Speaker A
By the way, that was Claude back then. But with Astra, he achieved, I quote, 99%. The next day, he added another scene: septal puncture, heart under ultrasound, and released it all for free into the public domain. The story
Speaker A
gets even more interesting from here. This simulator, which we mentioned earlier, was spotted by a cardiac surgeon who is the editor-in-chief of a specialized journal. He decided to test Astra with his own work and, in 27 minutes, obtained a 3D model of a
Speaker A
neutral valve, both healthy and diseased. It’s a heart valve that allows blood to flow in only one direction. At first, the surgeon wrote that it was insane how fast it was, but then corrected himself, saying, "I've performed over a thousand such
Speaker A
surgeries, and for a model to be suitable for treatment, it must be verified by an expert." But previously, for such a model, I needed a separate specialist, and now I can do what I see in my mind all by myself. That is how,
Speaker A
in my opinion, one should look at this. It didn't replace the doctor. It simply gave him extra hands that he lacked before, and consequently, increased his efficiency. Now we leave the doctors and head to the American channel Busy
Speaker A
Works Beats. He teaches people how to make beats. And so he gave Astra a folder of his drums and a task to make a beat in the style of Bryson Tiller.
Speaker A
And then he just sits there, not touching the computer with his hands, but watching. Astra opens the music software, finds a synthesizer, and scrolls through presets all by itself.
Speaker A
It composes the melody, lays out the drums, and then renders it. It even notices that the kick drum in his own sample is clipping and fixes it. And the human stepped in just once, when it chose the wrong bass. All of this took
Speaker A
about 30 minutes. And the author says a phrase: "I remembered it: someone spends 2 years learning to make beats, and this was done in 12 minutes." Well, okay, this is all beautiful, of course, but the most important and interesting
Speaker A
question is: where is the money in this whole story? Now we move on to sixteen-year-old programmer Bilal Bakr.
Speaker A
He gave Astra one task: try different ways until you earn at least some money. And he went to sleep. Astra worked for over 16 hours without him. It found open-source projects that pay a bounty for fixing bugs. It solved these tasks,
Speaker A
corresponded with project owners, and sent fixes, after which he had 200 dollars in his account. And this is why I respect him. He later added himself that there were hundreds of attempts.
Speaker A
Most were rejected or simply ignored, only three worked out. So, there is no "get rich quick" button here. It's more like an intern who spent all night sending out resumes and got three responses. But it sent them out by
Speaker A
itself while its owner was asleep. And next, there's something I didn't expect myself. At the beginning, I promised a story about 10,000 dollars. American blogger Nethverk gave Astra 10,000 real dollars for a week of trading. The task was simple: to beat the index, that is,
Speaker A
just normal market growth. Astra built a strategy itself, scheduled when to wake up and check the market. It created a trade log, but on the second day, it stopped and wrote to him, saying, "I'll help with the strategy,
Speaker A
but I won't execute trades for you myself." He tried to persuade it, but couldn't convince it. It stood its ground. So what did he do then? He connected a separate bot that just pushes the buttons. Astra became the
Speaker A
brain, and the bot became the hands. And what follows is funny. The first two days, Astra traded so cautiously that it earned 21 cents in a day. Then they hit a rate hike, and the portfolio dipped. And a week later, there was
Speaker A
9,901 dollars in the account. But if he had just put the money in the index and done nothing, it would have been 9,980.
Speaker A
It turns out that the smartest model in the world lost to the option where you do nothing. Frustrating, isn't it? A week is not long enough. And he himself says that last time, it beat the index in a month. But that moment where the
Speaker A
neural network said, "I'll suggest the strategy, but I won't carry out trades with your money myself." To me, it
Speaker A
It set a boundary itself and defended it. But this isn't the only time Astra says "No." And in this case, it proved useful. Imagine you are building a house and need to calculate how many windows, tiles, pipes, and so on you'll
Speaker A
need. This is called an estimate. It is calculated based on a blueprint. If you make even a 1%error, you can end up significantly overpaying. This is the task a blogger who talks about neural networks in construction gave to Astro.
Speaker A
It was an actual architectural blueprint where he erased the labels with apartment areas so the neural network could calculate them itself. It did the calculations, but not for everything. For some of the apartments, it honestly wrote that it wasn't sure.
Speaker A
On the drawing, furniture was placed behind the glazing line, and it was unclear if it was included in the area or not. In other words, it didn't just guess; it found an inconsistency in the drawing itself. And where it did
Speaker A
calculate, its error was less than 1%. Claude gave answers for all those same apartments and made a bigger mistake.
Speaker A
And I understand why this example spread so quickly among builders. For this kind of work, a neural network that honestly says "I don't know" is more valuable than one that lies confidently. Now for the story I started with. A guy nicknamed Animem,
Speaker A
who has a small startup, wanted a mini DJ controller in the style of Teenage Engineering. These are expensive, beautiful gadgets for musicians. He essentially gave Astro his credit card and the task of recreating it. Astro designed what the controller would look
Speaker A
like. It selected the parts, read their Chinese technical specifications, built a 3D model of the casing, ordered everything, and even made an animation of how to assemble it later. And five days later, the Chinese factory—whose name I won't read, but will show on
Speaker A
screen—sent an email with a question about the printed circuit boards. The author honestly writes that he didn't understand the questions himself. But the most important thing is that Astro understood these questions. It created assembly drawings with dimensions, refined the gaps, and answered the
Speaker A
questions itself. Granted, I haven't seen a working controller in his hands yet. The parts are still on their way, so we will find out how it ends later.
Speaker A
But you understand what this is about. Before, to order a custom circuit board from a factory, you had to be an engineer; now, you literally just need a credit card and Astro. Well then, while the controller is on its way, we
Speaker A
have something next that you can actually touch. An ordinary guy with a small account, Souri Sharma, took an MRI scan of his head, gave the files to Astro, and it prepared a life-size model of his brain including the
Speaker A
cerebellum and a piece of the spinal cord. And then, an ordinary home 3D printer, which many people have at home , was put to work. The printing took 35 hours and used 400 grams of plastic.
Speaker A
And there he is, sitting and holding his own brain in his hands. A brain resting in his palm. A brain that was previously just a file from a medical scan. But let's be honest, people already knew how to turn a scan into a
Speaker A
printable model before. But that required knowing special software, while here he just asked the neural network, and it did everything for him.
Speaker A
Let me know in the comments, would you print one for yourself? I probably would, and I’d put it on my shelf.
Speaker A
Before we move on to the last two cases , I’ll tell you why I even put together these compilations. My channel's goal is simple: to share the most important things you need to know about AI, both good and bad, so you
Speaker A
understand where it's all heading in real-time. Science fiction is becoming reality much faster than we can imagine . What seemed impossible yesterday can now be installed on an iPhone today.
Speaker A
And if you find this interesting and necessary, please hit subscribe. It's free and takes just a second. If you hit it, thanks. In return, I’ll keep digging through hundreds of posts and checking every figure myself. And if you get tired of it, you can
Speaker A
unsubscribe at any time. Let's move on. Now for the second-to-last case about science. Jake Bruchman and his colleagues spent a whole month trying to prove a theory. It’s part of a famous unsolved problem in graph theory , the Seymour hypothesis. Don't ask me
Speaker A
what that is. I read the definition three times myself, but I didn’t really understand it. He left Astra running overnight. And overnight, it finished the next part of their work and proved the theorem they had been struggling with for a month. And here
Speaker A
is my favorite detail. It didn't wake him up. He writes that it apparently didn’t think it was anything important and just kept working.
Speaker A
Imagine, you can't solve a problem for a month, and a colleague solves it at night without even texting you. Not only that, it found a way to make the proof from their previous paper just one paragraph long. And Bruchman writes
Speaker A
that this is the first time he's seen a model show creativity in the method itself. It tries different approaches like a mathematician and finds what works. It is important to note that the result hasn't been published yet, they
Speaker A
are still verifying it, but in his own words, there's already enough for a new paper. And the last case is the one that gives you goosebumps. The Laps lab let Astra pilot a drone in their office . The task was stated in one sentence:
Speaker A
"Find a person and follow them." And Astra does it. It pilots the drone through the office, finds the right person among others, and flies after them. New York Times journalist Kevin Roose quoted this video with an old internet joke. Like, congratulations,
Speaker A
you've built an autonomous drone pursuer from a classic horror novel. Well, you get what I’m talking about, right? We’ve watched this in movies for years, and now it's just a Twitter post between memes. And the lab replied they did it on purpose so people would
Speaker A
debate where we need AI and where we don't, based on facts, not movies. The fact is, they have five tasks for the drone. And Astro became the first neural network to beat a human pilot in at least one attempt for each of the
Speaker A
five. So it turns out that in nine out of ten cases, people finally did what they dreamed of, from medical simulators to their own controllers.
Speaker A
And the tenth one we used to see only in the movies. It’s what we were afraid of, and now it’s become a reality. Uh-huh.
Topics:GPT-6 Astraneural networkAI automationmedical simulationmusic production AIAI trading3D modelingdrone AIopen-source bug fixingAI productivity
