Georgi Kostov · Studio Kostov

I build apps, games and interactive experiences. Sharing the work behind Studio Kostov. Unity developer, educator and game designer.

[Rishabh Karnawat @rishabhkio]

Many of you don't understand how serious this is, so let me break it down.

In computer science there is a famous kind of puzzle called 3-SAT. You are given a long list of yes-or-no switches and a set of rules about how they must be arranged, and you have to find a combination that satisfies every rule. The obvious way to solve it is to try every combination, but each new switch doubles the number of possibilities, so a puzzle with a few hundred switches would take longer than the age of the universe to brute force.

For decades, the smartest people in the field found clever tricks that made this a little faster, but every trick still hit the same wall: the time required kept exploding as the puzzle grew.
About 25 years ago, two computer scientists, Russell Impagliazzo and Ramamohan Paturi, working with Francis Zane, turned that wall into a formal idea.

They proposed the Exponential Time Hypothesis, which says this explosion is not a failure of human cleverness but a law of nature. No algorithm, however brilliant, will ever solve 3-SAT without the time blowing up exponentially. Nobody could prove it, but almost everyone believed it, and it became one of the foundations of modern computer science.

Researchers used it as a measuring stick. If you could show that solving some other problem quickly would also crack 3-SAT quickly, you had shown that problem was hopelessly hard too. Hundreds of papers, across scheduling, graphs, logic and much more, rest on this one assumption.

Now an internal AI model at OpenAI claims to have broken it. It says it found an algorithm that solves these puzzles meaningfully faster than the hypothesis allows, and it has checked the proof in a computer verification system called Lean.

And here is the remarkable part: the very people who proposed the hypothesis, Impagliazzo and Paturi, along with MIT's Ryan Williams, are now leading the public review of whether a machine has disproved their own idea.
So what happens next? If the disproof holds, a large part of theoretical computer science has to be revisited.

Every result that said "this problem cannot be solved faster, assuming ETH" loses its foundation, and researchers will have to work out which of those walls were real and which were simply where we stopped looking. It does not mean everything becomes easy. The new algorithm is still slow in absolute terms and this is not the famous P versus NP problem being solved.

But it means our map of what is possible was wrong in a place we were very confident about, and it was an AI that redrew it. If the disproof fails, the hypothesis survives and the field breathes out, but we will still have watched an AI produce serious, checkable work on one of the deepest questions in the discipline, closely enough that the founders of the idea had to take it seriously. Either way, something has shifted.

[Georgi Kostov · Studio Kostov @StudioKostov]

Powerful and true. That's why I've been building https://t.co/9NAvil2aiN to preserve the most precious memories and life stories.

[Horse @TheFlowHorse]

Son of a bitch this stings.

[Georgi Kostov · Studio Kostov @StudioKostov]

@BitmundFreud Used to make parties like that irl, legendary times 😄

[Jay Anderson @TheProjectUnity]

Gandalf is arguably the greatest character in cinematic history, and the older you get, the more you understand why.

He is one of the most powerful beings in Middle-earth, yet he chooses humility over domination.

He could walk among kings, but prefers the company of Hobbits.

He possesses ancient wisdom, but never pretends to know everything.

He inspires courage without demanding obedience.

He confronts unimaginable evil, not because he believes himself invincible, but because someone has to stand against it.

He understands that the fate of the world is not determined solely by armies, kings or great powers, but by the seemingly insignificant acts of ordinary people.

He sees greatness in those who cannot see it in themselves.

He falls into the abyss fighting a demon from the ancient world, sacrificing himself so that his friends might live.

And when he returns, more powerful than ever, he remains fundamentally unchanged in character.

No arrogance. No hunger for power. No desire to rule.

Just wisdom, compassion, humour, courage and an unwavering commitment to what is good.

And perhaps his greatest lesson is that evil is not ultimately defeated by becoming more powerful than it, but by refusing to become like it.

Gandalf isn’t simply a great fictional character.

He’s a representation of everything humanity should aspire to become.

[Georgi Kostov · Studio Kostov @StudioKostov]

Some people will call it slop, lack of human touch. Others will realize it's the greatest democratization of game design, freedom from corporate control. A new age of ideas!

[Doug McNabb @mcnabbd]

i retired from making video games several years ago. i hated it. i love everything about the craft. but the industry absolutely sucks. getting paid to work on a game is a dream come true. but the teams were so big that they have to be activel managed. and then the managers rule the world and its all over.

consider ai as the solution to the problem.

[Georgi Kostov · Studio Kostov @StudioKostov]

@hthieblot StarCraft, WarCraft 3 TFT, KOTOR

[Thierry from arvy 🇨🇭 @ThierryBorgeat]

This is it.

Everyone is excited about AI.

I'm not.

AI is the biggest commoditizer of human capability we have ever built. Doctors earn more than cleaners because expertise takes a decade to acquire and judgment to apply. AI is going to make most of that expertise available to anyone, at marginal cost.

The gap doesn't go to zero. But it shrinks. And the same thing happens in almost every knowledge profession. Programmers. Lawyers. Analysts. Traders. Designers. The taxi driver replaced by a robotaxi is the easy case. The harder ones are people who spent twenty years training to do something a model can now do in a second.

The one corner of human life I don't see this hitting is sport. Nobody pays to watch a robot beat a human. The whole appeal is humans competing against each other. That's why nobody watches Deep Blue play chess anymore.

From an investing perspective, the next decade plays out in three phases.

Phase 1 is the buildout. Hundreds of billions in capex on chips, data centers, and energy. This is what's driving today's market.

Phase 2 is the deployment. Productivity rises sharply, then deflation sets in as the cost of knowledge work collapses. Jobs disappear faster than the economy can replace them.

Phase 3 is the social adjustment. Capital becomes more powerful than labor. The biggest winners are likely to sit where AI cannot reach. Real assets. Energy. Commodities. Precious metals. Scarce land.

And there's something deeper than the economics. The Japanese have a word for it: ikigai. The reason you get up in the morning. Your vocation. Your purpose. The work that makes you feel needed. AI is coming for that too. Nobody has yet figured out what a healthy society does when most of its members no longer feel needed by anything.

I might be wrong. I hope I am.

If I'm not, the world is about to change faster than most people are ready for.

[Georgi Kostov · Studio Kostov @StudioKostov]

@OpenAI This reads like out of a sci-fi book. Using frontier models to contain next gen models, wtf are you guys up to 😂

[Georgi Kostov · Studio Kostov @StudioKostov]

@geoffkeighley 6 years later mass effect was still in early development, wtf?

[Georgi Kostov · Studio Kostov @StudioKostov]

@johncrickett nvidia just buys them up ;)

[Jacob Navok @JNavok]

If you have not spent the last 1.5 months with Astra or Opus 5.5 in daily dev you may have a skewed view of what their capabilities are based on social media reading.

I'll break down why there is such a big wave happening in games as a result of these specific models, which are substantially better than what had come before.

Coding a game requires an understanding of the language you are coding in, the structure of the code you want to write, the logic by which you lay down the code, the architecture it sits in, the components it connects to / affects, and if you are writing in an engine, the engine and its editor and plugins.

Earlier this year some of these components- code structure and language- were good, but the LLMs lacked good logic, components, and editor understanding. Plus many of the systems lacked good MCP support; UE didn't add a lot of theirs until July, for example, and even then it required a lot of engine restarts.

In just the last few months most of this had been fixed. Astra was the first model, launched in the second half of August, where I felt like it had the understanding of Unreal at the level of a mid-career programmer, but it blew away tokens faster than Scarface.

Then we got Opus 5.5 in the second half of September and it had the capabilities of Astra 6.1 but at 1/5 the token cost, meaning I could go several days before hitting my weekly usage cap.

Not only did the model properly understand how Unreal, Godot, Threejs and other systems worked, but the editors actually played nice with them. More importantly, the logic and structure of their code was nearing senior dev, and if you put it at the highest settings, could actually develop strong architectures, though it could still jump to wrong assumptions if it misunderstood you.

Anyone complaining about bugs is probably misunderstanding an actual bug for incorrect logic. I find it very, very rarely makes standard bugs (like forgetting a character or putting the wrong string in, something that my best devs still do all the time).

The logic issues will get better as time goes on and the models self correct. They will always be limited by an understanding of what you want to build, but they are superior at the speed by which they can code and test their own coded.

Concurrent with this the recent models became excellent at Blender usage. This means they can adjust models and animation themselves, without you needing to do a lot of the legwork. This does not mean they are great at them yet; I find its animation quality to be at the level of a junior animator. It still lacks understanding of key body parts and smooth movement. But the path to getting there is clear and I suspect this will be solved within six months.

They are also still bad at lighting and visual/gameplay fidelity feel. I suspect lighting it will get better at, I think gameplay feel may take a long time.

These components combined are what has brought us to our current point.

[Georgi Kostov · Studio Kostov @StudioKostov]

@davidkaye Definitely agree, but selling is already nightmare level and it will get worse 😄 it's good to be an influencer

[Georgi Kostov · Studio Kostov @StudioKostov]

@BeardWhoCodes @hany99dev Just used to it since i had 200 last month 😂 switched to sol still drains super fast

[Tim Sweeney @TimSweeneyEpic]

When I was writing Unreal Engine 1, my time was divided as follows:

- Around 30% went to seriously deep coding on hard problems: assembly language texture mappers, C++ BSP trees, data persistence, object oriented systems design, and networking.

- Around 70% was just typing and refactoring large bodies of code needed for a complete engine feature set. This was not hard, and while I was doing it my mind was mostly on the deeper coding topics, thinking and planning for what's next. It didn't feel much different from vibe coding today, except that it was just me typing code on an IBM Model M keyboard rather than prompting it and reviewing the results.

So, it has always seemed perfectly logical that we would rely on AI for that 70%, once the AI could do it with a similar quality standard as we could, which has never been perfection. It came as no surprise that, after a few years of AI model evolution, it was able to take the reigns on that 70% and free up programmers' time for more valuable tasks.

What's more of a surprise is the emergence of AI as a sort of insta-Carmack able to solve hard technical problems quite cleverly and often cleanly, and that it's able to reason about and work in complex 3D scenes both by operating the tools and by directly understanding the raw data files.

I abandoned assembly language in 1998 when out-of-order CPUs made much simpler C/C++ code faster, and never I missed it. If prompting through architectural documents, concept art, and design plans can transcend typing code and tweaking vertices, that will enable developers to spend their time on higher-level, higher value technical and creative tasks.

What matters is the end product, whether it's a game or an engine. Anything developers can do to try out more ideas and make better software and games faster is a win for everyone.

[Aaron Levie @levie]

The biggest misread of anyone predicting the end of software engineering was imagining that the world only needed a certain amount of software, and bringing automation to the field would reduce the need for labor.

Exactly the opposite is happening where essentially most things are actually turning into software. Because agents can use tools, work with computers, write software on the fly, and process data like humans (but far faster), we can now throw “software” at problems that never could reasonably be digitized before.

What that leads to is the role of engineering taking on greater importance in the economy, because we can now engineer far more systems and processes than were possible before.

If you’re a bank trying to do new forms of trading, engineers are often your first bet. If you are in life sciences, you expect more and more research and lab work to be engineering centric. If you’re in manufacturing, you’re trying to automate more of the supply chain, all that is engineering. More and more work is shaped like software problems.

Yes we’re not going to be coding manually in the future. But almost everything will be code, which means those that understand it are in a fantastic position.

[Georgi Kostov · Studio Kostov @StudioKostov]

@ElitzaVasileva People to a lot of computer work, 3D, video, costs more tokens. If you do web stuff it's quite cheap on tokens. Just start a side project where you can dump your extra tokens :D

[Georgi Kostov · Studio Kostov @StudioKostov]

Is 20$ sub broken for anyone? I gave 2 prompts to Astra and 5h limit vanished instantly. It looks like a bug or something fishy, since it barely started working and the limit was gone. I used the 20$ Claude sub for a while and with Opus I can do quite a bit of work before the 5h limit runs out.
I had 200$ openai last month but switched to claude now, and sorry to say it I've been using both a lot and Claude is currently a lot better, faster and quality of output is next gen compared to Astra. I can't even believe I say that since Astra was already shockingly good in the beginning of september and blew my mind. Then everything changed in a month.
I hope ChatGPT strikes back soon, right now it seems they have some big compute constraints.

[SightBringer @_The_Prophet__]

⚡️AI is destroying the economic relationship between discovering something valuable and owning the value of that discovery.

For centuries, intellectual advantage was protected by the difficulty of reproduction. A company could spend decades developing proprietary algorithms, accumulating expertise, and refining technology because competitors faced enormous costs to replicate that knowledge.

AI is dismantling that protection.

An innovation that required years of specialized engineering can increasingly be reconstructed in hours.

The original developer bears the cost of discovery while the next competitor pays a fraction of the cost to reproduce its functionality.

The market doesn’t compensate anyone for how difficult something was to invent once alternatives become cheap to produce.

This changes the economics of entire industries. Intellectual property loses pricing power, competitive advantages decay faster, and businesses built around proprietary knowledge face relentless margin compression. Even AI companies will eventually confront the same force they are unleashing, as increasingly capable intelligence makes competing models and systems cheaper to develop.

The deeper shift is that intelligence is becoming abundant while the resources required to turn intelligence into economic power remain scarce.

Energy, computing infrastructure, physical production, capital, distribution, and control over real-world systems become increasingly important sources of economic leverage.

And the same mechanism reaches human labor. A professional can spend fifteen years acquiring expertise that once commanded extraordinary compensation, only to discover that machines can reproduce an increasing share of that expertise at a fraction of the cost.

For generations, superior knowledge was one of the most reliable paths to wealth and independence. That relationship is breaking as intellectual capabilities become reproducible at scale.

The coming economic hierarchy will increasingly favor those who control the scarce resources through which intelligence becomes productive.

The greatest fortunes of the AI age will accrue to those who control the bottlenecks that intelligence cannot eliminate.

The ultimate scarcity is migrating from knowing how to do something to controlling what is required to make it happen.

[Georgi Kostov · Studio Kostov @StudioKostov]

@theo Same, it's ridiculous