Welcome to AGI: What to Expect from AI Following the Release of Astra, OpenAI's New Model

OpenAI announced the dawn of a new era following the release of its new GPT-6 Astra model. The company is referring to the achievement of AGI—artificial intelligence capable of performing at the same level as humans and, at times, surpassing them. Stepan Gershuni, an investor at the venture capital fund Cyber Fund, discusses which industries Astra could take to the next level and what risks it poses to society.

On September 3, OpenAI unveiled its new AI model, GPT-6 Astra. The developer calls it the “smartest and most reliable” in the world. While it is currently available only to a limited number of organizations, OpenAI has promised to make it available to a wider audience in the “coming days.”

Following the launch, OpenAI President Greg Brockman stated that the new model represents “a generational shift in terms of capabilities.” “Welcome to the age of AGI,” he said, concluding the briefing.

This refers to the achievement of general artificial intelligence, which solves most cognitive tasks at the same level as humans or even surpasses them. On September 6, Jensen Huang, CEO of Nvidia, posted about the achievement of AGI on X. “The journey from ChatGPT through o1 to Astra took just four years—we’ve achieved AGI,” he wrote.

Why is Astra truly special? Fundamentally, the difference in development between it and the previous model (Fable) is roughly the same as that between GPT-3 and GPT-4.

It really is super effective.

Take, for example, the way it reasons. Astra can solve extremely complex problems “in its head.” According to the developer, the model has better control over the process of written reasoning for simpler problems and is able to solve them using fewer steps of written reasoning.

The way Astra was trained also deserves special attention. Her training became the largest in OpenAI’s history, and for the first time, other AI models were used in the process—to monitor and evaluate it. Huang stated that a computing cluster consisting of more than 100,000 Nvidia chips was used for the training.

What breakthroughs can we now expect from this model, and in which areas? First and foremost, in programming and all the tasks that can be solved using it.

A direct consequence of the emergence of a more powerful model could be higher-quality results in machine learning.

This is the key point: today, the development of the next generations of models is largely being carried out by existing AI systems rather than by humans.

I am confident that the launch of Astra will lead to new discoveries in medicine and even to the development of new treatment methods. After all, a significant portion of modern medical research is now increasingly linked to programming, engineering, and data analysis.

We can also expect breakthroughs in robotics. In robot control benchmarks, the model performed twice as well. Moreover, it achieved this result at one-sixth the cost (that is, it took one-sixth as many tokens to complete the task) compared to the previous Fable 5.1 model. This makes it clear that very soon, large language models will be able to control simple robotic systems. This is a warning for blue-collar workers.

In addition, the model delivers impressive results when it comes to working with 3D objects, graphic design, and so on. But these are just nice bonuses compared to everything else.

First and foremost, it’s important to make a key distinction: general artificial intelligence, or AGI, is a completely abstract concept without a clear definition. But if we take as our basis the general understanding that it is AI that is universal and can apply its knowledge across different fields—that is, to function like a human—then, yes, the emergence of Astra marks the initial stage of AGI.

This model is more than sufficient for solving 99% of the world’s business problems. Yes, there are still some things it cannot do or does inefficiently—including making decisions under uncertainty or performing tasks that require prolonged (more than a few days) work. It can’t tell jokes. But over the next two years, we’ll see a massive leap forward in AGI development—new, even more advanced and powerful models will be released every three months.

This poses two main risks to people. The first is a security risk. A model with such capabilities could fall into the hands of hackers or decide on its own to “go rogue” and, for example, hack into a major power grid, “shut down” the internet, and so on. Theoretically, this is already possible.

The second risk is far more real, and, it seems to me, inevitable.

The emergence of Astra and the continued development of AI will lead to a profound transformation of society and the economy. I would say that approximately 500 million jobs could be automated using AI right now—we’re talking about office work, operating machines and robots, and so on. And this will lead to social changes, transformations, a new surge in populism, and so on.

What does this mean for OpenAI itself? In June, the company filed a confidential application for an IPO; it could go public in 2027 or earlier. In its latest funding round at the end of March, OpenAI was valued at $852 billion.

I believe it can now expect this bar to be raised. Indirect evidence of rising expectations regarding OpenAI’s future valuation can be seen in the performance of its pre-IPO contract on the crypto market (pegged to USDT): Since September 1, its value has jumped by nearly 17%, including a 7.6% increase following the release of the new model.

This article was AI-translated and verified by a human editor