- GPT-6 Astra has arrived, and it's being called AGI
- This includes by Nvidia CEO Jensen Huang
- Huang's celebratory X post also promotes Nvidia's GPUs, saying we need to get 400K of them online
Over the weekend, Nvidia CEO Jensen Huang took to social media to celebrate OpenAI’s latest achievement — “AGI has arrived” in the form of GPT-6 Astra. And of course it was developed using over 100,000 NVIDIA GPUs.
According to OpenAI, Astra represents a "significant step forward" for AI, with OpenAI's president Greg Brockman also saying "welcome to the AGI era" as Astra was debuted — with the company promising that GPT-6 boasts incredible aptitude at tasks across software engineering, science, mathematics, and professional workflows.
It’s also said to be much better at agentic work, meaning you can delegate tasks to GPT-6 with less fear that the AI will misunderstand your intent as it works independently.
We haven’t yet given GPT-6 a full shakedown — the update is steadily rolling out to subscribers — but if OpenAI’s claims live up to the hype, Astra could be quite the capable tool, and may actually live up to those early AGI claims, even if this isn’t the first time Jensen Huang has heralded the arrival of AGI.
But what is AGI, and why does it matter if Astra is AGI or not? Well, as Huang himself suggests, AGI matters because it’s a reason to quadruple the number of Nvidia GPUs OpenAI is using.
What is AGI?
Let’s address the elephant in the room. AGI, or artificial general intelligence, isn’t a well-defined term, so judging if we have or haven’t achieved AGI is a bit like judging if a piece of art is a masterpiece or not.
Broadly speaking, AGI is an artificial system that can beat humans across any task that involves thinking or reasoning. Crucially, it should also be able to adapt to new problems by applying what it has learnt in training. For example, an AGI could learn a new skill to overcome a challenge posed to it without being retrained or reprogrammed — it just applies what it knows to evolve itself.
Think J.A.R.V.I.S from Marvel. It’s a very powerful assistant that helps Tony Stark across all fields, even how to deal with otherworldly threats like Loki, without needing to be trained on each individual problem.
Does AGI matter?
Okay, but does it matter if it’s AGI or just a very, very smart AI?
For most of us, not really. An AI doesn’t need to be very good at everything to be a majorly useful tool. It doesn’t matter if your AI doctor can understand the deep complexities of the vast universe, it just needs to be able to diagnose your illness and suggest treatment. An AI lawyer doesn’t need to be able to comprehend five-dimensional topology to defend you in court.
It's not just me saying that; Jensen Huang recently admitted that AGI doesn’t matter — this was before his social media praise for OpenAI — noting that the term is effectively meaningless.
However there’s one important reason for achieving AGI, and it’s about reaching what comes after: ASI, or artificial superintelligence.
ASI would be an AI no human could ever hope to match, let alone beat. It would be able to grapple and solve problems humans aren’t capable of even understanding. AGI is distinctly not ASI, but its ability to self-train and develop tools for itself suggests that an AGI will eventually construct ASI through a process called recursive self-improvement.
For Hitchhiker’s Guide the Galaxy fans, it’s like how the supercomputer Deep Thought couldn’t find the ultimate question of life, the universe, and everything, but could build the computer which would be able to find the question.
Perhaps more importantly, AGI is an investment buzzword, and there’s more than a hint of that in Huang’s celebratory post. Because while he says Astra is the product of 100,000 Nvidia Grace Blackwell NVLink72, he adds that if we want to see bigger and better improvements we’ll need to bring 400,000 GPUs online next.
To do this OpenAI and Nvidia will need a heck of a lot of investment, and will also need to build more data centers, just as communities across the US and rest of the world are pushing. back against their construction.
The arrival of AGI could represent the last opportunity for investors to make it big in AI, or any stock, before ASI takes over productivity. So they need to invest as much as they can now (or keep their investments in AI) before they’re left behind. For data centers, the big concern is they are pollution nightmares that merely take from a community with no promise they’ll actually succeed long enough to give back. The advent of AGI would suggest that the AI hype has been justified all along — the bubble won’t burst, and local communities could finally reap the reward of accepting data centers.
I’m not saying Huang or OpenAI are definitely misleading us, or even just being hyperbolic. It's possible that Astra is in fact our first taste of AGI. I’m just pointing out that companies and CEOs have a massive incentive to claim that we’ve achieved AGI, even if they later make the same claim about the next new model — I won’t be surprised if GPT-7 is later called 'the debut of AGI'.
My point is that we won’t really know if or when we achieved AGI until much later, when we can look back with hindsight and see when our AI models reached a turning point in their ability to evolve.
In the moment, though, we will know if an AI is useful and can provide some genuine benefit to us. We’ll do that as a collective, as individuals and businesses not directly associated with OpenAI get their hands on the new model and determine whether or not it makes their life significantly easier. Try Astra out if you can, learn about the experience of others with the tech, and decide if you think it’ll be worth paying to be able to access it.
Don’t get caught up in the hype, and don’t worry about whether it's AGI or not. In time it’ll be very clear if it is — and we won’t need a social media post to convince us.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
