Hello friends! Last week Jensen Huang, Nvidia’s CEO, the leather jacket guy, posted on social media that “AGI has arrived”, because of OpenAI’s new model. And my dev friends group chat caught fire: has it? Has it not? And what is AGI, anyway?
If that question got stuck in your head too, this text is for you. I’ll explain what AGI is with no runaround, where this “it has arrived” talk came from, and give you my opinion in plain words at the end. Spoiler: it involves someone trying to sell you something.
1. What is AGI, without the alphabet soup
AGI stands for Artificial General Intelligence. The word that matters there is “general”. Bear with me, I’ll explain with a metaphor from my own kitchen.
My blender is spectacular at one thing: making smoothies. But if I ask it to fry an egg, things fall apart. The AI we use today is exactly that: a kitchen full of amazing appliances, each one great at its own task. ChatGPT writes, the translator translates, your bank’s app detects fraud.
AGI would be a different category: no longer an appliance, but a cook. Someone who learns any new recipe, copes when an ingredient is missing, and improvises dinner with whatever is in the fridge. In technical terms: a system able to reason and adapt to new problems, in any field, applying what it has learned without being retrained for each task.
Today’s AI is a specialist; AGI would be a true generalist.
2. Where the “AGI has arrived” talk came from
The hook was the launch of GPT-6 Astra, OpenAI’s latest model, which is truly impressive at coding, science, math, and that agent-style work where you delegate a task and it executes on its own. On top of that, Jensen Huang declared that “AGI has arrived”, mentioning in passing that the model was trained on more than 100 thousand Nvidia GPUs and that the next ones will need 400 thousand.
See the move? The guy selling the shovels is announcing that the gold rush paid off. Seriously, stop and think: if AGI arrived with 100 thousand GPUs, imagine what’s coming with 400 thousand… which coincidentally are for sale at his store 😛
And there’s a tasty detail: Huang himself said, a while back, that the term AGI is practically meaningless. Then his biggest customer’s model launches, and boom, AGI has “arrived”. Can you see why marketing loves this acronym?
3. Why it’s so hard to say it “has arrived”
The underlying problem is that there’s no official AGI exam, like a standardized test the model takes and walks out with a “general intelligence” diploma. Every company and every researcher uses a different yardstick.
The most cited definition, OpenAI’s own, talks about highly autonomous systems that outperform humans at most economically valuable work. Other researchers prefer to look at adaptability: solving genuinely new problems that weren’t in the training data. And there are serious people saying the right question isn’t even “has it arrived?”, but “how much is missing in each area?”, because progress is uneven: models are already monsters at math and code, and still stumble on long real-world tasks.
I’ve shown that unevenness in practice when I wrote about AI replacing programmers: on a well-defined task, AI flies; everywhere else, it still needs a human holding the wheel.
Without an agreed yardstick, anyone can be the first to shout “it’s here”.
4. Okay, but bottom line: is it here or not?
My opinion, in plain words: no, AGI has not arrived. What arrived was an excellent model, maybe the best ever released, and along with it a billion-dollar interest in declaring victory. The price and narrative war between AI companies (in Portuguese) explains a lot of the noise: announcing AGI sells GPUs, sells subscriptions, and sells headlines.
That doesn’t diminish the progress, which is real and impresses me every single week. But an AGI declaration in a social media post, made by someone who profits from it, is worth what it’s worth. If AGI ever truly arrives, we’ll probably only be sure looking back, calmly, and not on launch day.
Meanwhile, the cook doesn’t exist yet; it’s the kitchen that keeps getting better.
5. Frequently asked questions
What’s the difference between AI and AGI?
Today’s AI is specialized: each system solves one kind of task well. AGI would be a generalist system, able to learn and solve any intellectual task, including ones it has never seen, at the level of a competent human or above.
Is GPT-6 Astra an AGI?
No. It’s a very advanced model, with strong results in code, science, and agent work, but it remains an AI with strengths and weaknesses, not a general intelligence. The AGI declaration came from an executive with a direct commercial interest, not from an independent evaluation.
Is AGI the same as AI becoming conscious?
No. AGI is about the ability to solve general tasks, not about consciousness or feelings. They are different discussions, and mixing them up is the favorite shortcut of anyone trying to scare you or sell you something.
When will AGI arrive?
Nobody knows, and be suspicious of anyone naming a date. Expert predictions range from a few years to decades. The most honest path is to track concrete capabilities, area by area, instead of waiting for a triumphant announcement.
Real AGI won’t need a post to announce itself.
Cheers, Fellipe Soares
6. References
Subscribe
Get the next articles by email.