Yooo, I think people might be looking at AI completely wrong.
We always ask which AI is smarter.
ChatGPT? Gemini? Claude?
But NVIDIA just showed us that the AI model itself is only part of the story.
THINK OF AI LIKE A BRAIN
You can have the smartest brain in the world, but what happens if I keep making you forget what you were doing?
Or I don’t give you the right tools?
You might be smart as hell and still struggle to get the job done.
That’s basically what researchers are figuring out with AI.
The model is the brain.
The system around it controls what the AI remembers, what tools it can use and whether it can learn from mistakes and keep working.
That system is called a “harness.”
NVIDIA JUST SHOWED WHY THAT MATTERS
NVIDIA built a system called AVO and connected it to Claude Opus 5.
Then they put it through ARC-AGI-3.
Think of it like dropping somebody into games without telling them the rules or even what the goal is.
You gotta try something, see what happens, remember it and figure it out yourself.
NVIDIA’s system completed all 183 levels across all 25 environments in the public test.
100%.
Now before somebody says NVIDIA figured out AGI...
They didn’t. ?
This was one specific public test.
HOW they got 100% is what matters.
The AI could try something, see what happened, remember it, change strategies when it was wrong and keep going.
NVIDIA even had AVO work continuously for SEVEN DAYS on a real GPU engineering problem, exploring more than 500 different ideas.
THIS EXPLAINS SOMETHING I’VE NOTICED
I recently talked to Namecheap’s chat support and didn’t even realize I was talking to AI until it asked if I wanted a live agent.
Does that mean Namecheap’s AI is smarter than ChatGPT?
Nah.
It shows what happens when AI has the right setup for the job.
With mainstream AI, sometimes you get exactly what you need.
Other times you gotta ask five different questions before you finally get there.
The brain might already be smart as hell.
The setup around the brain matters too.
OPENAI SAW IT TOO
OpenAI tested GPT-5.6 Sol on ARC-AGI-3 and improved how the system kept its previous reasoning and managed context.
Its score jumped from 13.3% to 38.3%.
Same AI model.
Better setup around it.
THAT COULD BE THE REAL GAMECHANGER.
Imagine instead of asking AI 10 questions, you simply tell it what you need DONE.
It remembers what it tried.
Uses its tools.
Learns when something doesn’t work.
And keeps going until the job is finished.
NVIDIA didn’t figure out AGI.
But they showed us something important:
Making AI better isn't only about building a smarter brain. You gotta build the right system around the brain too.
Source: NVIDIA — AVO Reaches 100% on ARC-AGI-3
Source: OpenAI — How Two Settings Tripled Our ARC-AGI-3 Scores
— King Coop
Content & Social Media Manager
