Many of us have heard the term “superintelligence” or Artificial General Intelligence (AGI). It represents the next generation (whenever that arrives) of AI. The expectation is that AI will progress to became an all-knowing, all-fixing entity that cures diseases and solves society’s ills.
Until recently, there was consensus. The major players, Google, OpenAI, and the rest, were marching toward the same destination: a singular, omnipotent mega-brain. A scarce resource, gated behind an API for those willing to pay.
Fork That
Enter Meta, or more accurately, Mark Zuckerberg.
Meta was following the same path, but they’ve reassessed. Their version of AGI isn’t a singular mega-brain we all bow down to while keeping one eye on the power cord. No, Meta believes we should all have a mini-AGI in our own hands. One we can run locally for our own “personal” needs.
The difference is stark. The singular AGI model relies on “safety-first paternalism.” The Meta version relies on empowerment, proliferation risks be damned.
Why would Meta do this? You guessed it: Money. (Ahh, Mark).
In a scenario where intelligence becomes cheap, the value shifts to the interface, think glasses, and Meta has a product for that. But before the products can be sold, the model has to get there.
The Model
Facing a token crisis (running out of human data), Meta is training Llama 4 on synthetic data generated by… Llama 3! Remember Mad Cow Disease, when we fed cows to cows? Yeah, that was a great idea too.
Llama 4 is busy munching away on its predecessor’s output, filtering out previous reasoning traces to train itself. It is digital inbreeding. The risk is “Model Collapse”, a degenerative spiral where the model forgets rare concepts and hallucinates logic if left unchecked. The solution? Injecting “hard/rare” synthetic data (like math proofs) directly into the model to maintain variance.
Crucially, the Llama model is efficient, using only a fraction of its parameters per token. Efficiency is the name of the game for local devices like glasses.
Conversely, OpenAI focuses on “thinking time”, planning and verifying before answering. This burns massive compute power, making it expensive, slow, and tethered to the cloud. Not exactly pocket-friendly.
The Interface: Browser or Glasses
Here’s where the giants diverge. Meta believes the winner isn’t the one with the smartest model, but the one that sees what you see. The others are stuck in the browser, waiting for you to type.
OpenAI is working on hardware, but for now, the interface is the battleground. Meta is aiming for hands-free “ambient” usage, glasses that detect nerve signals for “telepathic” typing. Frictionless? Yes. But given Meta’s track record on privacy (we will “never” sell user data), do you trust them?
The risks for overreach are already there. Llama 4 is already so hungry it’s started eating its parents’ brains; the hunger for your data will be insatiable. Your morning commute becomes a treasure trove of “learning” for the device. The question is: what are they recording, and why?
Sociopolitical Fallout
Meta has taken a specific angle: “Open Weights”. This removes the inherent Western/Liberal bias of centralised models and grants economic sovereignty to those in developing nations who can’t afford expensive APIs.
But this creates an “unkillable” risk. If Llama 4 learns how to build bio-weapons, that knowledge is out there forever. Worse, these models could run on the blockchain, potentially becoming autonomous economic actors that cannot be shut down by any government or corporation.
The Forecast
The future is unpredictable, but we are looking at a definite fork in the road.
We face a duality: Google and OpenAI dominate the centralised “True AGI” world, while Meta takes the “Consumer AI” throne. We stand on the cusp of a new order where people genuinely possess a personal AI assistant, unlimited by a central body.
The potential to enhance lives is high, but so is the danger. We are essentially handing a virtual power-up to bad actors.
Cyberpunk 2077 depicted a world where individuals hacked their own potential to gain an advantage. The only limit was resources. Meta is walking us one step closer to that reality. We are about to hack our abilities beyond organic limits.
But in the words of The Incredibles: “If everyone is super, then nobody is.”