Meta CEO Mark Zuckerberg is making some very grandiose statements amid his ludicrously expensive quest to gain a lead in the AI industry.
It’s easy to get caught up in the human drama of his talent hunt. But in the very first lines of a new public letter about his so-called Superintelligence Lab, the tech founder made a claim that’s either a staight-up fib, a very generous reading of a situation, or a clue that he’s closer to enacting profound change in the world than almost anyone realizes.
Specifically, the exec boasted that in recent months, Meta has “begun to see glimpses of our AI systems improving themselves” — a bold claim that Zuckerberg suggests is a precursor to “developing superintelligence.”
Because the Meta CEO declined to go into specifics with that vague claim, it’s hard to say exactly how revolutionary that progress is.
Taken at face value, “self-improving AI” has, to some extent, already been observed, though only so far in limited domains.
Known as recursive self-improvement, this formerly-theoretical process involves AI systems that develop the ability to improve themselves without human intervention. Back in 2023, researchers from Nvidia and a consortium of American universities did exactly that when they built Voyager, a Minecraft bot that learned to rewrite its own code continuously using OpenAI’s GPT-4 large language model (LLM).
More recently, Google DeepMind researchers launched AlphaEvolve, another self-improving AI system that was similarly helmed as a step towards artificial general intelligence (AGI) or even superintelligence.
In Zuckerberg’s letter, which was posted in tandem with an eerie video of the exec droning on about how “personal superintelligence” will one day be your best friend, details about Meta’s purported “glimpses” at self-improving AI are quite literally nonexistent. Mentioned as an aside and then abandoned in favor of uninspired copy that very much sounds like it was written by a chatbot, the concept of an AI that can improve itself without humans isn’t even defined, much less explained.
If an AI really could improve itself in meaningful ways, it would be an immense accomplishment — and one that may well precede or even mark the beginning of what some experts call the “technological singularity,” a hypothetical future moment at which technology starts to self-improve at a rate so rapid that it outstrips human understanding.
But left as vague as it was, it seems like Zuckerberg likely wants readers to draw their own conclusions that his efforts are making immense strides, without being accused of actually saying as much. Indeed, in an investor call that occurred the same day the CEO dropped his video and accompanying letter, he didn’t even mention those self-improvement “glimpses” and instead said Meta is in the process of developing such models.
We’ve reached out to Meta to ask for more clarity about those alleged glimpses, but if they really did happen, we kind of doubt the company is gonna share too many details about them with us.
More on Meta: AI Researcher Declines $1 Billion Offer From Mark Zuckerberg
The post Zuckerberg Says Meta Is Now Seeing Signs of Advanced AI Improving Itself appeared first on Futurism.
