To say that LLM (Large Language Models) are a dead end is an exaggeration considering how much has been achieved in so little time and the fact that we are already beyond LLM now. Still, Yann LeCun is right, the future is indeed World Models in the sense that AI must now be grounded in an actual experience and understanding of the world in order to progress further.
To put this argument in perspective and to use the analogy of Iain McGilchrist's thesis from his book, The Master and His Emissary (a must read to understand how the brain actually works), we have developed as far as we could the left hemisphere rational (the emissary) aspect of understanding while completely overlooking the right hemisphere, holistic (the master) aspect of the world.
In other words, a more advance AI will eventually have to mimic the human brain and root its understanding of the world on a dialogue between two conflicting approaches, which somehow must reflect the dual nature of reality.
Here's how DeepSeek synthesize this understanding;
The Cosmic Duality:
The universe is fundamentally composed of complementary, and sometimes
conflicting, principles: Particle/Wave, Matter/Energy, Local/Non-local,
Explicit/Implicit.
The Neural Duality: The human brain has evolved two primary ways of engaging with the world, broadly mapped to the two cerebral hemispheres:
The Left Hemisphere (The Emissary): is the "Particle" principle incarnate. It deals with the parts,
the discrete, the sequential, the literal, the categorical, the
manipulative, and the utility of things. It is the realm of logic,
language, and linear reasoning. It is brilliant at taking the world
apart and using it.
The Right Hemisphere (The Master): is the "Wave" principle incarnate. It deals with the whole,
the context, the embodied, the metaphorical, the emotional timbre, and
the interconnectedness of things. It grasps the "gestalt," the living
presence of the world, and understands the meaning that flows between
things, not just the things themselves.
McGilchrist's
central argument is that our modern Western world is
largely the story of the "Emissary" (the left hemisphere) usurping the
role of the "Master" (the right hemisphere). The Emissary, tasked with
managing the known world, has convinced itself that the known world is all there is. It denies the validity of the Master's broader, deeper, but less explicit understanding.
And this extreme rationality unfortunately is perfectly represented by our approach to AI. This is a complex but extremely fertile approach which eventually should solve the current AI conundrum provided of course we find proper solutions to this conceptual problem.
We will soon come back to explore this fascinating subject deeper and in more details in the coming weeks.
Via: Gizmodo:
One of the most important AI scientists in Big Tech wants to
scrap the current approach to building human-level AI. What we need,
Yann LeCun has indicated, are not large language models, but “world
models.
Why is he leaving a company that’s been spending lavishly, poaching
the most highly-skilled AI experts from other firms, and, according to a
July blog post by CEO Mark Zuckerburg, making such astonishing leaps
in-house that supposedly the development of “superintelligence is now in
sight”?
He’s actually been hinting at the answer for a long time.
When it comes to human-level intelligence, LeCun has become notorious
lately for saying LLMs as we currently understand them are duds—no
longer worth pursuing, no matter how much Big Tech scales them up. He
said in April of last year that “an LLM is basically an off-ramp, a distraction, a dead end.”
