If This Is What It's Like at Four…

When I read the remarks, they reminded me of a line from Copernicus: “I cannot make sense of the heavens while believing the stars revolve around the earth. So I must try the opposite.” ☄️

In an earlier article (curious readers can find it in the blog) I wrote about AI and consciousness coming together. Beyond my own views, I'd shared a video of an interview with Geoffrey Hinton; while trying to draw attention to the cliff we're approaching like a “truck with no brakes,” I quietly built the piece on Hinton's words. 😃 By the way, in my view he is still by far the “Godfather of AI.” 🕶️ I can't move on without giving him his due.

Recently, another authority — Yann LeCun (Head of AI, Meta) — made a statement emphasizing the obstacles on the way to that future, and criticizing the approach many have adopted for reaching artificial general and super intelligence (AGI and ASI).

According to LeCun, today's biggest AI models have been fed, at most, data equivalent to the visual cortex of a four-year-old child. 👶 In short, he says:

We won't reach real intelligence through large language models scaled with ever more data and compute. There's no untouched data of that kind left anyway! LLMs have hit a plateau, or are about to. For these models to do more than produce text or images, they need common sense and persistent memory, and they need to understand the physical world. Yann LeCun · Head of AI, Meta

By understanding the physical world, he means not just linguistic context but grasping interaction with objects and physical logic too — having intuition-based skills that can rapidly analyze the kinds of problems they have no training data for. The gist of the gist: “That's as far as the LLM hype goes; don't expect more — it's time to turn the wheel.” In other words, the approach is changing.

In my view LeCun has put his finger on exactly the right spots, and he doesn't contradict Hinton's warnings. He's simply arguing that the approach some have adopted to get there isn't the right one — and I think he's very right. 👍

To think this idea — supported not only by LeCun but by authorities like Wang, Gomez and Kalinin — has only just emerged, or isn't already being tried by many of the leading AI companies, would be “naïve” to put it mildly.

From this angle, if we're only now hearing about what's already being done, then here comes today's crazy question: “Could these statements be a way of slowly preparing people for truths that are hard to swallow?” 🤔

The “conspiracy theory” crowd was already ringing in my ears as I wrote this. But even if the ringing deafens me, rather than burying my head in the sand, I'm on the side of those shouting that international ethical-oversight mechanisms must be put in place effectively and fast.

Which side are you on?