Yann LeCun argues AGI is nonsense, pointing to failure of robotics and autonomous systems to match animal intelligence

It's not merely physical agents, it's anything that deals with something else than sequences of discrete symbols.

Any data modality that is high-dimensional, continuous, and possibly noisy is completely out of reach of current generative models.

That includes pretty much all real-world signals (aside from human language, computer languages and mathematics).

Furthermore, you can't have reliable agents unless they have the ability to predict the consequences of their actions and plan accordingly. LLMs simply don't.

If @ylecun can really make a 10yr old a first-time listener, this will dwarf his contributions to machine learning. Sign me up for the alpha!

@caleb_shepherd @andrewgwils 1. Today is a holiday 2. I suggest you look up how long it takes to go from a research concept to a technology breakthrough and a deployed product, and learn what the process involves.

@das_rdsm @andrewgwils Yes, humans are missing the G. Humans are quite specialized. But they are also adaptable and can learn new skills very quickly. Machines are doing many things better than humans. That's why they are useful and why we build them.

@ylecun @andrewgwils What is weird is that you have to spell this out when every AI bro can see it.

Their reality denial is handled by redefining what AGI means. It is now about solving artificial and contaminated LLM benchmarks.

@CrunchyAI @andrewgwils Thank you, you are making my point

@ylecun @andrewgwils I partially agree with Yann here. But I do think we are on the precipice or near car AGI

@ylecun @andrewgwils Well, we also dont have humans that can do everything those machines can do… are humans also missing the G? (Not trying to be smart here , really got me curious)

@ylecun @andrewgwils A teenager doesn’t learn to drive a car in a few hours. They learn to drive a car with 15 years of in world learning and a few hours of practice on top of that.

@MatthewWnuk @andrewgwils That's exactly my point

@ylecun @andrewgwils Intellegence is likely nonsense too, then. What is intellegence exactly? Until we agree to a common definition, how can we suggest something is solved?

@caleb_shepherd @andrewgwils Also, I don't complain. I try to rally people to working on the Right Things.

Based on Tesla FSD v14, we are very close to Level 5 self-driving cars. 1-5 years.

Whether or not it learned it in a few hours is irrelevant to the user.

But your broader point about the G is correct.

@ylecun @andrewgwils The letter G looks like an arrow being diverted too

@ylecun @andrewgwils "Learn to drive in a few hours like a teenager" is a bold oversimplification for someone who has spent decades preaching unsupervised methods of pretraining

@ylecun Yes, physical agents are not generally competent yet, and certainly new paradigms are needed for significant further advances in AI. But autonomous vehicles are now generally safer than human drivers under typical driving conditions.

@ChrissGPT @ylecun @andrewgwils Stop saying: "something AGI". Car AGI, coding AGI, Math AGI. AGI by definition has to be general otherwise it's just a "narrow" AI that is good at one specific task. That means it's not able to transfer skills to multiple domains and lack holistic understanding of its environment

@ylecun @andrewgwils If the G is nonsense, what’s the right term for what a teenager has that our best systems don’t?

@TonyZador @andrewgwils The Moravec paradox is 38 years old and we still need to remind every new generation of non-physical AI researchers about it.

@BrianCHolt @ylecun @andrewgwils That's my point really. Much of what I see people arguing online is just due to their differences in what something actually means. It gets a little ironic when discussing LLMs.