Metas AI Team Just Revealed Their Secret To AGI

00:00:00 Yann's talk
00:00:23 Human-level AI
00:01:11 AI requirements
00:01:54 Current limitations
00:02:25 Proposed architecture
00:03:47 Moravec paradox
00:05:27 Data comparison
00:07:26 Visual data
00:09:02 Objective-driven AI
00:11:33 JEPA architecture
00:12:57 Video prediction
00:15:31 Future implications
00:17:49 AGI timeline
00:19:25 Expert disagreement

Links From Todays Video:

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[SapienSpace]

JEPA is strikingly similar to the application of Fuzzy Logic, a method that merges statistical math with language (symbolic representation).

[thelastshaneperson]

What the hell happened to Zuck? How is he becoming the "liberator" of AI?

[MrBillythefisherman]

A 10 year old has not learnt to clear a table and fill up a dishwasher in one shot: they are 10 years old and have had to learn an array of skills such as standing, walking, speech, understanding others speech, fine motor control: most of which is done before 5. Then they spend the next 5 years learning about all the edge cases. Sure when they get to ten they have generalised and that's fine but give our latest LLMs continuous training on multimodal data and they could well be like a massively superior Einstein. We're not missing a lot if we are missing stuff.

[Wizardess]

AI that is only text driven, recorded video driven, and/or aurally driven can never become intelligent. It may simulate but not be intelligent. It must be equipped with the better part of the fill suite of animal and human senses, including "I have a booboo on my finger and it is bleeding and hurts a lot so I am crying." It must learn consequences in a way that means something to it. That would be how it learns humans are good to have around, we kiss that finger and make it better. Works for child. Should work for AGI we want to be like a social human rather than a "Spock" like me. It must act, feel, experience, and react to a dynamic world that changes gradually enough action and consequence can be experienced in real time. "I am tipped off balance, . I must fix that. This must be done at a level below my thoughts, too."

Yann is spot on as far as he goes. He needs to think forward to the next steps. He should do this while raising a real human and imagining it knows everything he knows except how its actions afecet itself and the world around it. "The burners on the stove are often very hot and hurt like hell." That will mean something with the right AGI training. Without it you have a fancy huge library and a useful librarian. The equipment the AGI has to work with matters as much as its library and librarian.

{^_^}

[picksalot]

Trying to make AI learn and work like humans is like saying we need to put legs on a car so we can travel around. 🤪

[woolfel]

my guess is something like V-JEPA that combines statistical with symbolic is needed. It's pretty clear that transformers are powerful, but they have fundamental limits. All of the attempts to extend context window run against physical limits and they burn asssston of power to train on gigantic clusters. I also don't trust Sam Altman.

[AustinGlamourPhoto]

Yann is definitely the most based in reality of the main AI researchers. He doesn't have it figured out but at least he looking in the right direction.
As time goes on the limitation of LLM will become more and more obvious.

Human intelligence developed to solve real world problems in the objective environment and LLM have no objective environment at all.
So they can solve SOME hypothetical text based problems, but they can't think in terms of the real world or solve problems in the real world because of the LLM there is no real world.
How smart would you be if you still had all of your knowledge but your lost all your perceptions of the world. You could still "think" inside your head but there wouldn't be any real world outside your head. That's basically where LLM are at and are actually stuck at.

[yoyo-jcqg]

the 17yr old that learns to drive in 20 hours has been training for 17 years in the real world

[Ad]

Nice video, and I feel that Jann's general observations aligns with mine on how far away we are from AGI. It is not feasible with LLM's as of today and need a radically different architecture to create something resembling AGI. When some people, like Altman, Jensen Huang and others hype up and saying AGI is near, it can be two reasons: one fundraising/stock price and the second is that there is no set definition of what AGI is. Jann's introduction is good, as I believe it does lay out the complexities of what a true AGI would need to be capable of, which is just not feasible with current technologies.

[Justin_Arut]

15:27 Indeed, young Lincoln is one of the brightest AI researchers. No question about it.

[samahirrao]

We need human level democratization of computing power, without it we do not see any AGI.

[styx]

Thanks again. Important to consider a new generation of people who super educated by nanny robots from birth And the type and manner of socialisation . The learning and maintaining of multiple languages could be the basis for this mind expanded generation. When we super educate people they won't suffer the malaise of bordem or be misled by junk media. In my opinion discovering what it means to be human is the reason to be human.. This becomes obvious when we live into our 60's ; the different phases of experience are then comprehended as an overview of periods of different passions and illusions. So certain at the time but in hindsight 'just' human developmental stages .

[txzgmzp]

Government (military) will obtain AGI/ASI before all.

[maxss]

I think some of these predications take into account what the companies are working on behind closed doors. Like openAI was just setting on The chain of thought model waiting for the hype to die down around the previous GPT models etc... before releasing it.
I think there are different types of architecture they are setting on in regards to models structures.

[fromduskuntodawn]

I really like these videos with a blend of experts and then your commentary.

[eviv]

He says as if all AI will be under his control lol

[georgewashington]

We're missing the embodiment of AI.

[qwertyzxaszc]

Yann is the only adult in the room where everyone else in AI come off as children.

[TheAIExplorer-oj]

16:14 Wow, I have to admit, your moves are on point here! Respect. 🙌

[matthew]

Why people keep thinking " a few thousand days = a few years" ? How ? Did someone skip kindergarten school ?