Yann LeCun: Self-Supervised Learning Explained | Lex Fridman Podcast Clips

[@oldnepalihippie]

aw man where is the rest?!? such a clear explanation so far of the hard problem. i want more!

[@Jfantab]

in the description, you can find the link to the full podcast episode

[@hanyanglee9018]

That is because images, videos contain way more information than text. If you have trained to draw, you'll definitely know how big the gap is between description and the final image you are gonna draw.

[@eelcohoogendoorn8044]

Not clear that this should make it easier rather than harder though. Frankly I am a bit surprised that video hasnt made more progress on this front i the past few years; a shame the clip cuts there.. now im going to have to watch the whole episode to see what more yann had to say on the topic.

[@the_disruptor_deagan_ferland]

Lex is so smart he is half asleep explaining everything the professional is saying but in a more simple and smother sense.

[@aromax504]

If we cannot create an AGI then can we use the human brain circuitry itself to generate a NN with human comparable intelligence?

[@rb8049]

We don’t have true L5 driving is because we are stuck on old concepts of how the neuron works and how the brain works. Once we stop thinking this way and open our minds, it will come overnight.

[@nicktheodorou3474]

I don’t see consciousness as mysterious. We are just a very complicated quantum machine and consciousness was just the next logical step in evolution. Being aware of the past and future. I think this is the self learning mechanism which is the key.

[@Saad-xo6yc]

lol

[@linkin543210]

"quantum machine"...are you a marketer ? 😂

[@nicktheodorou3474]

@@linkin543210 Lol! The brain and body is composed of atoms, and atoms follow the laws of quantum physics 😉. But probably not the best way to refer to something like cognition.

[@linkin543210]

@@nicktheodorou3474 I'm sorry I was a bit rude, and I totally agree that conciousness was the next level in evolution to survive.

[@cashh2906]

what about Pulsechain

[@hanyanglee9018]

We have self driving horses but we don't have self driving cars.🤣

[@thomasgilson6206]

Horses have 5 billion years of self-supervised training. (Evolution).

[@billbrett365]

They need to eject the lithium electro-ceramic bio-filter. should eject the subatomic vacuum node.

[@cathodeactive5932]

DOJO will be a paradigm shift. Maye the force

[@eerievibes6854]

I'm supposed to believe this guy is a genius? He spits out like 4 words a minute.

[@thelonespeaker]

Both speak at a reasonable pace YLC is French btw

[@JohnSmith-bb1sv]

I don't mean Lex any harm but he is not a genius not by a long shot. He's slow, he asks dumb questions, and he is obtuse. His interview with Vladimir Vapnik was what did it for me. The interviewer needs to move at the same intellectual pace or faster than the audience. Machine Learning Street Talk is a much better podcast and they interviewed Prof LeCun twice.

[@septemberamyx]

We also have several quantum dimensions that our brain uses. Do you dimensionally train AI?

[@saltyfish157]

This is the best.

[@christopherprobst-ranly6357]

Humans learn while sleeping. Maybe we just run a lot of examples in our head 🤔

[@lukassteindl1914]

think so too but you dont let it run 1 million times. when you dream, it feels more like a single realtime replay per night, so it must be much more efficient than reinforcement learning type episode replay. and sometimes the mental video shows strange things that are unrealistic. might be some calibration / regularization going on while one sleeps.

[@christopherprobst-ranly6357]

@@lukassteindl1914 Well, .. remember Inception and the in-sleep time ;-)

[@shilashm5691]

Man with the specs, talking about meta learning, meta learning is different from self supervised learning

[@campervanman5340]

All you need to do us create a system with desire, as soon as an AI actually WANTS to learn something it will need to be programmed and therefore can never be as smart as us.

[@artpinguin]

Why don't you make longer vid???too short not wordy of my time...

[@samosa9488]

Watch the full interview on his other channel. Lex fridman

[@artpinguin]

@@samosa9488 Okey thnx friend

[@isak01234]

Why does lex sound so sleepy and boring? Can't he make an effort and learn to articulate and project some energy and enthusiasm?

[@michelcote]

Building world model is done trough an interacting physical body. To duplicate our intelligence I beleive that for training we will need to provide a phyical body with same sensors. A humanoïde robot

[@eerievibes6854]

Everything about this channel is a JRE rip-off.