Seems like old video. Please add the video recording date to your video's description. Thanks.

He is 100% correct. The models are not smarter than everyone. But they are smarter than huge amount of people.

@antonystringfellow5152

9 күн бұрын

Most people are a lot smarter than the best models in some areas. That is also 100% correct. "smart" does not give a good description of a person's or an AI's abilities unless a particular area is specified. We don't know what it means to be generally smart or intelligent. We only understand it in the context of comparing one human to another, which is very limited. LLM's have made this point very clear as they surprise us both by how "smart" they are and how "dumb" they are. Some things that we thought of as simple are still beyond them while things we considered difficult, they can excel at. What this demonstrates is just how bad we are at determining what's easy and what's difficult (what's simple and what's complex). This is because we've based these beliefs on what's easy and difficult for us and we ourselves are a form of narrow intelligence - we have been shaped into what we are by natural selection. Optimization of efficiency was essential for survival and our brains are optimized for survival and reproduction in the African savannah. Our brains are not good with math (a chip from a cheap 1980s digital watch can perform better) just as they're not good at understanding quantum physics and the nature of spacetime. It doesn't necessarily mean these things are very complex, our brains are just not designed to model such things - they're not essential abilities to have when you're a hunter gatherer, living in a world where the most complex piece of tech is a stone-headed spear. I think the day is drawing close though, when calling them "smart" will be a good description - when they can perform all mental tasks better than any human.

Great talk - )

Saying human brains use much less data than current SOTA neural networks isn't really a fair comparison. We randomly initialise the weights of our neural networks, whereas our brains come initialised from billions of years of evolution. Many animals are born able to walk, run, or fly straight away, and have many of their behaviours built in. Same goes for humans but it's less obvious I think as we take much longer to develop.

@goldnutter412

12 күн бұрын

Saying the brain is everything is just lacking any clue about the universe honestly. We'll get there, and it will be rapid progress once we start treating the brain as a construct, a programmable cache. Our mind is something else, an information system.. we're separated but.. only by these mathematically/computationally created constraints. The concept of god is just one of early stories.. now we have the information age and clarity is right there hiding in plain sight. We made this place.. we're far more than we think. Iceberg analogy works for most.. the "habits" and beliefs are just things we do to manage the raw data, solo, no cheating experience. If you don't earn personal growth.. you don't get anywhere.

@goldnutter412

12 күн бұрын

Oh, I forgot to mention the placebo effect makes sense, and why dude missing most of his brain has a normal life.. etc etc.. it all falls out so easily once you get past the stumbling blocks. A great bridge is being built though, please check this video and share if you agree/like or are interested in discussions ! Is Consciousness Fundamental? Conference Talk, September 2023 Annaka Harris

I'd like to see one of these LLMs take the place of a math professor. One problem these models have is that they never just think and dream, with no question pending.

@jeffsteyn7174

2 ай бұрын

Don't worry. They will and sooner than what you think. Chatgpt using tools can already teach math.

@RalphDratman

2 ай бұрын

@@jeffsteyn7174 I agree it might be able to teach math even now, though not very well yet. In fact I have asked it some math questions and it was helpful, I think. But suppose it had to solve a math problem it had never seen before?

@MrFujinko

Ай бұрын

@@RalphDratman Probably will fail. It depends on data it has been trained on, so if the data about "self reflect about this particular thing" was never seen, then it can't do it. Solution is probably mixing a LLM with another mechanism for learning and self reflection.

@goldnutter412

12 күн бұрын

@@MrFujinko LLM's are just an interface module. You're all being used as free testers and extra input data for their benefit, lol~ We need to make modularized, and DECENTRALIZED systems. There was actually just a merger of three projects which is now the ASI Alliance. The amount of industry experience and insight across the three teams is beyond comprehension. But any system will be better if you use the language model as only the interface to trigger other subsystems. This is the only way to start at the top and go down. The real way to win is to actually try and remake how we did what we did, to get this far. But it has to be raw pattern matching and completely on it's own. But then when it creates a language, we won't know WTF it means. Herein lies the reason I say Google is winning. People disagree, but people are morons. I'm a moron so many times in the past.. look at how arrogant a species we are, keep realizing we were WRONG Cosmology broken af because physics still trying to avoid the obvious truth about matter/energy. *theoretical physics mumbo jumbo Applied physics ? lets go.. materials science of this paradigm (2020-2040) will shock everyone, it already is and this is NOTHING. The compute went up 1million times the last 10 years - nVidia CEO..

There was a book superintelligence which compares einstein and village idiot

I get it, click bait title. Saying everyone was wrong is a much stronger title than saying a lot of people were wrong. If you someone who simply say whatever A.I can do now, will be able to do it faster with more accuracy year after year …and you just leave it at that- then you’ll still be right about A.I.

In this matter of AI not being able to make these connections over the many parameters it has, I think it's because it is passive, as he said, it is very good at making these connections to make sonnets like Shakespeare, but only when asked, it doesn't You have the desire to go out looking for connections in all this, when you start doing this the game will change completely. I also have a completely different view of intelligence than ordinary people, I believe that everything is standard and there are several ways to create smart devices

Just subscribed. The Demis video was the first I saw. You’re a great interviewer

I wanna buy Claude Pro from EU when will I be able to do it ????? Please.

@berkk1993

2 ай бұрын

get a vpn

@ChuckNorris-lf6vo

2 ай бұрын

Payment was declined because address of card is also locked to Non EU countries.@@berkk1993

Loving Claude 3… love Dario Amodei, BRILLIANT

cant access claude 3 opus, but sonnet will gladly and authoritatively misapply mathematics. i expected more.

Can you please start a second channel for these clips, or alternatively make one for the full interviews only, so we can avoid having our feeds spammed with this. You are risking an unsubscribe from me.

My prediction is that when humans try to play the role of God the ending won't be very good . And if these robot's have the ability to learn just like a human being ?? wouldnt that be a form of slavery ??? just a thought .

dont post old videos please

@gball8466

2 ай бұрын

Clips of longer videos is how KZread works

spoiler alert, it's click bait

Kurzweil.

Videos where someone just talks and talks without being asked anything important, and reigned in a bit so they don't go all over the place, frustrating and irritating. I like Dario Amodei but I think he could go on forever and this seems to need a focus.