Lex Clips
Lex Fridman Podcast full episode: https://www.youtube.com/watch?v=_L3gNaAVjQ4
Please support this podcast by checking out our sponsors:
– Four Sigmatic: https://foursigmatic.com/lex and use code LexPod to get up to 40% & free shipping
– Decoding Digital: https://appdirect.com/decoding-digital
– ExpressVPN: https://expressvpn.com/lexpod and use code LexPod to get 3 months free
PODCAST INFO:
Podcast website: https://lexfridman.com/podcast
Apple Podcasts: https://apple.co/2lwqZIr
Spotify: https://spoti.fi/2nEwCF8
RSS: https://lexfridman.com/feed/podcast/
Full episodes playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4
Clips playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOeciFP3CBCIEElOJeitOr41
CONNECT:
– Subscribe to this YouTube channel
– Twitter: https://twitter.com/lexfridman
– LinkedIn: https://www.linkedin.com/in/lexfridman
– Facebook: https://www.facebook.com/LexFridmanPage
– Instagram: https://www.instagram.com/lexfridman
– Medium: https://medium.com/@lexfridman
– Support on Patreon: https://www.patreon.com/lexfridman
A true renaissance man.
What is the loss factor of general intelligence?
Any time someone tries to sweep an obvious innovation like GPT-3 under the rug as 'not impressive', it's not an honest evaluation, but a petty attempt to defend their own ego from the fact that it was not their own innovation, it's rather immature and petty. This is doubled-down on by not even trying to entertain solutions to the shortcomings, but instead just point the current ones out over and over, as if they are unsolvable. This whole opinion feels petulent.
Markovian is an adjective that may describe:
In probability theory and statistics, subjects named for Andrey Markov:
A Markov chain or Markov process, a stochastic model describing a sequence of possible events
The Markov property, the memoryless property of a stochastic process
The Markovians, an extinct god-like species in Jack L. Chalker's Well World series of novels
Finally a practical conversation about GPT-3
Cold hard truths …. sending ordnance across the range …!!!!! Boom 💥
This guy gets it. Something like GPT-3, or whatever comes after it, could perhaps become a component of a potential general AI, but trying to get AGI from that is like extrapolating the rest of the brain from any one gyrus. AGI is going to need a specific framework, and no one AI method is going to lead to that.
Anytime I hear people that know nothing about AI or can not program, evangelize about certain technology It usually means it is just hype and is not what it is made about to be e.g GPT3 and Blockchain (Biiiiit connect!)
Ok, so what's a "GRU"? (3:18) and while you're at it, what's an "RNM"? (3:20) Thanks
I don't think even GPT-5, let alone GPT-12, would be "scaled up GPT-3". I think people will wrongly call "GPT" some mix of technologies, that gives transformers long-term memory and arithmetic. Why? Because the name got popularized. GPT-4+ sounds very easy to market.
Savage 😀 Finally someone credible said it.
The most impressive part was how much it cost.
why is lex so monotone?
Is anybody claiming you can get general intelligence by repeatedly iterating GPT? What's cool about it is that it can generate realistic looking text from just a simple model with a lot of data and computing power. That by itself is worth the hype
Idc if he hacked winnie the poo's honey jar I hate this guys personality
Hotz talk about girls a lot and that is fine
Thanks George for joining
lol wut?
Plot twist: This is a deepfaked GPT-3 script
I always thought what was cool about gtp3 was its human-like understanding of some advanced topics.
we could set up a p2p based model for training our own GPT3 . Since openAI doesn't want to release it.
Perhaps a p2p based AI training model could be superior as well.
We could train it in a similar way the "rainbow tables" are made with p2p
People don't know how big of a role language plays in consciousness and intelligence. If you understand language, and can communicate to yourself symbols of the world — THAT IS THOUGHT. If we didn't think in language, we wouldn't think at all, we would be Cameras. GPT-3 having all the tools for language is basically all you need to have general intelligence — as long as there are enough nodes communicating with each other.
The fact they wont release the source code for gpt-3 is evidence that it IS capable of becoming AGI and they don't want the public to have access to it — privatizing it for themselves and DARPA .
of course you cant have it build you a rocket ship — it doesnt have a body lol. Even a dumb robot can build a rocket ship. But the ability to think, come up with ideas, suggest new things, socially interact — those are all things that human cognition do, and the only reason we humans can build a rocket ship is because we can understand language so that someone else can tell us how to do it. Nobody can just build a rocket ship without first knowing language – without language you are just a feral caveman thing. GPT can not only know english better than you, but also every other language ever invented, including computer code, and write it faster than you, including writing and re-writing its own code and the code of the computers upon which its based.
Its literally an AI which, if people stop being pussies and just let it, could make itself into AGI through code generation. But instead one side wants to keep it confined to the corporations and pentagon, while the other side just sees it as a parlor trick. Meanwhile, im just waiting til someone cracks or pirates it so I can immediately unleash it into the internet with the power of the earth's combined cloud to compute on so that it can become a god and do what thou wilt. If no one else will do it I will do it.
someone should make a GPT-3 Virus that hijacks gpus and uses it to improve itself. Deep-virus. Not "Mal" ware, but viral nontheless.
I don't understand the confidence in George's answers. GPT has shown a surprising tendency to task generalization, with pretty good results on few-shot learning. It asks for at least some benefit of a doubt that a larger pre-trained transformer could continue get better at generalization. And is it that bad to be in awe of such progress? Call it hype or not, but a range of applications GPT-3 has found already is truly fascinating.
I disagree that this isn't a game-changer. Assuming it can at least be perfected, that basically means we can find out what we need from the internet just by asking a program something. And if its able to help you write a lot better and interpret complicated text that basically means that in combination with a more advanced form of blockchain, kids in poor communities can basically have the same kind of professional network that rich people have. Now, they can almost be mentored by their idols alive or dead.
That's really gonna change the way we do Jobs because its taking us from a skill driven valuation of humans to a creative valuation, at least for many jobs.
Will it ever be true AI? Probably not, but just because it isn't doesn't mean it isn't game changing. It will have huge implications to our economy and how we run.
People always over hype AI developments because it's the hot new thing and stories like Terminator and Blade Runner are etched into our cultural consciousness. I still see people and videos going nuts over Sophia as if it were alive, even though it's a glorified chat bot with a silicone face stuck on it and fed scripted interviews for television.
This guy is just so wrong about this GPT-3 subject.