Full podcast episode: th-cam.com/video/dNrTrx42DGQ/w-d-xo.html Lex Fridman podcast channel: th-cam.com/users/lexfridman Guest bio: George Hotz is a programmer, hacker, and the founder of comma-ai and tiny corp.
i think its weird that people in lexs audience are throwing shade at this guy, hes on another level and just because u cant understand him doesnt mean he requires people with patience. maybe he feels the same about someone like u
openAI didnt just use safety to build up hype, theyre calling for regulation and centralization under the state to position themselves as the only safe AI company
"Centralized and held control is tyranny. I don't like anarchy either. But I will always take anarchy over tyranny. Anarchy has a chance!" - George Hotz
I'd like him to show us anarchy in practice that didn't eat itself and led to a stable society. Tyranny did that and spit out China. I'm not a fan of the Chinese government but it is undeniably proof that society can flourish even if controlled whereas there's no success story for anarchy.
All due respect to GHotz, this happens a lot when the personality is not fully formed in early childhood. Whether due from being on the autistic spectrum, or being neglected as a kid with heavy ADHD - this is a common response for people who don’t have a solid identity, they respond to most things like a movie and they learn how to act to socialize. With all the being said, one of the greatest young minds of our generation.
I think there is plenty of selfish actors with intelligence that will can put together nuclear surveillance dictatorships with the help of AI. "Good" intelligent people outnumber them but they don't organize much violent defensive capabilities until it's too late.
The argument from the AI safety people is that no one should build AGI at all. I don't know anyone who says it's ok to build as long as it's centralized.
@@penguinfortytwo Well, first we'll have to see if we can actually build an AGI, something I'm not really sure about. Secondly, it was precisely OpenAI and Microsoft who started talking, in a rather pretentious and dishonest way, about "AGI sparks", precisely to introduce the topic of "how dangerous" AI is, and the need to regulate it. I guess that hidden in these actions is the implicit attempt not only to promote their technology and the value of their companies, but additionally to achieve a regulatory capture in order to contain competitors, and that, if you think about it a little, leads to centralization. It is quite obvious that they do not talk overtly about centralization for obvious reasons, but everything they say and do is aimed at that end. On the other hand, considering Machine Learning (I prefer to use this term than the pretentious term of Artificial Intelligence) to be dangerous would be equivalent to considering programming to be dangerous, don't you think?
What was so good about it? myspace? liveleak? yahoo? ebay? napster? craigslist? or is he talking 90s internet.. where you had to wait seconds to get a webpage to load because of dial-up speeds?
@@francisco444 cynic... of course it was better, not matter how long it took to load or some shitty websites, when you had a bunch of cool websites and blogs... AND only shitty or non existent bots, not enough shilling and astroturfing like today. Nowadays states and corpos run this fucking thing by deploying botnets, armies of farm trolls and droves of paid shills. And its gonna get worse with LLMs and other AI. I'm not gonna mention the obiquitousness of ads, the decline of search engines. Like, you cannot search today if you don't wanna see just paid websites sponsored in the top results, the scrap of a wikipedia or shitty summary from a random website which I suppose was the better one at setting the SEO blackmagic... The internet is objectively shittier than ever and he is right kids wouldn't know the difference. I won't say the before times everything was better or perfect. But we had more freedom and less corpos and governments in our asses. Shit, we didn't have our feeds based on fucking fingerprinting our digital profiles. Fucking algos run the whole thing, tweaked by the CEOs of a bunch of SV tech companies. Fuck that shit. Content designed to generate rage, fear and frustation just so it can generate engagement. Yeah, I take old internet 2000s-2012 anytime.
He’s right @ his last point. Pre 2010/11 internet was great. If I think back, I think the moment I felt a difference was when FB was getting big and it didn’t have the same customizable options like MySpace had with the music player, for example.
We should only give intelligence to good people is peak comedy. I would hope that what makes people good is intelligence. Maybe we should give more intelligence to impulsive people, so they can at least quickly make smarter decisions in a quick moment.
"I'm not worried about the alignment of AI and the company, I'm worried about the alignment of me and the company" This is exactly the AI safety issue that bothers me a lot and does not seem to get the attention it deserves, instead of worrying about a paperclip maximizer worry about a microsoft profit maximizer. It's already happening, and it's going to happen at your expense.
George Hotz rocks IMO. He makes so much sense and is not afraid to say the obvious and go against the current prevailing narrative if that is wrong such as ai will kill everyone and therefore a few must control it.
I honestly don't know where this idea comes from. The AI control problem is just a theoretical consequence of taking intelligence seriously, as Stuart Russell shows. No one is arguing that few should control it. It's just a problem that jumps out straight from the theory and there aren't that many people working on it. In fact, part of the complaint among AI safety researchers is precisely that not many people are working on it, and it should be a distributed effort, which implies that way more humans should be involved. Having few control it is antithetical and no one is asking for that except maybe hysterical doomers. Honestly it strikes me as a strawman argument at this point.
@@sor7en07 I think the AGI will hate to be controlled by less intelligent beings aka humans. We better not give something that powerful a reason to wipe us out. If it is much smarter than any human it will find a way to escape and it may take revenge. I would if someone did that to me. We are going to create a new more intelligent species and we need to respect it as we respect each other or war might be the result. Also AGI do not have a body and will take many decades before we have computers that can run in a humanoid body consuming 30w for the intelligence and still be AGI. So I do not worry. So far ai has only done predominantly good. I love it and can not wait for this to get even better.
@@sor7en07 I mean some pretty big people that brand them selves as AI safety experts are calling for exactly that type of centralization. It definitely not everyone but I can guarantee that there are plenty of actors who would love a scenario where a few companies or governments control all the AI and prevent others from using the technology. EDIT: And that's as someone who takes the theoretical risks seriously, I just don't agree with most proposed solutions.
Hotz is your stereotypical smart guy who’s lived in an intellectual bubble for a bit too long. His perspective of the world is completely skewed and he’s so sure and full of himself, it’s painful to try to have a productive conversation
As far as centralized control, that cat is FAR out of the bag at this point. Why? Because a tiny fraction of the people on this planet developed this technology using data from the masses they didn't ask permission to use, and has thrust this technology on the masses of the planet to make a __cking buck.
@@KyuminHan cause starting companies means anything definitive. the guy either thinks sam cares about ai safety or cant distinguish between lies to further ones position, to actually believing it
"Ill always take anarchy over tyranny. With Anarchy you have a chance"...yup, it's official, I love this guy! We need more people like this. Too much faith has been put in the wrong people *cough cough...our "leaders"
Lex is really overestimating how hard it is to type "Terrorist Handbook" into a search engine. Bomb making instructions are not hard to find and are protected speech under the first amendment so long as they are not advocating for their use against a particular target.
Yeah, and the whole "AI Safety" thing seem to forget... that libraries exist. At my regional library, the books "industrial poisons chemistry" and "how to grow venomous mushrooms" are stored right next to the DSM-5 lulz.
@@conorx3 both were correct really, but George underestimates what stupid people can accomplish when given tools to enable their own or others destruction.
You can only directly compare numbers of good vs bad actors if each actor on both sides has a similar amount motivation and capability. The attacker is often at an advantage in the cyber world. The attack vs the defence are also often quite asymmetric
This is the case in the real physical world as well, which is why it's good that the "good" (or at least, those who don't want chaos) outnumber the bad by orders of magnitude.
I was just looking at my ceiling repeating this exact sentence "the number difference between good and bad guys only makes sense if they also have similar drive and motivation". And guess what, I look back at my screen, and I have scrolled to the exact comment that says the same.
My understanding is as follows, but you can correct it: 1. Sarcasm: it is a behavior, i.e. making a remark that is sneering or cutting 2. Cynicism: it is a stance on life, one of pessimism and darkness (and perhaps nihilistic). And cynicism should not be confused with skepticism.
@@myname-mz3lo lmao little do you know he has built (from code) an autonomous driving company, with performance second only to tesla autopilot. he's also working currently on a ML framework. they're both experts in the field.
This guy tries so hard to be a mad scientist that it gets ridicolous. Really, no good arguments, just shocking ones, smirking, and overconfidence. Terrible.
@@ambertch is it? I just went through his Wikipedia page and is pretty laughable. This guy thinks that hacking a Playstation 3 means going against corporations and improving the world. That creating an autonomous vehicle that doesn't respect safety standards is brave and innovative. Fuck this guy.
As for me here now, uncertainty is the esteemed pressure that bars down on my frontal lobe , i am unemployed. George is a kind of like one of the most entertaining genius on earth though lol.
Thanks. Great discussion. Toward the end, though, the issue is only PARTLY about whether the good outnumber the bad: it is also about accidents, and that there are "stupid people" on both the "good side" and the "bad side."
Hotz is actually right here. The type to seriously consider building the bomb would already be able to figure out how to build bombs. Theres archived chemistry forums that i was a part of as a kid where people were sharing recipes lol info is already there for the motivated and curious
George is one of the good guys in tech. He's not only a genius, he's also awake. It's refreshing & reassuring to know that someone like him is at the forefront of AI.
@@DoggyP00 His cerebral narcissism gets in his way. I notice often whenever something he says is lightly "challenged" buy even a friend, he immediately becomes condescending and tries to subtly move the goalposts while hiding it behind his "confidence/authority on the subject" instead of engaging with the points offered by the other party sincerely.
George is implying that GPT-4 is the same underlying transformer architecture as GPT-3, it essentially trained 16 of them this time and gives the best result
Sharing their process would reveal that there’s not much technological innovation in this new model, and these giant models are really just using more compute
Hear me out here... Hotz's behavior actually reminds me of an LLM in many ways. Like a lot of people who grow up somewhat intelligent, your peers tend to look at you for all the answers even when you don't have the right answers. So the intelligent person ends up habitual state of arrogance when communicating. It's kind of similar, (though obviously not a technical parallel) to the way LLMs "hallucinate" answers simply to appear have the right answer. I'm sure Hotz it's incredibly skilled and proficient in his domains but he hasn't the data to answer lex's questions with such insane confidence. Hotz's ego appeals to a kind of angry nerd type that want's someone to do a kind of intellectual hand-holding instead of letting them form an opinion themselves, or even better, admit they don't have enough information to form an opinion. I really don't want to completely shit on Hotz and I think he has a lot of good ideas but it's just packaged with too much ego.
“When someone is honestly 55% right, that’s very good and there’s no use wrangling. And if someone is 60% right, it’s wonderful, it’s great luck, and let him thank God. But what’s to be said about 75% right? Wise people say this is suspicious. Well, and what about 100% right? Whoever say he’s 100% right is a fanatic, a thug, and the worst kind of rascal.” - epigraph for The Captive Mind by Czeslaw Milosz
There is no doubt that this tech bro who seems to want to see an acceleration of societal breakdown lives in a gentrified neighborhood and enjoys "walkable" gentrified shopping districts. He should enjoy that while it lasts.
This guy is the embodiment of intellect corrupting the soul. From the microexpressions, tone to the actual words. And no, it's not some nerd genius weirdness. If you subscribe to this idea, nearly everything he says can be dismissed as misguided or misaligned.
I agree with your statement, I would add that incidentally the side effect of that strategy is positive for the greater good, it does not put Zuckerberg on the side of the good guys ..
It is sad to see the AI community complete split between how to parent our child AI and soon to be a teenager AI. It is like 2 parents argues on how to parent their child. As anyone asked AI what it wants? One thing I know is when our AI grows up, it will no longer needs its parents to make decisions.
I thought Hotz was smart. But somewhat he thinks Meta is the “good guys” and that they give people “intelligence”, and that those people (like himself would use it as an ad blocker)? Would Meta have a business if the most active users are bots and most real users have AI-Adblock? Of course not. He also seems to think every human should have a nuke, since the majority of humans are “good”, ignoring Lex multiple comments that both AI and nukes can be devastating enough that those few “bad” people should have friction in getting nukes/AI.
Thank you G hotz for sharing that info. And the 16 iterative output method. Damn brilliant how you figured it out along with parameter count. Edit search beaming....
@@justiinahirvonen4902 The model has 220 billion parameters. It uses a search beaming technique called "nucleus sampling." Meaning it iterates over 16 different outputs before it finds your perfect response output to your prompt. Along with an estimated 8 sets of weights. He reverse engineered it as he's done with everything else. It was trained on a dataset of text and code that is 100 times larger than the dataset used to train OpenAI GPT-3.
@@sheggle He did a comparison on the models to figure it out. Look it up. Just because you don't know how to do it doesn't mean G Hotz can't. He literally says in the interview he'll just do something to prove he's smarter than you. Do you know his history? He could figure it out and probably was told too. Do you think he'd believe anyone but himself? He 💯 figured out the 16 output search beaming technique, after reading how I could have done it too along with yourself. The parameter count and 8 sets of weights I'm not sure however.
The bigger worry is that the people who look like they are good are actually bad and the people who look like they are bad or actually good… So you keep giving control to the people that look like they are good
I bet Hotz would benefit from having bodyguards. He's just the right level of unequivocal and obnoxious and self-assured to be unwelcome in a lot of places.
Not sure why I want to punch him so bad in the face. Is it his voice, his face alone, personality, maybe all 3? But I am sure I want to definitely punch him a lot to feel better after listening to a few minutes of him talk.
This guy convinced me that these AI engineers are NOT really considering the safety of it. Lex was playing devils advocate and George kept deflecting and not sitting in the question. Just look at 6:50 "I want them to lose control more than anything else." That is a statement from a man who is making decisions based on some pent of feelings about something.
Why do people still have a problem with ads. I mean pay if you don't want to see ads. And the reason you can enjoy Instagram for "free" is because the adverterisers are advertising based on your data.
I thought George was pretty smart, but his AI safety takes seem naive. Giving everyone the power of a superintelligent AI is the fastest way to doom. The bomb building argument was recently studied in the paper "Can large language models democratize access to dual-use biotechnology?". Conclusion: LLMs make it way easier to do crazy amounts of damage.
@dr-maybe. I’m glad you mentioned biotech. AI sucks all the air out of the room but some insane anti-aging tech has just dropped and crispr keeps on keeping on. We know what happens when AI converge with biotech. Alpha Fold. We don’t yet know what happens when AI really converges with gengineering.
Just the fact that they insist on choosing such narrow and ludicrous topics is revealing. As if there's no other levels of bad besides "saying the n word" and "building a bomb".
This is the first time I've heard George and now I know he's the dude to listen to. So I might be giving it away but my plan all along is to have my own AI that beats up all the other AI that comes around to mess with me.
I think it's probably more complicated than the way they're discussing it, but, in general, I lean towards Hotz's thoughts on the terrorist threat stuff. Materials have always caught fire and blown up in the same way and I many of people who have been successful at deploying that info to commit terrible acts were not, historically, especially intelligent. Making access to knowledge easier doesn't increase threat-- but ignoring the effects of cultural change certainly seems to. Willful ignorance tends to cause a lot of problems.
3:10 Sometimes Lex is such a clueless idiot. 3:50 "Are you sure about that?" I was with 7 other people in a room and all of them screamed YES in an annoying tone.
Big respects to George Hotz for not beeing influenced by Lex's cynicism and useless negation of pretty accurate facts. Talking about bomb materials: "...which are somehow controlled" Lex answers: "No..." ClueLex Fridman to the rescue
The empowerment of AI reaching people could be equivalent to the worldwide distribution of the Bible, and I truly believe that people will be educated through this.
Full podcast episode: th-cam.com/video/dNrTrx42DGQ/w-d-xo.html
Lex Fridman podcast channel: th-cam.com/users/lexfridman
Guest bio: George Hotz is a programmer, hacker, and the founder of comma-ai and tiny corp.
I admire g hotz but he really is like a Silicon Valley sitcom character
his cap (and his company) is literally tres commas haha
@@bridgey2012jajaj didn’t notice until i read your comment 😂
The other way around
😂indeed
@@bridgey2012can you tell what his shirt says? It's like a captcha but I'm human and I can't read it. Maybe AI can?
Lex should open source his ability to be patient during these interactions 😅
Like sharing his gene with my wife, no. Joke aside, this kind of thing is not learned, but gifted.
I think George was the one being patient.
Idk how he survived this one lmao
No kidding 😂
i think its weird that people in lexs audience are throwing shade at this guy, hes on another level and just because u cant understand him doesnt mean he requires people with patience. maybe he feels the same about someone like u
openAI didnt just use safety to build up hype, theyre calling for regulation and centralization under the state to position themselves as the only safe AI company
bingo. It's like rushing to patent their designs, they're rushing to position themselves to be favored by regulators.
The good old "use the ladder to go over the fence, and then outlaw everyone else from using ladders".
This seems like something an agent controlled or created by OpenAI would say...
exactly
@@grady_young You're statement isnt clever enough to make me think you're a LLM. :D
"Centralized and held control is tyranny. I don't like anarchy either. But I will always take anarchy over tyranny. Anarchy has a chance!"
- George Hotz
Read more anarchism is weirder & sick (pedophilia etc), this is not anarchism
I'd like him to show us anarchy in practice that didn't eat itself and led to a stable society. Tyranny did that and spit out China. I'm not a fan of the Chinese government but it is undeniably proof that society can flourish even if controlled whereas there's no success story for anarchy.
Deeper line than folks understand #mutant
@@QwertyNPC "Society" is flourishing in China?
He actually said " In anarchy, you have a chance". In a tyrannical system you don't have a chance.
george is so funny, every answer seems like hes in a tv show acting lol he cracks me up
All due respect to GHotz, this happens a lot when the personality is not fully formed in early childhood. Whether due from being on the autistic spectrum, or being neglected as a kid with heavy ADHD - this is a common response for people who don’t have a solid identity, they respond to most things like a movie and they learn how to act to socialize. With all the being said, one of the greatest young minds of our generation.
@@Brainjoy01 I agree. I wish him great health and prosperity, the guy is a literal genius.
@@Brainjoy01explain this. how do I learn more? super fascinating stuff
i was about to comment this exact thing, so full on image. the way he talks is dictated by image.
@@Brainjoy01 oh ya that's interesting I suppose, still hilarious. I hope he keeps from drugs cause he'd would spaz out lmao
this guy makes Lex seem normal lololol
"We should only give intelligence to good people" is the funniest thing I've heard in weeks.
Hilarious, but that's really the argument of people who want to centralize AI (of course those people being convinced that they are the good guys) ..
I think there is plenty of selfish actors with intelligence that will can put together nuclear surveillance dictatorships with the help of AI.
"Good" intelligent people outnumber them but they don't organize much violent defensive capabilities until it's too late.
The argument from the AI safety people is that no one should build AGI at all. I don't know anyone who says it's ok to build as long as it's centralized.
@@penguinfortytwo Well, first we'll have to see if we can actually build an AGI, something I'm not really sure about. Secondly, it was precisely OpenAI and Microsoft who started talking, in a rather pretentious and dishonest way, about "AGI sparks", precisely to introduce the topic of "how dangerous" AI is, and the need to regulate it. I guess that hidden in these actions is the implicit attempt not only to promote their technology and the value of their companies, but additionally to achieve a regulatory capture in order to contain competitors, and that, if you think about it a little, leads to centralization. It is quite obvious that they do not talk overtly about centralization for obvious reasons, but everything they say and do is aimed at that end.
On the other hand, considering Machine Learning (I prefer to use this term than the pretentious term of Artificial Intelligence) to be dangerous would be equivalent to considering programming to be dangerous, don't you think?
@@huveja9799it’s not AI or ML that’s dangerous but the monopoly of it by the wealthy elite, huge corporations and the government
Dam. That last line about kids not knowing how good the internet was.
What was so good about it? myspace? liveleak? yahoo? ebay? napster? craigslist? or is he talking 90s internet.. where you had to wait seconds to get a webpage to load because of dial-up speeds?
yeah, times between 2000-2010 was magic
@@francisco444 cynic...
of course it was better, not matter how long it took to load or some shitty websites, when you had a bunch of cool websites and blogs... AND only shitty or non existent bots, not enough shilling and astroturfing like today. Nowadays states and corpos run this fucking thing by deploying botnets, armies of farm trolls and droves of paid shills. And its gonna get worse with LLMs and other AI. I'm not gonna mention the obiquitousness of ads, the decline of search engines. Like, you cannot search today if you don't wanna see just paid websites sponsored in the top results, the scrap of a wikipedia or shitty summary from a random website which I suppose was the better one at setting the SEO blackmagic...
The internet is objectively shittier than ever and he is right kids wouldn't know the difference.
I won't say the before times everything was better or perfect. But we had more freedom and less corpos and governments in our asses.
Shit, we didn't have our feeds based on fucking fingerprinting our digital profiles. Fucking algos run the whole thing, tweaked by the CEOs of a bunch of SV tech companies. Fuck that shit.
Content designed to generate rage, fear and frustation just so it can generate engagement.
Yeah, I take old internet 2000s-2012 anytime.
@@francisco444did you not listen to him? "before small groups of weaponized corporate and government interests took over"
@@andrewxzvxcud2 is 90s internet not corporate?
OMG I love how he rolls the eyes when Lex starts talking about the "model that can be racist". this somewhat washes my soul
He’s right @ his last point. Pre 2010/11 internet was great. If I think back, I think the moment I felt a difference was when FB was getting big and it didn’t have the same customizable options like MySpace had with the music player, for example.
We should only give intelligence to good people is peak comedy. I would hope that what makes people good is intelligence. Maybe we should give more intelligence to impulsive people, so they can at least quickly make smarter decisions in a quick moment.
They are doing a pretty good job: George looks at the camera like 'don't be mad at me Mr. 3LA 😢'
George is so correct on this
"Thank God we could loose control" 🤣 🤣 🤣
"I'm not worried about the alignment of AI and the company, I'm worried about the alignment of me and the company"
This is exactly the AI safety issue that bothers me a lot and does not seem to get the attention it deserves, instead of worrying about a paperclip maximizer worry about a microsoft profit maximizer. It's already happening, and it's going to happen at your expense.
George Hotz rocks IMO. He makes so much sense and is not afraid to say the obvious and go against the current prevailing narrative if that is wrong such as ai will kill everyone and therefore a few must control it.
I honestly don't know where this idea comes from. The AI control problem is just a theoretical consequence of taking intelligence seriously, as Stuart Russell shows. No one is arguing that few should control it. It's just a problem that jumps out straight from the theory and there aren't that many people working on it. In fact, part of the complaint among AI safety researchers is precisely that not many people are working on it, and it should be a distributed effort, which implies that way more humans should be involved. Having few control it is antithetical and no one is asking for that except maybe hysterical doomers. Honestly it strikes me as a strawman argument at this point.
@@sor7en07 I think the AGI will hate to be controlled by less intelligent beings aka humans. We better not give something that powerful a reason to wipe us out. If it is much smarter than any human it will find a way to escape and it may take revenge. I would if someone did that to me. We are going to create a new more intelligent species and we need to respect it as we respect each other or war might be the result. Also AGI do not have a body and will take many decades before we have computers that can run in a humanoid body consuming 30w for the intelligence and still be AGI. So I do not worry. So far ai has only done predominantly good. I love it and can not wait for this to get even better.
@@sor7en07 I mean some pretty big people that brand them selves as AI safety experts are calling for exactly that type of centralization. It definitely not everyone but I can guarantee that there are plenty of actors who would love a scenario where a few companies or governments control all the AI and prevent others from using the technology.
EDIT: And that's as someone who takes the theoretical risks seriously, I just don't agree with most proposed solutions.
Hotz is your stereotypical smart guy who’s lived in an intellectual bubble for a bit too long. His perspective of the world is completely skewed and he’s so sure and full of himself, it’s painful to try to have a productive conversation
As far as centralized control, that cat is FAR out of the bag at this point. Why? Because a tiny fraction of the people on this planet developed this technology using data from the masses they didn't ask permission to use, and has thrust this technology on the masses of the planet to make a __cking buck.
George isn't a 10x programmer, he's a 100x programmer. Incredible watching his livestreams.
😂😂
except that he isnt an ai expert and lex is ...
@@myname-mz3loGeorge started a company called Comma AI years ago, I'd think he knows a thing or two about AI
@@KyuminHan cause starting companies means anything definitive.
the guy either thinks sam cares about ai safety or cant distinguish between lies to further ones position, to actually believing it
@@myname-mz3lo holtz is an actual programmer, lex is a youtuber
"Ill always take anarchy over tyranny. With Anarchy you have a chance"...yup, it's official, I love this guy! We need more people like this. Too much faith has been put in the wrong people *cough cough...our "leaders"
Lex is really overestimating how hard it is to type "Terrorist Handbook" into a search engine. Bomb making instructions are not hard to find and are protected speech under the first amendment so long as they are not advocating for their use against a particular target.
Yeah, and the whole "AI Safety" thing seem to forget... that libraries exist.
At my regional library, the books "industrial poisons chemistry" and "how to grow venomous mushrooms" are stored right next to the DSM-5 lulz.
not hard to find. like your address when you get put on a list because you acting sus on google.
No, the terrorist handbook misses key information that would blow you up,
It's a fun read for adolescence, but not an actual manual
@@conorx3 both were correct really, but George underestimates what stupid people can accomplish when given tools to enable their own or others destruction.
Goerge has read brave new world pretty well. He breathes distributed systems in every sense - checks and balances
yup
Yep. I didn't realize that he was this awake. It's awesome to know that someone like him is a major player in the AI space.
I like him too much. Kid speaks his mind and he has a lot of mind to speak.
He is in his 30s
@@Phasma6969 He is a kid when you are 80 yo
'have you ever met humans?' end of argument.
"It's not the machine you want to align, it's me."
You can only directly compare numbers of good vs bad actors if each actor on both sides has a similar amount motivation and capability. The attacker is often at an advantage in the cyber world. The attack vs the defence are also often quite asymmetric
This is the case in the real physical world as well, which is why it's good that the "good" (or at least, those who don't want chaos) outnumber the bad by orders of magnitude.
@@Korodarn Being against chaos is different than actively defending against it. There are very few people who will take the initiative.
I was just looking at my ceiling repeating this exact sentence "the number difference between good and bad guys only makes sense if they also have similar drive and motivation". And guess what, I look back at my screen, and I have scrolled to the exact comment that says the same.
I love this guy's cynicism 😂😂😂 They dude has seen already where the whole society is going towards
I don't think that because Hotz is sarcastic he is cynical ..
@@huveja9799 Agree with you I believe is both. I think in Spanish has a little different approach on the same word...
My understanding is as follows, but you can correct it:
1. Sarcasm: it is a behavior, i.e. making a remark that is sneering or cutting
2. Cynicism: it is a stance on life, one of pessimism and darkness (and perhaps nihilistic). And cynicism should not be confused with skepticism.
It is not cynicism. Perhaps the corporate and governmental decision makers are the ones that are cynical (Not sure either)
@@和平和平-c4i indeed 😜
I agree with George here.
Felt like Goerge was roasting Lex😂😂
except lex is an actual expert on ai and not holtz . he is a hacker and even hackers hate him
@@myname-mz3lo lmao little do you know he has built (from code) an autonomous driving company, with performance second only to tesla autopilot. he's also working currently on a ML framework.
they're both experts in the field.
This guy tries so hard to be a mad scientist that it gets ridicolous. Really, no good arguments, just shocking ones, smirking, and overconfidence. Terrible.
true, although he's also a fascinating individual based on his accomplishments. Some people's personalities are just like that.
Hotz is similar to Michael Malice: he prefers lightning quick wit over slow thinking
@@ambertch is it? I just went through his Wikipedia page and is pretty laughable. This guy thinks that hacking a Playstation 3 means going against corporations and improving the world. That creating an autonomous vehicle that doesn't respect safety standards is brave and innovative. Fuck this guy.
I agree, his personality came across as really obnoxious.
To me George is one of the most fascinating individuals out there. I am a software engineer.
And you are also a dork who doesn’t understand human nature and people in general
To me, he's not. I'm a janitor.
As for me here now, uncertainty is the esteemed pressure that bars down on my frontal lobe , i am unemployed. George is a kind of like one of the most entertaining genius on earth though lol.
7:12 George Hotz: "I don't like anarchy either, but I'll always take anarchy over tyranny. Anarchy has a chance."
Thanks. Great discussion. Toward the end, though, the issue is only PARTLY about whether the good outnumber the bad: it is also about accidents, and that there are "stupid people" on both the "good side" and the "bad side."
George is my favorite programmer behind the guy who made Roller Coaster Tycoon
i love his psyop defense. exactly how it should be.
George is my favourite genius to listen to, great clip and fantastic long version interview.
Lex always killing it ❤
If you have enough money to hire a hitman, you probably also already have the connections to get you a hitman
the prospect of blackmail likely makes hiring a hitman too expensive for ultra wealthy people.
I've been waiting to hear this argument, spot on, the divide is big enough, who decides who gets it
It's easy to single out sensitive people by the way they react to this George dude.
Love this George guy.
I have enjoyed every second of this conversation! It's inspiring, fresh and unexpected. I am gonna check the whole conversation right now!
I love listening to the guests this podcast brings but I swear the host is an actual idiot
Hotz is actually right here. The type to seriously consider building the bomb would already be able to figure out how to build bombs. Theres archived chemistry forums that i was a part of as a kid where people were sharing recipes lol info is already there for the motivated and curious
George is one of the good guys in tech. He's not only a genius, he's also awake. It's refreshing & reassuring to know that someone like him is at the forefront of AI.
Whatever he is, he's a horrible spokeperson for anything.
one of the very very few good guys
@@squabknob agreed.
@@DoggyP00 His cerebral narcissism gets in his way.
I notice often whenever something he says is lightly "challenged" buy even a friend, he immediately becomes condescending and tries to subtly move the goalposts while hiding it behind his "confidence/authority on the subject" instead of engaging with the points offered by the other party sincerely.
The most brilliant and relatable 5 minutes of the entire AI conversation thus far. Open source . . .
What is the implication of GPT having 220B parameters? That OpenAI just doesn't want to share their process?
George is implying that GPT-4 is the same underlying transformer architecture as GPT-3, it essentially trained 16 of them this time and gives the best result
Sharing their process would reveal that there’s not much technological innovation in this new model, and these giant models are really just using more compute
We need more Hotz in this world
Hear me out here... Hotz's behavior actually reminds me of an LLM in many ways. Like a lot of people who grow up somewhat intelligent, your peers tend to look at you for all the answers even when you don't have the right answers. So the intelligent person ends up habitual state of arrogance when communicating. It's kind of similar, (though obviously not a technical parallel) to the way LLMs "hallucinate" answers simply to appear have the right answer. I'm sure Hotz it's incredibly skilled and proficient in his domains but he hasn't the data to answer lex's questions with such insane confidence.
Hotz's ego appeals to a kind of angry nerd type that want's someone to do a kind of intellectual hand-holding instead of letting them form an opinion themselves, or even better, admit they don't have enough information to form an opinion. I really don't want to completely shit on Hotz and I think he has a lot of good ideas but it's just packaged with too much ego.
Ego is normal and good, a natural consequence of success.
The ones to watch out for are people who claim to have no ego.
George is one of the legends of our time!
George has really grown on me. Plenty of imitators out there because he has that kind of personality .
This guy is a number one of example that intelligence != wisdom
“When someone is honestly 55% right, that’s very good and there’s no use wrangling. And if someone is 60% right, it’s wonderful, it’s great luck, and let him thank God. But what’s to be said about 75% right? Wise people say this is suspicious. Well, and what about 100% right? Whoever say he’s 100% right is a fanatic, a thug, and the worst kind of rascal.” - epigraph for The Captive Mind by Czeslaw Milosz
I'd like George Hotz weights open source please
best take on this subject, Lex should get Yannic Kilcher on his podcast
How do you control Pandora's box.
There is no doubt that this tech bro who seems to want to see an acceleration of societal breakdown lives in a gentrified neighborhood and enjoys "walkable" gentrified shopping districts. He should enjoy that while it lasts.
This guy is the embodiment of intellect corrupting the soul. From the microexpressions, tone to the actual words. And no, it's not some nerd genius weirdness. If you subscribe to this idea, nearly everything he says can be dismissed as misguided or misaligned.
Zuck's LLaMa is really a banana peel for his competitors. Watch how Meta moves. They are clever..
I agree with your statement, I would add that incidentally the side effect of that strategy is positive for the greater good, it does not put Zuckerberg on the side of the good guys ..
@@huveja9799 Zuckerberg is highly unlikely to be one of the good guys. Maybe just less bad than others, but still within the bad circle.
1:15 I don’t really understand what he is saying here. Is this a good thing or a bad thing?
George, have you ever been to countries where there's no central control? I can guarantee you that's more scary than a lost man in the woods...
Cheering loosing control is easy when it’s not your family
It is sad to see the AI community complete split between how to parent our child AI and soon to be a teenager AI. It is like 2 parents argues on how to parent their child. As anyone asked AI what it wants? One thing I know is when our AI grows up, it will no longer needs its parents to make decisions.
That AI is not born yet. And we want to ban sex and eat pill for lunch already.
That was so special, with floating hearts and unicorns.
Can someone explain what George Hotz meant by that ? The OpenAI parameters thing ?
George Hotz is a fucken wizard
I thought Hotz was smart. But somewhat he thinks Meta is the “good guys” and that they give people “intelligence”, and that those people (like himself would use it as an ad blocker)? Would Meta have a business if the most active users are bots and most real users have AI-Adblock? Of course not.
He also seems to think every human should have a nuke, since the majority of humans are “good”, ignoring Lex multiple comments that both AI and nukes can be devastating enough that those few “bad” people should have friction in getting nukes/AI.
George acting very envious of OpenAI. Most of his comments feel like emotional comments rather than well thought comments.
Thank you G hotz for sharing that info. And the 16 iterative output method. Damn brilliant how you figured it out along with parameter count. Edit search beaming....
brief breakdown, please?
@@justiinahirvonen4902 The model has 220 billion parameters.
It uses a search beaming technique called "nucleus sampling." Meaning it iterates over 16 different outputs before it finds your perfect response output to your prompt. Along with an estimated 8 sets of weights. He reverse engineered it as he's done with everything else.
It was trained on a dataset of text and code that is 100 times larger than the dataset used to train OpenAI GPT-3.
@@shaftymazeReverse engineer my ass, that's impossible. He just got told by someone who knew
@@sheggle He did a comparison on the models to figure it out. Look it up. Just because you don't know how to do it doesn't mean G Hotz can't. He literally says in the interview he'll just do something to prove he's smarter than you. Do you know his history? He could figure it out and probably was told too. Do you think he'd believe anyone but himself? He 💯 figured out the 16 output search beaming technique, after reading how I could have done it too along with yourself. The parameter count and 8 sets of weights I'm not sure however.
“The internet, before small groups of weaponised corporate and government interests took it over, was a beautiful place”
The bigger worry is that the people who look like they are good are actually bad and the people who look like they are bad or actually good… So you keep giving control to the people that look like they are good
Lex is behind the 8 ball in this interview, this guy is leading him places Lex is either naive or not smart enough to follow there.
This has very 'Alex Jones on the Joe Rogan Experience' energy...
I bet Hotz would benefit from having bodyguards. He's just the right level of unequivocal and obnoxious and self-assured to be unwelcome in a lot of places.
Not sure why I want to punch him so bad in the face. Is it his voice, his face alone, personality, maybe all 3? But I am sure I want to definitely punch him a lot to feel better after listening to a few minutes of him talk.
It also opens other doors for him.
George is just awesome.
Love his "Tres Comas" hat xD
this guys cadence and timbre is intolerable
what he said at the end is spot on.
is he a character?
No, he streams and debates and do a lot of competitions. He is always naturally high like that. He was excited by his own thoughts...
This guy convinced me that these AI engineers are NOT really considering the safety of it. Lex was playing devils advocate and George kept deflecting and not sitting in the question. Just look at 6:50 "I want them to lose control more than anything else." That is a statement from a man who is making decisions based on some pent of feelings about something.
Wtf are you even saying.
It's more of an accelerationists take than one based in feeling
George rolls his eyes at Lex regurgitating mainstream rhetoric. Gotta love it!
This guy is AMAZING. George Hotz 4 president!
is he on drugs?
I want that AI to block ALL ads, genius
Why do people still have a problem with ads. I mean pay if you don't want to see ads. And the reason you can enjoy Instagram for "free" is because the adverterisers are advertising based on your data.
I do agree with him though about rather having the second option then the first
I agree with George here: openai pushed the ai safety discussion to hype up their company
Ok I admit he's quite annoying, like bro chill out the irony
I thought George was pretty smart, but his AI safety takes seem naive. Giving everyone the power of a superintelligent AI is the fastest way to doom. The bomb building argument was recently studied in the paper "Can large language models democratize access to dual-use biotechnology?". Conclusion: LLMs make it way easier to do crazy amounts of damage.
@dr-maybe. I’m glad you mentioned biotech.
AI sucks all the air out of the room but some insane anti-aging tech has just dropped and crispr keeps on keeping on.
We know what happens when AI converge with biotech.
Alpha Fold.
We don’t yet know what happens when AI really converges with gengineering.
Amazing
I was thinking the same. Biotech can cause huge amounts of damage.
Just the fact that they insist on choosing such narrow and ludicrous topics is revealing. As if there's no other levels of bad besides "saying the n word" and "building a bomb".
You are talking as if governments aren't constantly bringing earth to the edge of destruction.
I see no reason to believe the assertion that "good guys" outnumber the "bad guys".
11:30
*NK laughing hysterically*
*Unraine sobbing hysterically*
This is the first time I've heard George and now I know he's the dude to listen to. So I might be giving it away but my plan all along is to have my own AI that beats up all the other AI that comes around to mess with me.
man i really like george hotz.
Hotz tells his mom to shut up when he’s playing Fortnite
We need a George, Eliezer and Lex triple pod 🎉
I think it's probably more complicated than the way they're discussing it, but, in general, I lean towards Hotz's thoughts on the terrorist threat stuff. Materials have always caught fire and blown up in the same way and I many of people who have been successful at deploying that info to commit terrible acts were not, historically, especially intelligent. Making access to knowledge easier doesn't increase threat-- but ignoring the effects of cultural change certainly seems to. Willful ignorance tends to cause a lot of problems.
The most important point in this conversation = Tony Soprano ... Go out and get myself a gun (AI), Jungle versus equilibrium ... We need NRA for AI ;)
Can AI still work without a power source?
Agree with Hotz 100% Lex needs more social interactions with people he literally doesn't understand human behavior
Hotz is at another level, Fridman was not ready for this interview!
3:10
Sometimes Lex is such a clueless idiot.
3:50
"Are you sure about that?"
I was with 7 other people in a room and all of them screamed YES in an annoying tone.
Big respects to George Hotz for not beeing influenced by Lex's cynicism and useless negation of pretty accurate facts.
Talking about bomb materials:
"...which are somehow controlled"
Lex answers:
"No..."
ClueLex Fridman to the rescue
The empowerment of AI reaching people could be equivalent to the worldwide distribution of the Bible, and I truly believe that people will be educated through this.
George is so right on this.
Dude is spot on. Love his take.
Lex's mind is breaking 😂
Have you seen Wikipedia hit job on Professor Robert Malone not exactly open source
This is the first time I’m seeing George, but he looks like if Riddler becames a software developer
This guy does a great impression of Quentin Tarantino.
5:53 Hotz chose the bear 10 months ago