/sci/ - Science & Math

>>11387381
>Obviously top 1% intellect in his field
podcast hosting?

>>11387388
I knew people like you would show up.
>https://lexfridman.com/

I liked the podcast where Elon Musk dismissed his research in human-assisted driver AI as irrelevant and refused to talk about it

>>11387394
I’m sure he does great work but he is not a star at the top of the field, that’s for sure

>>11387417
better than us at least.
from his video I found out a poo is behind Alexa.

>>11387411
... isn’t Tesla autopilot driver-assisted ai?

>>11387430
Yeah, he was researching into if driver-assisted AI would outperform the AI alone and Musk basically said the AI will eventually be held back by driver assistants to researching into it is moot since the technology will quickly make it irrelevant. He was really smug and dismissive about it as well.

>>11387441
i think i agree with Elon in the long term, but i don’t know the current SOTA and how far away that is in the future. but if Tesla is currently using human-assisted it’s probably still a useful research direction...

>>11387441
AlphaZero(completely from scratch with no database) beat AlphaGo (learned from initial human database)

AlphaZero(0 human database) beat Stockfish(expert level AI with human rules/database of best chess moves)
LeelaZero(based on AlphaZero) beat Stockfish.

Human input will only be holding back AI imo. Elon wants to streamline their current Autopilot to have even less human input and going straight from raw sensor instead of converting shit to video/frames.

>>11387411
https://youtu.be/dEv99vxKjVI?t=1582
26:25

>>11387486
Yes he is not wrong.

>>11387485
i agree with elon but these comparisons are horrible. alphazero is just trained from self play on a simple game with simple rules and full knowledge of the state of the game (although yes complexity does arise from the number of possible games)

>>11387381
I like him. I enjoy his podcast because he introduces me to some tech leaders I would have otherwise not have known before, like Jim Keller (aka Silicon Ronin). Wish he would talk more into the architecture of TPU and how they differ to GPUs. Would be an interesting topic.

>>11387555
OpenAI with dota2
DeepStar with starcraft

Go game does not show its full state with AI or human. It's all paternity recognition. Chess isn't solved either. NN AI don't have full knowledge of the game, they make inferences based on data. They can't calculate all the possibilities.

I don't know, his success seems to come more from dedication and hard work rather than having a superior intellect, which is admirable of course. But he definitely is not quite on the genius level as some of the people in his podcast. Still like him though.

>>11387411
Definitely agree with Elon. History has shown fully automatic systems tend to beat "human assisted" AI in the long run (also see http://www.incompleteideas.net/IncIdeas/BitterLesson.html).).

>>11387560
based JSR anon

>>11387381
>Obviously top 1% intellect in his field.
He religiously believes that we live in a simulation. Stop mocking AI

>>11387381
one of the few based /sci/ approved podcast

>>11387629
>>Obviously top 1% intellect in his field.
>He religiously believes that we live in a simulation. Stop mocking AI
Source?

>>11387624
UNDERSTAND UNDERSTAND
UNDERSTAND UNDERSTAND
THE CONCEPT OF LOVE

>>11387641
https://twitter.com/lexfridman/status/1125432152697528320?lang=en

Just a little taste. I'm pretty sure he also talks about it on one of his recorded lectures about deep learning on MIT's youtube page

>>11387560
Watch the tesla autonomy day video on youtube. They go into their chip a bit more indepth.

>>11387486
what did he say?

>>11387661
The only one who talks about simulation theory is Elon Musk. I'll keep an eye open for it if you don't have a real source.

>>11387629
I think of this kind of stuff as something to draw more audience, make subject more popular, make things more intresting. The podcast is obviously not intended to be purely academic.
Wheter that is something harmfull or helpfull is another thing, maybe it spawns pseuds more than it inspires people to pursue their stem related passions (i hope there is a plural form) more seriously. I tend to think it is more of the latter, i really do no t know any other better place where i can find thst much of intetest-specific simmultaneously informational and entertaining content, by that i mean it's better than watching 3 hours of jre talking about ancient aliens, while still lex podcast provides a way to turn brain off after intelectually demanding work.

>>11387624
>>11387643
best JSRF song
https://www.youtube.com/watch?v=gdl-LIH1XgM

>>11387560
search for online college lectures on gpu architecture on youtube or something

>>11387682
Couldn't find the part where he mentions it in a lecture, but here's an interview with Joe Rogan where it's heavily implied that he believes in it

https://www.youtube.com/watch?v=ZFID_rgzE_Y

>>11387716
Thanks, fair enough. What was the point of your original post if I may ask, are you against his beliefs or do you like them?

https://www.youtube.com/watch?v=dEv99vxKjVI

Can anyone explain Elon's quick eye movement? He look's crazy

>>11387381
fuck off with your e-ceb youtuber shit

>>11387872
At least he's more intelligent than you. NEET midwit

>>11387866
He's thinking at an accelerated rate and does that to maintain situational awareness

>>11387381
Well, I'm listening to the P vs NP part of this interview https://www.youtube.com/watch?v=2BdBfsXbST8

When Knuth talked about classification of graphs, he spoke of classes of graphs that are closed under taking minor (shrinking edge to a vertex or deleting edge), one such class is planar graphs, but the description of such classes is clearly complicated, so kinda analogous to NP problem

Then he mentioned Robertson Seymour theorem https://en.wikipedia.org/wiki/Robertson%E2%80%93Seymour_theorem which tells you for any such class there is a finite list of graphs that can serve as a test, if it's not in the class, you can shrink it down to one in that list of bad minimal graph. When the class is planar, the list simply contains two graphs K(5) - everyone among five neighbors or K(3,3) - the connecting 3 utilities to 3 houses without crossing meme problem on sci

1:04:00 this dude asked "why do you bring up this theorem" lol obviously the finite list of bad minors is analogous to the finite exponents in a hypothetical algorithm to solve an NP problem, it's finite but it could be 2, 20 or 20000 so it might not be useful. I'm a layman in this topic but I can see that. Knuth brought this up as a case where P is not useful in practice.

So Knuth explained the obvious, then 1:06 this dude asked again why Knuth holds the intuition that P = NP is possible when it could be so counter intuitive, this annoys me, Knuth just explained that above (he gave an example that can give an intuition as to why P = NP might not be so paradoxical) Then the interview became a bit meandering about some Aliens shit lmao, idk this guy might be a good AI coder but he doesn't strike me as particularly clever.

>>11387381
He's clearly outclassed by most of the people he interviews on his podcast. Actually commendable as he can use it to improve himself unlike other grant chasing MIT professors.

>>11387381
he talks like a dimwitted retard, like his brain is slow

>>11387388
sure he's not hinton or bengio but those guys are the top 0.001% of the field. fridman is def in the top 1%.

>>11387381
yeah we get it, you watched the new 3b1b video. kys nigger

>>11388106
high iq?

>>11387394
woaaa.... he has a website..... top 1% material for sure.....

>>11389166
what's the issue with having a website, anon?

I can assure you that no one in the 1% intellect of deep learning is making podcast

>>11389188
based sauce but still better than you.

>>11389211
Actually not

>>11387769
Well, he does sort of state that his speculations in the simulation theory is mostly a thought experiment, but even so, I would say that noone in the top 1% in the field of AI even would consider simulation theory to be correct.

My point is that he's an AI superstar, sort of like how Jordan B Peterson is a psychology superstar, not that they're necessarily the best at what they do. Above average, sure, but not the best.

>>11387669
>I dont care
>im not in the mood to talk about that

Elon is chad autist

>>11389322
proof?

>>11388588
I agree with this.

He's also rather bad at playing guitar. He did a civet of the Joe Rogan theme and says he's an avid listener. top kek

>>11388588
based

>>11387560
Search for books if you want to understand that not easy to follow videos.As Keller said he reads few books a week for 50 years

>>11389162
High IQ = rapid eye movemen = thinking hard

>>11389687
How can this NEET, sitting at home, typing this very message if it's not already obvious, become like him?