# MEMES IS ALL YOU NEED - Deep Learning Meme Review - Episode 2 (Part 1 of 2) ## Метаданные - **Канал:** Yannic Kilcher - **YouTube:** https://www.youtube.com/watch?v=7DGlElSVYGo - **Дата:** 24.12.2020 - **Длительность:** 14:28 - **Просмотры:** 22,689 ## Описание #memes #science #ai Antonio and I critique the creme de la creme of Deep Learning memes. Music: Sunshower - LATASHÁ Papov - Yung Logos Sunny Days - Anno Domini Beats Trinity - Jeremy Blake More memes: facebook.com/convolutionalmemes Links: YouTube: https://www.youtube.com/c/yannickilcher Twitter: https://twitter.com/ykilcher Discord: https://discord.gg/4H8xxDF BitChute: https://www.bitchute.com/channel/yannic-kilcher Minds: https://www.minds.com/ykilcher Parler: https://parler.com/profile/YannicKilcher LinkedIn: https://www.linkedin.com/in/yannic-kilcher-488534136/ If you want to support me, the best thing to do is to share out the content :) If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this): SubscribeStar: https://www.subscribestar.com/yannickilcher Patreon: https://www.patreon.com/yannickilcher Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2 Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n ## Содержание ### [0:00](https://www.youtube.com/watch?v=7DGlElSVYGo) Segment 1 (00:00 - 05:00) yanny just kidnapped me and now i i'm and he told me okay antonio just pretend everything is fine just tell about the papers tell about the memes what's going on yannick we're gonna look at pictures and go haha all right we're back antonio's back welcome back to meme review well how dare you to assume i'm back antonio never left how's the channel going the channel's going fine it's at like 60 some thousand subscribers 60 million subscribers 60 million subscribers this is not financial advice he uses machine learning machine learns me oh it's still a bit like magic machine learning honestly like you understand everything it's still a bit like magic you understand everything i don't i'm not at youtube i don't even watch yannick i mean um what my own videos so yeah password mom can we have pie torch we have pizza at home pie torch at home psychic learn that was always the best after matlab of course and you know every time we do this actually there's a matlab email coming up now just for me an email just says just for me where is the new matlab 2021 a pre-release it's going to be hard for them for all the matlab users to make individual releases yeah exactly there must be at least like seven matlab users in the world jim just unsubscribed yesterday major revenue draw yeah they have to fire half the team 200 people oh so you're a human yes name every picture traffic lights i was like i think that's genius i feel enslaved yeah it's genius it's so genius to do that the first time i saw that i was like that's genius i don't have glasses literally anything if state is this interpretable ai yeah what is this thing fuzzy logic what is that what is that i think that's right if you write your code on wool if you sew it on speaking of wool oh yeah of course oh yeah because it's uh this is uh it's christmas today it's just christmas edition christmas we're going to do a copy edition later when are there cove with machine learning memes that was the effect of copied on machine learning we have conferences in gather talent uh yeah which is also the virtual pretzels that made me laugh in new york what's the virtual threat there was like um an event called okay at four we're gonna have virtual drinks and pretzels so in gather town so there's a function to follow someone if you click on the name of someone you can follow them so i stalked a bunch of people if someone walks by you just follow them and it's super creepy because it'll be like walking and you'll just be always walking behind them i don't like this run but i have to say i quite enjoyed gather town yeah i liked it i stopped a bunch of people it was like i was at my poster i wanted to talk to james martens now you're watching james and you're watching then every time he was like oops i have to go sorry a little p i have to go pee yeah sure it would be funny if there's toilets in gather town you know are there so i don't know there are you can also you can only like you know the things you pee like the things how's it called hey you already know the urinal yeah then you can only talk to the two on the left and right the video connection activates by the way uh thanks to all the discord members who are largely responsible for these memes thank you very much bunch of criminals double blind review gpt-3 paper it's open ai who knew oh my god that's how you do papers yeah well gpt-3 is now the best paper in europe was it in europe like it was an europe's i missed that i remember still last last year in europe's but you had this uh banjo person you know that guy the boxing the boxing he does boxing professional boxing nice and even though he does boxing people just you know stay very close to him yeah just i mean desire to die like you know ### [5:00](https://www.youtube.com/watch?v=7DGlElSVYGo&t=300s) Segment 2 (05:00 - 10:00) the society desire to die they asked him question and he was like i don't care i just want to do a fight and then anyone new at ai any new ai technology can it beat the stock market i think yeah i think this one this new one this one you will be able to beat the stock market transformers will beat the stock market you know that gpt3 you just ask it what's the price tomorrow it will tell you really it won't be correct but it will tell you we do have a channel on our discord about stock market prediction it's easily the most exciting channel i will check it out promise um check it out no you can't just not say prayer work you have to give proper recognition what about artificial curiosity what about predictive really next layer wx plus b smaller than zero stop please please enough good guys enough of you good guy really which model is this i am state of the art do you have the slightest idea how little that narrows it down okay so i watch all your videos and i know them all by heart all of them by heart and also i know them in reverse and basically i was wondering how much does improvement over state-of-the-art mean like really it means one paper but like percent wood it's if you have if you write the magic letters s-o-t-a with the first and the last capitalized the reviewers magically will lift from their chairs and up to the sky where they'll be treated to a massage come back down their hand will be guided to the accept button researcher is often obtaining sota performance by replacing rnns with transformers future is now old man future is now yeah the funny part is this is already old it's already older yes now people are replacing convnets with transformers and getting state of the art i never quoted a transformer yeah never did you from scratch no see no also i meant from scratch of course what do you think about multi-head attention ah that's the best it's just the best kind of attention between any kind of attention yeah and uh also like sometimes i i think it's better than i don't know eating and sleeping what's your favorite transformer multi-head attention i would also count bumblebee it's from the movie it's a car that can also be a robot transformers optimus i have been this guy and yeah sometimes i'm very like on some papers i must say but i'm very very um how's it called bloody yeah somebody yeah yeah i just don't do anything they're getting ready there's a little bit of a joy right yeah and just being what's once the last review i did it was like okay ah this was already done in and i decided i took the time to put the citation of like 10 papers to do this one two four all of them just to destroy them yeah but yeah it was not a good paper that is from xkcd and when you train predictive models on input from your users it can leak information in unexpected ways the person types in long live the revolution our next meeting will be at and the model completes it to the docks at midnight on june 28th see the interesting thing is that this the meme is about or the comic is about i'd say six months old at least but just this week a paper came out doing exactly this yeah crazy yes yeah this is like perfect prediction where should i find a paper is there a ### [10:00](https://www.youtube.com/watch?v=7DGlElSVYGo&t=600s) Segment 3 (10:00 - 14:00) video link to that uh it's going to be yeah there's going to be a video on that paper why are you late i um had to put on gabbert oh okay so this is cat or croissant and i have actually made for you a presentation where i'm going to test you first one cat or crossover i was a cat i was definitely a cat indeed a cat okay next one that was a croissant damn you're good okay elsa it was a cat okay that was a croissant okay next one was ever a very good croissant or a cat though it was a cat normal people can't go to the gym to work out because of lockdown me and ml engineer phd students europe's reviewers a well-written research paper our model is simple and easy to implement implementation when you hear someone still referring to europe says nips now that's a name i haven't heard in a long time i got used to that the question is do you say i was at nips 2016 or do you neep some sounds weird now it's on i know it sounds weird yeah i had the same experience why yeah we did it we time traveled but to what here let's ask that guy over there hey what's the coolest deep learning framework tensorflow we're in 2016. paddle 2021 it's going to happen i believe it paddle 2021 it's best framework when you're a python user and it's been five minutes since you haven't told anybody how it's better than tensorflow i don't know i don't even know what you use yannick but i mean i have to say i'll just don't ask just not to get angry at you otherwise and you'll stick with matlab well this one can also be applied to other things yes we'll make the title of this video meme review is all you need there's a paper on your desk saying like logarithmic bounds and where to find them yeah no it's like ah yeah it's like fantastic generalization measures and where to find them you should be like electro shocked when you submit this to archives like i think they got very accepted click bait oh it's also by banjo the brother okay by torch google tensorflow enable eager execution this was a disaster a design i didn't know tensorflow eager mode so pi torch was is always like dynamically constructing your graph you explained it to me yeah yannick you don't remember but you explained it to me probably i actually gave summer schools on this topic summer school yeah the best kind of summer school if you actually look at the tensorflow source code it is littered with if statements if eager then this part piece of code if not eager than this piece it's like two frameworks just bumped together into one because they wanted to copy pythord so much and uh so in a weird statement at that time ai was actually full of if statements now i understand the mean way better see it gives it a new meaning theoretically well understood the deep learning practices all the pages are the black quantified no yeah deep learning is not a thing this is me this is totally it's gonna be fuzzy logic i told you what do you think the future is gonna look like --- *Источник: https://ekstraktznaniy.ru/video/13242*