DeepMind’s New AI: History In The Making!

preview_player
Показать описание

📝 The paper "Faster sorting algorithms discovered using deep reinforcement learning" is available here:

My latest paper on simulations that look almost like reality is available for free here:

Or this is the orig. Nature Physics link with clickable citations:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Benji Rabhan, Bret Brizzee, Bryan Learn, B Shang, Christian Ahlin, Eric Martel, Geronimo Moralez, Gordon Child, Jace O'Brien, Jack Lukic, John Le, Kenneth Davis, Klaus Busse, Kyle Davis, Lorin Atzberger, Lukas Biewald, Martin, Matthew Valle, Michael Albrecht, Michael Tedder, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Richard Sundvall, Steef, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.

Károly Zsolnai-Fehér's links:

#alphadev
Рекомендации по теме
Комментарии
Автор

This is incredible! The problems this can solve when joining it with GPT4, Wolfram Alpha, and web browsing are limitless!

DTrackFinder
Автор

2 minute papers has turned into 5 minute AI papers...

And I'm not complaining haha. Love it!

carlosreyesf
Автор

This sadly isnt as cool as it may sound. libcpp sort3 was unoptimized as some people pointed out. On the other hand, it confirmed that what we already have is likely the fastest solution there is.

nulled
Автор

It’d be interesting to hear what the AI’s implementation actually was, what was the wild invention which exceeds the current algorithms?

slothsarecool
Автор

an AI to recursively improve compilers would be nice

swaystar
Автор

As a dev, I am thrilled and terrified at the same time

FaffyWaffles
Автор

This video contains extreme misinformation. Quote from HN:
'As someone that knows a thing or two about sorting... bullshit. No new algorithms were uncovered, and the work here did not lead to the claimed improvements.

They found a sequence of assembly that saves... one MOV. That's it. And it's not even novel, it's simply an unrolled insertion sort on three elements. That their patch for libc++ is 70% faster for small inputs is only due to the library not having an efficient implementation with a branchless sorting network beforehand. Those are not novel either, they already exist, made by humans.'

_midinette_
Автор

0:40 to be honest, alphastar(starcraft 2 vrsion of alphazero) wasnt able to beat human pros under restrictions(ie having a cap on key presses, like pressing 500 keys per minute and not 6000), but it was still remarkably good, it reached top 200 europe, and even invented some fun strategies. just saying that, unlike chess, it didnt achieve superhuman level in strategical sense

shame
Автор

Let's train a AI to create better algorithms to run neural networks next.

stefang
Автор

If anyone has ever read the Ender's Game series beyond the first book. Remember the first AGI? "Jane"? She was created by Ender's Game...a video game. What's interesting is this is literally happening right in front of our eyes when this book series was written over 30 years ago.

artificial-ryan
Автор

Well, it was a good run fellow programmers. The time has come. o7

GnuSnu
Автор

Kinda funny. If you let humans have fun with games that represent work/learning they perform much better, AI seems to be more similar to us in this regard apparently than one might think considering it learns via gamification xD

DreckbobBratpfanne
Автор

That’s freaking incredible. I think they have a good plan… but why not teach an AI to make a more efficient coding language?

Quantum_Nebula
Автор

Incredible! From what it seems like here, this requires knowing exactly where you're going with the program (knowing the end result), which is something most LLMs struggle with. How long can these programs get?

bupp
Автор

Worth asking: How much more efficient than a brute force was the process to find those solutions.

DrN
Автор

Any plans to cover the “Textbooks are all you need” paper? It covers the new Phi-1 model which has very high coding proficiency despite small training sets and low training cost/time.

genericwannabe
Автор

Can't wait for the audio version of that sorting algorithm

azafreak
Автор

*Amazing❤️🙌Being able to provide all my needs without the help of the Government is really a dream come through and I’m getting $50, 000 returns from my 10k investment, Glory to the everlasting God almighty.*

marcusjohansen
Автор

I unironically play BitBurner where the verb you use to play the game is writing actual JavaScript against an API that represents what you can do in game

andre
Автор

In 4:08, what is the name of that tool/web/application? I want to try it too.

kingki