Two Minute Papers - 2021-01-16
❤️ Check out Perceptilabs and sign up for a free demo here: https://www.perceptilabs.com/papers 📝 The blog post on "DALL-E: Creating Images from Text" is available here: https://openai.com/blog/dall-e/ Tweet sources: - Code completion: https://twitter.com/gdm3000/status/1151469462614368256 - Website layout: https://twitter.com/sharifshameem/status/1283322990625607681 - Population data: https://twitter.com/pavtalk/status/1285410751092416513 🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible: Aleksandr Mashrabov, Alex Haro, Alex Serban, Alex Paden, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bruno Mikuš, Bryan Learn, Christian Ahlin, Eric Haddad, Eric Lau, Eric Martel, Gordon Child, Haris Husic, Jace O'Brien, Javier Bustamante, Joshua Goller, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Michael Albrecht, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Ramsey Elbasheer, Robin Graham, Steef, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Tybie Fitzhugh. If you wish to support the series, click here: https://www.patreon.com/TwoMinutePapers Thumbnail background image credits: https://pixabay.com/images/id-3202725/ Károly Zsolnai-Fehér's links: Instagram: https://www.instagram.com/twominutepapers/ Twitter: https://twitter.com/twominutepapers Web: https://cg.tuwien.ac.at/~zsolnai/ #openai #dalle #dalle2
WHOA THIS IS CRAZY
The clocks are the most impressive to me, because some of them have literally all the numbers 1-12 in the correct place! Some of the clocks... well, not so much... but still. That must take so much data to learn patterns like that!
Hi cary! Nice to see you, and waiting for your next videos!
1. hi cary kopenhagen i love your vids
2. i literally was about to comment the same thing lol
I think the problem with the hour marks comes with some of the images of clocks used for training were ones with regular Arabic numbers and others were with Roman numerals and then some were with just dots or nothing at all. So it gets confused about what the signs should look like and it draws something that resembles Arabic as well as Roman signs.
carykh! I'm so happy you found this channel! :)
This will revolutionize the meme community. I can't imagine how crazy will be the memes of the future
You get a meme idea then "Hey Siri" it into existence
"In the future, humors will be autogenerated"
@Matthew Boyd Thanks <3
Hold into your papers, like that cat was doing
"Cat here likely holds a piece of paper"
This AI has become too powerful.
It even gets our memes.
noooo not the memes
"In the future, humors will be autogenerated"
The cat is holding onto its paper.
I'm waiting for the day we hear the phrase: "Hold onto your papers, the entire episode was generated by this algorythm"
At least the script by GPT-3
2040: "Generate a good movie".
2 papers down the line
@GS4444 I had a concept in my head but it would be slow
We don’t have to worry about (or waste time wishing for) an AI generating a good movie. AI needs to be trained with an enormous amount of appropriate data in order to become effective. There is not an enormous amount of good movies. What we need to do instead is fear the near future when some movie exec says “churn out 20 cheap, easy, revenue-generating summer blockbusters”. 🤣😭😂
Why not 2028.
I doubt it will be realized in anytime soon. Beyomd imitiating a sequence of frames, learning the underlying emotions and narratives represented by 2 hours of high-quality images and sounds would be so much difficult.
The stock photos industry is gonna tank lol
@Darin Krumov Wait until this thing can animate too, and animate 3D, and make sounds too... Also the filmscript because why not, right?
@koyint Renamon and Rouge on a beach.
They'll most likely use this to generate images on the fly for folks.
Exactly what I was thinking 😂 I wanna get in on the gold rush how do I implement dall-e?
@Robbie P probably won’t be used in businesses for years though
Glad to see the honor of the bestagon has been restored.
Onwards to a world with more hexagonal clocks! ⏰
And i'm glad to see i'm not the only CGP Grey fan here.
Hexagons are the Bestagons
Shenanigans beget shenanigans
@Jacob Griffin that's my one of if not my favorite video
This would be amazing in 3d for procedurally generated games.
@Karen Reddy Have you played the <AI dungeon>? It is close to that
This already could be sufficient for things like UI elements such as ability or item icons.
@안해찬 imagine ai dungeon with DALL-E
@안해찬 Would not recommend. The creators are breaching user privacy, I heard they are even deleting adventures.
@Karen Reddy AI driven characters could also have an arbitrary body, meaning that you could make them adapt in that way too (got ideas for an open world game in which the (wildlife) NPCs (probably including plants too) evolve completely over time, globally (not just isolated to your own computer)
god this can be such an amazing tool for design, like its literally a "this, but..." machine
@May the Science be with You We can fire all designers along with all the artists and lock them up somewhere. Let's call it perma-quarantine. Their lives are no longer required, and it will be better for the climate to get rid of them altogether.
@clray123 this is why ai is going to far people are losing jobs lol. And it’s not a good thing
@Ben cgi and origami Basically there are too many people... the great majority of humans (myself included) don't produce anything essential - the "corona" politics are proof of that:
You can lock up significant portions of population and basically nothing happens - because they were "unproductive parasites" to begin with and the "services" they provide to each other in the end are just wasting energy / natural resources (not to mention the endless bickering among themselves). Since the consumable natural resources unlike the number of human consumers are limited, having more "AI" and fewer humans might actually be reasonable (though only if the AI can provide for itself and consumes less of the energy/resources which is not clear at all at this point - biological organisms are remarkably energy efficient compared to the machines we produce).
Also, it appears that the superrich know that and wish to basically enslave and gently eradicate humanity before it grows so large that it wipes itself out in an uncontrolled fashion.
@clray123 jeez are you alright?
@Dominic Stocker well, do you have any counterargument to what I said?
Can't wait to see games with these procedural signs.
@Stanislav Teliatnikov It's a different thing, not better not worse.
@FunkyPrince imagine ai dungeon with DALL-E
@RADINIUM update the game 😉
@Kennneth ummm oh god i will do that rn
@Kennneth what is different exactly?
Aw, I wish we could play around with custom text ourselves. :(
In my opinion one of the best channels on YouTube!
this one and Ben Eater's :)
and by cloud
Dr. Karoly: “Dear fellow scholars,”
Me: Day 42 and the scholars still haven’t noticed I’m a uni dropout
I've never even /been/ to uni, but this stuff is just so damn interesting
One thing I found funny about this is that the AI doesn't seem to know what a heptagon looks like. Presumably because there aren't that many images out there.
Henry Ambrose Hey I remember you lol
@Poll Person oh shit lol
Would be cool if it accepted input like GPT-3, so you can give it a couple of images of heptagons and then ask it to generate more.
@Luke Haulin That would probably take a LOT of images.
"It could figure out that this cat likely held a piece of paper" The future is an awe inspiring and terrifyingly scary place full of wonders I wouldn't not have thought would exist as a child.
"not two minutes, and not about a paper"
That's why we are here
Thank you for such brilliant and consistent content! Every new video makes my day
I hope the folks working on AI Dungeon get a hold of this at some point for some AI-illustrated adventures
If you've ever had a lucid dream, you know that text and especially (digital) clocks look very similar to what this A.I. creates (unreadable, yet familar). This phenomonon is universal to all who has lucid dreams.
Avacado Chair. This is the future
This is the way
Most of this shit is nightmare fuel NGL.
Torus of porcupine.
I was legitimately afraid of mixing some of the terms around in the interactive blog. The unexpected consequences of changing simple terms were enough to generate anxiety and terror. The suspense was palpable. Will you get a sensible image or will you get something that violates your subconscious? Wowee, Salvador Dalí ain't got shit on machine learning's naturally occurring surrealism.
Once you can manually input anything to this AI online who knows what are the limits of what you could generate?
this is the way
insert surprised avocado emoji
This looks amazing. I can’t wait to try it.
Also, I think we’re gonna see a lot of unemployed graphic artists in the future.
Ok, I was not holding onto my papers hard enough for this one
Mind-blowing, it's hard imagine that what we write comes out as a picture.
This is absolutely insane, some of the most impressive results I've seen from an AI. A lot of jobs are going to be eliminated/reworked over the coming decade to be sure, this AI can already create fully usable cartoon art and illustrations, and certainly original and creative avocado-inspired chair designs. I can already see designers using this AI to significantly improve and speed up their work. Use this AI to get the general idea then just polishing left. Eventually AI's could make us entire movies, musical albums, artwork, symphonies, etc, and they would be completely free for the public.
Another chunk of humanity is going to be left out in the cold because there will be no place for them. Everyone thinks AI and robotics will take over menial, repetitive jobs, but even those pursuing art, design, etc will lose out in some form. There will be a few still doing their thing and being paid. But no one will pay the majority of them and their value will drop even further.
@CheapSushi Yes I agree totally, I'm not sure if its a bad or good thing. In a way it would really be a shame but it could also help us as a species in the end. I do not believe humanity will stop designing / creating art completely, only it will end up mostly being for entertainment rather than a professional business opportunity.
@nilsragnar I think one benefit is that Universal Basic Income will become a real thing. But then I worry about after. I feel like we might end up in some kind of Gattaca kind of situation, where only the "best" or "superior" get to truly exist and through time less and less people will have a role in the world. Everyone person on Universal Basic Income might be seen as a long term drain and population control and immigration control I think would become stricter. And then through time, even probably birth in the first place might become very controlled. It could lead down a very dark path. Good for the human race long term, in a way, good for the Earth in a way, but lots of ethical and moral issues and humanism goes out the window. And since the world itself is unbalanced, a lot of people in many other countries are going to suffer tremendously because those countries probably won't be able to handle the economic burden unlike the USA or Europe. I mean creative wise, how many people are using services where you hire 3rd world people to write things, to draw things, the edit things or even design things because the US counterpart is more expensive? I think in the future, the US counterpart becomes more ideal because of location and the grunt work goes to the AI, so all those previous 3rd worlders getting some income are completely out of the equation.
Thanks so much for showing me PerceptiLabs. It's such an amazing tool. I hope the devs get enough support and investors. This DALL-E is also damn amazing. Can't wait to play with this if it becomes publicly available.
Very grateful for your videos man! Love the quote "Fighter jet for the mind"
Thank you for this video, otherwise I probably wouldn't have seen this absolutely amazing techology!
Woah, tons of applications for this. It feels like I'm peering into the future. Unfortunately I don't understand the limitations of this model, but seeing how ml algorithms are becoming more able to generate content based on descriptions, I wouldn't be surprised if we were able to use this stuff in real applications in the next 5-10 years. I'd really like to see something like this to be used in procedurally generating game stories. I did try ai dungeon, and it demonstrates that a huge room for improvement but hopefully they'll be fixed in the coming years.
I really wonder how this type of technology will impact artists and designers.
This just blows my mind, the accuracy of it is honestly impressive.
The AI almost seems creative in a human way, if someone told me to come up with a storefront for a particular name I can see the thought process being similar.
I feel like this is all leading to "Star Trek"-like programming.
Enter the Holodeck: "Computer" beep beep "Create a cozy cafe, location New Orleans. Add an armchair in the shape of an avocado" beep "Make it a bit less bright, and play some Jazz"
Combine it with a 3D-printer and we almost get a replicator (- well, no „Earl-Grey hot, but still: make a chair sounds possible...)
I'm just imagining running the comic book scripts through this to see what the output would be like. Many comic book writers are pretty detailed about what the scene and framing should look like. Of course, that means creating a series of these would need some way to preserve the style between panels in a more automated way.
This would be amazing in combination with something like AI Dungeon.
Next step: Generate one high quality emoji avocado, click on the one you like and generate all the missing emotions for the same exact style to make a full set. From there, ist only a small step to AI-generated real-time video-chat avatars, which display the proper emotion of the what you just said with proper facial expressions. .. I'll propably be a talking cup of coffee. :O
GPT-3 seems to have opened the door for creating practical, highly effective A.I. systems. I wonder how many industries will be able to use this model to create groundbreaking new tools. Being a audio engineer and sound designer. I would love to have a tool similar to this, translated for working with audio files.
Truly impressive.
What a great paper! Thank you for the videos, I love your channel.
Also, PerceptiLabs looks amazing, thank you for introducing that tool!
Fantastic! Thank you so much for continuously sharing the future with us!
This is great. I saw GPT-3 presentation video also. As a tech guy, I understood its value. I would request you to specify "how to apply this" or "how it can be used" in general terms, that will make video more appreciated by layman or common people.
so now we can answer the question: "what if a dog wore pants?"
But how would a dog walk a dog ?
@John Doe yes
@Crumble just like Goofy and Pluto
yesss
@Sansirow Makes sense.
Played with it for a minute and already laughing like crazy. These cat elephant chimeras are priceless.
Amazing! But I’m quite happy to say that I could not even find one creation that was not off aesthetically. Humans still winning 😉👍
I'm wondering is it possible to learn and simulate motion with this algorithm. Or do we have to wait for LeCun for that. I can see its phenomenal progess. I'm just wondering if these results mean what they imply.
I can't wait for this to be publicly available, having so many great ideas with this
You already can try it on their website, its fun
Meant with trying out with own words and examples, not just the pre-set ones
It feels like AGI will be created in this decade. I can already imagine an agent with textual stream of thought generated by GPT-3, fused with a visual stream of imagination generated by DALL-E.
How about focus on make one thing great instead of endless things half good?
@WebDev I can make one thing great, I can make endless things half-good, I can do practically anything within AI field, if someone will pay me competitive salary for doing it :)
As a long time Ai advocate, I can genuinely say this is the first time the results have scared me! Amazingly impressive... but thr implications of this line of research.... 🤔
I'm so happy i found this channel, so interesting and eye opening. I wish I was smart to create things like this :/
this is how work in future is going to look like. You need to be an expert to evaluate generated result, know how to tweak it, and you are going to be communicating with other parts of your team, while most of labor-intensive work will be done by AI
I’m currently doing an ML class in my masters program. I love this topic so much and have never been fascinated by a field of research so much. It’s just so demotivating that with my mediocre math knowledge I might never be capable of really advancing in this field..
I was waiting for your video on this! As a designer and dev this is super exciting, and hopefully productized in the future!
You should do a video about 15.ai. It does a good job synthesizing voice lines from various characters
Zachary Marion - 2021-01-16
This is insane, as a programmer and designer, I'm jumping up and down right now
Christian's Games - 2021-01-25
@Sam Blackmore I have to disagree or do I? I don't know what I think about your comment, but it scares me either way.
Martin Mclean - 2021-02-27
@Dan Iel lol. Really. Your lack of imagination makes me laugh and cry at the same time. Rather ironic that the topic is ai imagination...
Aee Bee Cee - 2021-03-24
almost the same reaction
coltin engle - 2021-09-03
@veggiet2009 Really? it's hard to beat the speed of light m8
coltin engle - 2021-09-03
@Dan Iel Hey why'd you stop responding to Karen, I was enjoying the banter