EPISODES:
Two Minute Papers: Perfect Virtual Hands – But At A Cost! Two Minute Papers: Virtual Characters Learn To Work Out … and Undergo Surgery Two Minute Papers: This is What Abraham Lincoln May Have Looked Like! Two Minute Papers: This AI Learned Boxing … with Serious Knockout Power! Two Minute Papers: Everybody Can Make Deepfakes Now! Two Minute Papers: AI Learns To Compute Game Physics In Microseconds! Two Minute Papers: DeepFake Detector AIs Are Good Too! Two Minute Papers: This AI Clones Your Voice After Listening for 5 Seconds Two Minute Papers: This AI Does Nothing In Games … And Still Wins! Two Minute Papers: OpenAI Five Beats World Champion DOTA2 Team 2-0! Two Minute Papers: 6 Life Lessons I Learned From AI Research Two Minute Papers: DeepMind’s AlphaStar Beats Humans 10-0 (or 1) Two Minute Papers: OpenAI Plays Hide and Seek … and Breaks The Game! Two Minute Papers: 4 Experiments Where the AI Outsmarted Its Creators Two Minute Papers: AI Learns to Animate Humanoids Two Minute Papers: Ken Burns Effect, Now In 3D! • Two Minute Papers: This AI Creates Human Faces From Your Sketches! • Two Minute Papers: Google’s New AI Puts Video Calls On Steroids! • Two Minute Papers: New AI Research Work Fixes Your Choppy Videos! • Two Minute Papers: Can an AI Learn Lip Reading? • Two Minute Papers: Two Shots of Green Screen Please! • Two Minute Papers: This AI Creates Dessert Photos … and more! • Two Minute Papers: NVIDIA’s AI Dreams Up Imaginary Celebrities #207 • Two Minute Papers: Beautiful Gooey Simulations, Now 10 Times Faster • Two Minute Papers: DeepMind’s New AI Dreams Up Videos on Many Topics • Two Minute Papers: How Do Genetic Algorithms Work? #32 • Two Minute Papers: AI Makes 3D Models From Photos #122 • Two Minute Papers: What is De-Aging? • Two Minute Papers: This AI Made Me Look Like Obi-Wan Kenobi! • Two Minute Papers: DeepMind’s AI Learns Locomotion From Scratch | Two Minute Papers #190 Two Minute Papers: DeepMind’s WaveNet, 1000 Times Faster | Two Minute Papers #232 Two Minute Papers: This is How You Hack A Neural Network Two Minute Papers: We Can All Be Video Game Characters With This AI ★★★★★ Two Minute Papers: DeepMind’s New AI Helps Detecting Breast Cancer • Two Minute Papers: Artistic Style Transfer For Videos #68 • Two Minute Papers: OpenAI’s Whisper Learned 680,000 Hours Of Speech! Two Minute Papers: Ubisoft’s New AI: Breathing Life Into Games! Two Minute Papers: How To Get Started With Machine Learning? #51 Two Minute Papers: Google’s New AI: Fly INTO Photos! Two Minute Papers: NVIDIA’s AI Removes Objects From Your Photos | Two Minute Papers #255 Two Minute Papers: Stable Diffusion Is Getting Outrageously Good! Two Minute Papers: OpenAI Dall-E 2 – AI or Artist? Which is Better? Two Minute Papers: Google’s New AI Learns Table Tennis! Two Minute Papers: NVIDIA’s New AI: Video Game Graphics, Now 60x Smaller! Two Minute Papers: New AI Makes Amazing DeepFakes In a Blink of an Eye! Two Minute Papers: How Does Deep Learning Work? #24 •

Two Minute Papers: OpenAI Dall-E 2 – AI or Artist? Which is Better?

Dear Fellow Scholars, this is Two Minute Papers with Dr. Karo Zonai-Fehir. Today we are going to see how open AI’s Dolly2AI’s images stack up to real works from real artists. This is going to be an amazing journey. So, hold on tight to your papers. Dolly2 is an AI that took the world by storm in the last few months. What does it do? Well, we can write it, attack description, and hand it to the AI that creates an amazing image of exactly that. And now, I am lucky to say that yes, I have ascended. Thanks to the generosity of open AI, I now have access to Dolly2. And my first thought was that if it can come up with high quality and truly novel images, let’s put those to the real test. In other words, see how they stand up against the works of real artists. Our first subject will be my favorite artist in the world, and that would be my wife Felizia Zonai-Fehir. She drew this chimpanzee. Oh my, that is beautiful. So, let’s try this against Dolly2. Okay, but how? Well, the first way of trying it will be through a variant generation. What is that? Well, since this AI is trained to understand the connection between text and images, the understanding goes both ways. This means that we can not only input text and get an image, but input an image too. Does it understand what is depicted in this image? Let’s see. Oh yes. Wow, it not only understands what is going on, but it can demonstrate that by creating really good variants of it. But not nearly as good as Felizia, I have to say. Let’s give it another chance. Here, I used a text prompt to get a hand-run chimp looking to the right. And I asked for lots and lots of detail. I have pleaded and pleaded for extra wrinkles to squeeze everything out of the AI, even cherry-pick the results, and these were the best I could get. You know, art is highly subjective. That’s what makes it what it is. But for now, I have to say, team humanity, team AI, one zero. And at this point, I thought this is going to be easy. Easy, easy, easy. And boy, you’ll find out in a moment how wrong I was. Here is a work from one of my other favorite artists, Yudit Shomujwari. Wow, that is one of her original characters. Very imaginative. I love it. So, let’s ask for a variant of that. Does the AI understand what is going on here? You know, there’s a lot to take in here, half ears, makeup, and the… Wow! I cannot believe that. I am lucky that the variants are working out so well because writing a prompt for this would be quite a challenge. I am truly out of words. These are so good. But not as good as Yudit, in my opinion. Team humanity, team AI, two zero. That’s a good score, but it is getting harder. Now, let’s look at one of Yudit’s other works, an ostrich that is also a Formula One racer. I am loving this, especially the helmet and the goggles. I think I can whip up a good prompt for this. So, what did the AI come up with? Holy matter of papers. Is this some kind of joke? It cannot be that good. I check the images over and over again. It is that good. Wow! Now, of course, scoring this is subjective, but clearly, both are excellent. I’ll give this one a tie. Team humanity versus Team AI, three one. I told you, it is getting harder. Make sure to check out both Yudit and Felicia’s works. The links are available in the video description. And now, for the next test, please give a warm welcome to the King of all donuts. Of course, that is the great Andrew Price, 3D modeler and teacher extraordinaire. Now, here are some of the donuts I came up with using Dolly too. And some from the prompt that I ran for my daughter. Loving the Googli eyes there, by the way. So, these are some great donuts. Now, Andrew, it’s your turn. Let’s see. Wow! Open AI’s was a good one, but this. This is a Michelin star donut right there. The AI made ones are no match for Andrew’s skills. So, Team humanity versus Team AI, four one. Now, note that Dolly too often isn’t as good as a real world-class artist, but the AI can create 10,000 of these images a day. That must be worth at least a point. So, four two. Whew! We made it. Now, all this testing is great fun, and of course, highly subjective. So, don’t take it too seriously. And here comes the even more important lesson. In my opinion, this tool should not be framed at all as Team humanity versus Team AI. I believe they should be framed as Team humanity supercharged by Team AI. Here is why through six inspiring examples. Look, first, this is my second favorite prompt ever, a grumpy cat in a spaceship. That is incredible. Now, almost as soon as I posted it on Twitter, it inspired Fabian to create an animated version of it. And now, our adorable little grump is ready for takeoff. Through the ingenuity of Team Human and Team AI together. So cool. What a time to be alive. So, I hear you asking if this was my second favorite. What is the first? Well, two have a look at this Fox scientist. I cannot believe how well this came out. It has tons of personality. And it also inspired others to repaint it. Here it is. You see, I don’t see this as a competition. I see this as a collaboration. Three, it also inspired the great beardy man who wanted to see a Platypus studying the schematics for a cold fusion reactor. These are absolutely amazing. Not just the quality, but the variety of the results. My goodness. Note that he asked for a one-eyed Platypus, but this iteration of Dolly is not that good with numbers. Yet, if it could talk, it would probably apologize for it. Four, I also ran a prompt for David Brown from Boy in a Band which got real interesting, real quick. He was looking for a visual pun. What is that? Is this a pen? Is this the pun that we got a pen instead? I am not sure. But this is very cool nonetheless. And I also wonder what the AI is trying to say here. Five, James from Liner’s Tech Tips also ran a superb prompt, a time traveler insensitive to humanity’s trivialities. What a great idea. Oh my, loving the framing and the thoughtfulness of these results. And finally, six, it inspired me too. I also made this. All hail the mighty cyber frog. Loving it. And we also have a tiny chihuahua in our home and in the two-minute paper’s labs. She is Rosie, the paper dog, if you will. This is how we see her. And now we can also show how she sees herself when meeting other dogs. Oh yes, I think that is very accurate. And of course, as always, plus one. You didn’t think we would leave without looking at some of you fellow scholars holding onto your papers, did you? Well, here you go. And as you see, these are some proper papers. You fellow scholars are loving it. But now let’s have an even more explosive paper like Dolly too. Oh yes, now we’re talking. This scholar is you who knows that this paper is history in the making. So much so that I cannot stop thinking about and playing with this thing. It really keeps me up at night. This is an incredible achievement that is going to transform the world around us. And fast. So artists, AI people, what do you think? Which results did you like better? Let me know in the comments below. And also I will be on Twitter taking some requests from my followers, so make sure to follow me there and send some prompts. The best ones may appear in a future episode here. The link is in the description. If you’re looking for inexpensive cloud GPUs for AI, Lambda now offers the best prices in the world for GPU cloud compute. No commitments or negotiation required. Just sign up and launch an instance. And hold onto your papers because with Lambda GPU cloud, you can get on demand A 100 instances for $1.10 per hour versus $4.10 per hour with AWS. That’s 73% savings. Did I mention they also offer persistent storage? So join researchers at organizations like Apple, MIT and Caltech in using Lambda cloud instances, workstations or servers. Make sure to go to LambdaLabs.com slash papers to sign up for one of their amazing GPU instances today. Thanks for watching and for your generous support and I’ll see you next time.

AI video(s) you might be interested in …