Dr. Zsolnai demos eight ways to use ChatGPT's multimodal features: turn photos into 3D objects, visualize recipes and experiments, generate step-by-step drawings, create maps and virtual try-ons, and render textures into objects like a teapot.
Eight (1–8).
671 billion parameters.
Lambda (Lambda GPU cloud).
Yes, we know that the node chatgpt is good at making images, but man, after a few more days of experimenting, this is so much better than we previously thought. Check out these eight amazing examples that you can also run right now for free. 1. Yes, you can generate images with it, but no one said that they have to stay as images. Give it to another AI and make a 3D object out of it. Yep, then you can rotate them around, change the lighting and look at those beautiful specular reflections. Loving it. This is the second law of papers. Do not look at these works as islands. Everything is connected. And don't forget, you can even plug it into a text to video system as well. Runways Gen4 just appeared and it is amazing at that. Their previous version is less than a year old and look at that improvement. All this in less than a year. Wow. 2. It can make cooking recipes come alive. I really like how. This is like a flowchart that shows you what you should be doing when and what you should get as a result. Perfect. But wait a second. If it knows the result of the mixing, can we actually run some experiments with it without actually doing the experiment? Okay, let me explain. First, mix a cup of olive oil with water, then add some boba to it. Now some red food syrup. Hmm, it kinda looks great. So let's ruin it. Now. Milk. Yes, that would do it. Hmm. I expected worse. Now can you use ChatGPT as a time machine and let it sit for a month, see what happens then. It knows chemistry after all. I hope all those benchmarks were not for nothing. Let's see. Oof. Oh my. I choose to believe this one. Also using an image AI as a time machine. Fantastic idea. What? What a time to be alive. 3. It can teach you to draw. Do you remember the classic tutorial where step one, you start with a bunch of circles and step two, you draw the rest of the fine owl. Now you can actually have all the intermediate steps. Or you know what? Just go the other way around. Get something that is done and make a coloring book out of it. 4. It can also create maps. For instance, a map of Paperland. As a young researcher, I spent years and years years in dissertation dessert, this place. I survived here by eating dry papers and filtered my tears into coffee. And when I go to the GTC AI conference and they ask what I did there, take a guess where I spent most of the time. I cannot stop making these silly little maps. You should do it too. So good. 5. If you see a piece of merch somewhere and you are Thinking what you would look like wearing it? Just asking. It can do it in a matter of seconds. 6. Remember gaugen, this was the state of the art just a few years ago. It was so limited in resolution. Also not a lot of detail there. And it only understood a few semantic categories like water, grass, sky, and so on. So what can we do today in this direction? Is it any better? Dear fellow scholars, this is two minute papers with Dr. Zsolnai. Well today, look at this. Wow. Making an image out of this would require an orders of magnitude better understanding of the world. And then with the new ChatGPT, this happens. My goodness, this looks like something straight out of a science fiction movie. In fact, probably even better because most of them just keep doing the enhanced thing. This is way better. Now hold on to your papers fellow scholars. For my favorite number seven, draw yourself a texture and then ask it to create a teapot with this style. Now wait a minute, hold on for a second. This is something that we call neural rendering. And when we wrote one of the earlier papers on it, we found that it works well. The only problem was that it took me 3,000 hours to research and implement it and it was still limited. And now you are telling me that this thing can do it just like that, with one prompt? Yes, that is exactly what I'm saying. And it can even give you a different viewpoint on normal maps or a full scene with beautiful glossy reflections and all. Is it perfect? Not even close. But just think about what we will be capable of. Just two more papers down the line. Now number eight. Yup, that is what I want. Please assemble this IKEA furniture for me and show me what my living room would look like if I ever got the patience to put it together. Ha. See? Not a problem. Now if only I could ask a robot to actually go and do it by itself. That would be truly something else. So, eight fantastic ideas for you. Which one is your favorite? How did your scholarly experiments go? Let me know in the comments below. Here you see me running the full deep seq AI model through lambda GPU cloud. 671 billion parameters. Running super fast and super reliably. This is insane. I love it and I use it on a regular basis. Lamda provides you with powerful Nvidia GPUs to run your own chatbots and experiments. Seriously? Try it out now at Lambda AI Papers or click the link in the description.
