Dr. Zsolnai walks through OpenAI's GPT-3 editing feature, which can insert or rewrite text anywhere in a document. He shows it rewriting stories across different settings, completing recipes, rewriting poetry, and improving code by compressing it or fixing runtime complexity.
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to have an AI write a story for us and even do our coding exam materials. And the coolest thing is that both of these will be done by the same AI. How? Well, OpenAI's GPT3 technique is capable of all kinds of wizardry as long as it involves text processing, for instance, finishing our sentences, or creating plots, spreadsheets, mathematical formulae and many other things. While their Dall E2AI is capable of generating incredible quality images from our written descriptions, even if they are too specific. Way too specific. Now, today we are going to play with GPT3, the text AI that just gained an amazing new capability and that is editing and inserting text. And if you think that this doesn't sound like much, well, check out these examples. Let's start a story. Today's the big day, it Sundays. Now the AI sees that the name of the section is high school graduation. And it infers quite intelligently that this might be a message from a school to their graduating students. So far, so good. But this was possible before too. So what's new here? Well, hold on to your papers. And now let's do this. Oh yes, change the story. Now, we don't know why this is the big day, but we know that the next section will be about moving to San Francisco. And oh my, look at that. It understood the whole trick and rewrote the story accordingly. But you know what if we wish to move to Istanbul instead? Not a problem, the AI gets that one too. Or if we have grown tired of the city life, we can move to a farm too and have the AI write out a story for that. Ok, that was fantastic. But when it comes to completion, I have an idea. Listen, how about pretending that we know how to make hot chocolate? Fill in the last step, which is, you know, not a great contribution, and let the AI do the rest of the dirty work. And can it do that? Wow. It knows exactly that this is a recipe and it knows how to fill that in correctly. I love it. The put hot chocolate part is a bit of a cop out, but you know what? I'll take it from the recipe. It sounds like a good one. Cheers. Now let's write a poem about GPT3. There we go. It rhymes to checkmark. And now let's ask the AI to rephrase it in its own voice. And yes, it is able to do that while carefully keeping the rhyme intact. And when we ask it to sign it, we get a letter to Humanity. How cool is that? Onwards to coding. And this is where things get really interesting. Here is a piece of code for computing Fibonacci numbers. Now translate it into a different programming language. Wow, that was quick. Now rewrite it to do the same thing. But the code has to fit one line. And here comes my favorite. Improve the runtime complexity of the code. In other words, make it more efficient. And there we go. Now, clearly you see that this is a simple assignment. Yes. But make no mistake, these are fantastic signs that an AI is now capable of doing meaningful programming work. And if you have a friend who conducted coding interviews, show this to them. I reckon they will say that they have met plenty of human applicants who would do worse than this. If you have your own stories, you know what to do. Of course. Leave them in the comments section. I'd love to hear them. And if you have been holding on to your paper so far, now squeeze that paper because here comes the best part. We will fuse the two previous examples together. What does that mean? Well, we will use the Hot Chocolate Playbook and apply it to coding. Let's construct the start and the end of the code and hand it away to the AI. And yes, once again, it has a good idea as to what we are trying to do. It carefully looks at our variable names and fills in the key part of the code in a way that no one will be able to tell that we haven't been doing our work properly. What a time to be alive. So an AI that can write a story for us, even in a way that we only have to give it the scaffoldings of the story, and it fills in the rest. I love it. So cool. And it can help us with our coding problems and even help us save a ton of time massaging our data into different forms. What an incredible step forward in democratizing these amazing AIs. So what would you use this for? What do you expect to happen? A couple more papers down the line? Please let me know in the comments below. I'd love to hear your thoughts. This episode has been supported by Lambda GPU Cloud. If you're looking for inexpensive cloud GPUs for AI, check out Lambda GPU Cloud. Get this. They've recently launched an Nvidia RTX A6000 with 48 gigabytes of memory. And hold on to your papers because Lambda GPU Cloud can cost less than half of AWS and Azure. Join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances. Workstations or servers make sure to go to lambdalabs.compapers to sign up for one of their amazing GPU instances today. Our thanks to Lamda for their long standing support and for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
