GPT 3 Demo and Explanation - An AI revolution from OpenAI

demo of some of the technologies and some of the things that GPT 3 is able to do and some of the demos circulating the internet in just the first few days and how this technology works and what it means for the future of technology, software and how we use the internet going forward. So GPT-3 comes from a company called OpenAI. So OpenAI was founded by Elon Musk and Sam Altman. Elon Musk is a prolific entrepreneur you maybe familiar with him from SpaceX and Tesla, and Sam Altman was one of the founders of Y Combinator, a very famous startup accelerator. Combined they invested over a billion dollars in OpenAI to advance the state of the art of artificial intelligence and make sure that artificial intelligence was being used for the betterment of mankind. So let's jump in and take a look at what the tech can do. So the first example is an application called Figma and the developer has built a plug-in to Figma that allows users to enter a description of what they want for an application. So in this example, a user is describing an application with a feed and some icons and a user interface and when he hit submit the application actually generates an application that looks very much like Instagram. it has all of the buttons all the UI in a scrolling feed. Just astounding in terms of what it's able to do. The next example is a coding example and so this is Debuild.co and they're describing an application that is a basic to-do list application. And so what the software is doing is it's generating a React application. All of the functions and procedures, all the things that are necessary to create a React component to handle to-dos. And you can see it works. You can add items to the to-do list there's all the events and triggers for handling the interactions of the application and it's all generated on the fly by artificial intelligence. Just amazing what we're able to do with AI both design and development aspects. The next example is in Excel, and so we have a table with headers of company ticker and year founded. And we ask OpenAI to complete the table and it's able to fill in the entire table with the data that it thinks we want. Again it's pulling this information from its corpus of artificial intelligence and it's trying to be predictive in terms of what it thinks we want in the table. If we don't agree or we want to train it a little bit differently we can cue it with different elements that start in the table. So here I actually want the table to be filled in starting from the newest to the oldest and so I put twitter and facebook, and then ask it to predict the table. And it's able to understand what I'm trying to do, and again goes top to bottom all the way in, filling in the year founded information with all the company information. So this example is within Wikipedia. So if you've ever seen a long article you can't really understand or find the information that you need, you can simply have a plugin where you ask the exact question that you want. So in this example, I'm asking why is bread fluffy and it's searching the data and information in that article and generating an answer to my question and giving me a citation and location that i can jump to. So really summarizing the data making it really easy for me to find exactly what i want from large volumes of text and information that are presented. Because of the vast amounts of information within GPT 3 you can really ask it anything. In fact you can almost use it like a search engine, where you ask it a question it can give you one result giving you the citation of exactly where to find that answer. So Google, and Apple and Amazon have thousands of engineers working on this problem and we're able to use this artificial intelligence to solve a very similar problem almost instantly. Just amazing what you can do with the tech. One of the common applications of GPT-3 is translation. So here we can give it an example of an English sentence that we want translated. "Your order will arrive in seven to ten business days" and ask it to translate into French. We can also use translation to translate things that are written in legalese. This is an NDA, a non-disclosure agreement and translate that into something that a second grader would understand. Really useful. So this is the paper from OpenAI it's called language models for few-shot learners. Now the idea behind the technology is that it can use the models in the text to train things on. Zero shot, one shot, and few-shot models. And so the example of a zero shot might be to translate English to French and you give it a prompt like the word cheese and then it translates that word. As the model is being fine-tuned it gets better and better at zero shot. The more examples of text you give it the better it can get. So here's an example of one shot where you give it one example, and here's an example of a few shot model where you give it a couple examples of what you're looking for. And it's able to do the translation for you. As you train more, the artificial intelligence is able to predict what you're looking for in terms of an answer, in terms of more and more models. As you train the model for more and more days the software starts to converge on what the answers are. And so this is the petaflop days as it's computing. So here's an example of one of the recent demos where people can speak in plain english and and the artificial intelligence is able to translate that into command line tools that are needed to run a linux operating system. So instead of an esoteric command, you can just say things like "list files recursively" and it knows exactly what to do or "list files in a human readable format" and it's able to return those results as well. There's lots of other applications for the technology you can use the artificial intelligence to generate quizzes based on the information that you have. So it can generate vocabulary or questions and then show you the context of where it's getting the information for the quiz that it's generating. It can also be used for entertainment such as role-playing and generating a story with multiple endings or multiple directions in which you can go. A really early exploration of what the technology is capable of but gets your imagination going in terms of what's possible. OpenAI has been developing its technology for a number of years and one of the very first papers was on generative pre-training. The whole idea of generative pre-training is that most artificial intelligence is trained on labeled data. And so you have a large collection of data and the data is manually labeled so that the artificial intelligence knows exactly that the data is correct, and what it means. And what genitive pre-training means is that they realized that there's a ton of data out there that hasn't been labeled. And in particular data like the entire internet or Wikipedia or books. And if they could train the artificial intelligence to essentially label itself, and learn from that existing data you could really accelerate the pace of artificial intelligence. The first generative pre-training model used about 7000 books in its collection and the newest generation the gpt-3 uses a much larger collection of data. How much more? Over 410 billion parameters taken from the web 67 billion parameters taken from books, 175 billion parameters in total and much, much, more. That's the sheer mountain of data that's being used to train this artificial intelligence. Because the data model is pre-trained the model gets better and better the longer you run it, and the openai team ran it for over a thousand petaflop days. So the data was trained on over an exaflop of computing power. And just to imagine how much computing power that is... One second of exaflop computing power is equal to 37 trillion years of adding numbers together. So it's an an absolutely staggering amount of computer horsepower. I'm really excited about the future of artificial intelligence i think that technologies like GPT3 and future iterations of this will change how we interact with computers. This technology makes things like Amazon Alexa, and Google Assistant and Siri look like real child's toys, because of the sophistication of what this technology can do. i'm incredibly excited by artificial intelligence i think this is going to be transformative. i hope you enjoy this type of content I'm Greg Raiz. I talk about technology, entrepreneurship, and design and i hope i'll see you on the next one.

Channel: Greg Raiz

Views: 256,414

Rating: 4.9241705 out of 5

Keywords: entrepreneur, startups, design, ux, raiz, greg raiz, GTP 3, GPT 3 Demo, GPT3 OpenAI, GPT3 Demos, GPT3, gpt 3 openai, gpt 3 paper, gpt 3 demo, computerphile, gpt 3 youtube, gpt3 design, gpt3 react, gpt 3 react, gpt3 excel, gpt3 demo, gpt3 ai demo, gpt 3 ai demo, GTP 3 AI, AI, Language Models, Open AI, Machine Learning gpt3, gtp3, gtp3 demo, gpt3

Id: 8psgEDhT1MM

Length: 8min 46sec (526 seconds)

Published: Mon Jul 20 2020