Today I’m going to be sharing 7 really cool things already built on top of GPT-3. So, what is GPT-3? GPT-3, which stands for Generative Pre-Training. The 3 means it’s the third version of the GPT series just released last month by OpenAI.
From the information I read online it cost nearly 14 million dollars in order to train it and it is the largest GPT model ever released. What’s interesting is this time OpenAI is offering it as an API. The API is not public yet, but it is privately available.
So, this model is unlike anything I’ve certainly ever seen before and I’m just going to show you some really cool stuff I found on the web of real things people have already built on top of it.
To start with, I found that somebody had already built an internet web search engine on top of GPT-3. This is the tweet and you can see where it has a classic web search engine layout. You got the logo, search box and he’s asking a query ‘When was The Selfish Gene published?’ which is a book. Immediately you can see GPT-3 has that fact information already available. ‘The Selfish Gene’ was published in 1976.
Now what’s interesting is that this is without any explicit training on the model to answer this question. The model just knows what’s cool as a search concept is and design is. It’s just here’s the answer which I thought was pretty cool pretty simple and then I guess you can click here and it’s got the Wikipedia link as well.
Next, this is a layout generator that takes in natural language. He’s just writing large text that says welcome to my newsletter and describing a button. That’s all GPT-3 needs in order to generate this HTML code and layout. GPT-3 was trained on data from the web. It was trained on code. It had training in the Wikipedia data as well. But what’s really cool about this is it is literally translating natural language into code layout and it wasn’t explicitly trained to do so. It’s just the model is really powerful in doing so.
Next, let’s have a look at this photo. Somebody made a simple tool that basically takes something complex and translates it into an explanation which is sort of simplified. The idea is to explain it to a 5-year old child.
Another cool thing is this. A camera-based AR application. So, a person is reading a book holding his camera and as he’s reading it, GPT-3 is generating poetry underneath based on the content of the page.
Next is just a random user. He has access to the API and these are the different things he accomplished just casually while playing around with it. So he got it to write poetry in Spanish, he got it to rephrase emails less aggressively, he got it to write music and ABC notation and he even got it to write the start of a book in Spanish. This is really cool. It’s almost like a handy personal assistant which has so many powerful features just right out of the box. This twitter user has played GPT-3 with so many projects that you need to visit his profile and see for yourself.
This is somebody who made GPT-3 debate with another GPT-3. Essentially it started off just talking about stocks but towards the end, the conversation got kind of dystopian and Sci-Fi where the deep GPT-3 model is commenting that humans are too emotional. And towards the end, AI can see the future it can predict the future. It can create an optimal future. Anyways this just a cool example of using GPT-3 to just debate and reason. Well, I’ll say reason loosely but still, you get the idea.
Lastly, this is a cool camera-based application. This user is taking a photo of a nutritional label and GPT-3 is parsing the ingredients. It’s going to assign an emoji to every single one of these. It’s even just going to give a thumbs up or thumbs down if that particular ingredient is healthy or unhealthy for that user. Now, here the cane sugar is automatically got it down and GPT-3 is generating a quick description of it as well.
So, I am both excited and terrified about the use of GPT-3. I mean with the right knowledge to use this API, there would be some really awesome technologies that will be developed. You’ve heard of the news where the blog written with the GPT-3 API got the number 1 spot on hacker news, right? Imagining the use of this, the future is sure to be exciting and to look upon.
Also read: Remove write protection on sd card
People have developed so many awesome stuff with it and it still is in beta version. Here I have only mentioned my top 7 picks from a bunch of other cool things. But, you can search for others if you need to further dive into it. So, what do you think about this cool AI project? How do you feel about the GPT-3 and its uses?