14 Cool Apps Built on OpenAI's GPT-3 API

what up guys today i'm going to be talking about gpt 3 uh which stands for general purpose transformer it's the third version of the gp uh t series just released last month by openai [Music] from what i read online it cost close to 14 million dollars in order to train it and it is the largest gpt model ever released what's interesting is this time open ai is offering it as an api that api is not public yet but it is privately available and so today i'm going to be sharing 14 really cool things already built on top of gpt-3 this model is unlike anything i've certainly ever seen before and so today 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 so to start with i found that somebody had already built an internet web search engine on top of gpt3 so let's take a look here this is the tweet let me let me load the video and so you can see here it has a classic web search engine layout you got the logo search box search and he's asking a query when was the selfish gene published which is a book [Music] and immediately you can see gpt3 has that fact information already available you can see it the self machine was published in 1976 now what's interesting is this is without any explicit training on the model to answer this question the model just knows and what's cool as a search concept is and design is like there's no page of results you know 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 it's got the wikipedia link as well i'm not sure how that works if that depends on the model or where it's getting that but still really cool next i'm going to be showing you this is a layout generator which takes in natural language so let's pause here take a look he's just writing large text that says welcome to my newsletter and describing a button and that's all gpt3 needs in order to generate this html code and layout so gpt3 was trained on data from the web it was trained on code it was trained i believe on the wikipedia data as well but what's really cool about this is it's literally translating natural language into layout code layout and it wasn't explicitly trained to do so it's just the model is really powerful next we'll be taking a look at this just a photo but somebody made a simple um tool which basically takes something complex and translates it into an explanation which is sort of simplified like almost like how you would explain to a five-year-old that's the idea and again i imagine when you have access to the api you're simply telling it to simplify this text just in natural language you're not doing any crazy coding to manipulate it to get it to simplify the text it just throughout the process of training it it can just figure out all these different things because it observes so many patterns next this is a cool uh you know camera based ar application so the camera is you know this person is reading a book and as he's reading it gpt3 is generating poetry underneath based on the content of the page uh next you see this is just a random user he has access to the api and these are the different things he 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 so this is really cool it's almost like it's like a handy personal assistant which has so many powerful features just right out of the box uh next so this is somebody who made um gpt three debate with another gp t3 essentially it started off just talking about stocks but towards the end i mean the conversation got kind of uh you know dystopian and sci-fi where the deep gbt3 model is commenting that humans are too emotional and towards the end and ai can see the future it can predict the future it can create an optimal future anyways this is a cool just example of using gpt3 to just debate and reason well i i'll say reason loosely but still you get the idea um another example is this person built a um [Music] this is a tool which could which takes in natural language description of a math formula and then it generates the corresponding latex that goes with it [Music] you see here it generated the latex formula or equation based on a natural language description again gb23 was not explicitly trained to do this but throughout the training process clearly it's picked up all of these really powerful patterns which allows it to do so many interesting things uh next this person is building something which translates natural language into i guess bash commands in linux and so you see here he's just explicitly telling it create me a file file that contains three colors [Music] list the files in the directory um again clearly it was you know tracking the code and reading it as it was trained and now it can it probably picked up some batch commands along the way and now it can do these kinds of interesting things next we're going to be talking about a figma plug-in somebody built it basically takes in natural language and generates a figma design this video is too long but basically he enters a description of what is basically instagram and gpt3 within figma is able to generate it so the full description is an app that has a navigation bar with a camera icon photos title and a message icon a feed of photos with each photo having a user icon a photo a heart icon and a chat bubble icon and then we can see this is what it creates all from one natural language description it's able to generate this next this is somebody who asked gpt3 to basically generate a religious doctrine based on some of the some of the ways that gp3 works like how it's trained like how a transformer is trained which is basically based on predicting the next word in the sequence at a huge scale and so these are some of the rules within this religious doctrine so words are things correctness is the beginning of sanctity to achieve it is to be rewarded i mean these are like you know pretty uh philosophical statements um and you'll also notice like this is automatically generated by gpt3 it's a pretty impressive generation uh it seems plausible it seems you know to follow the rules of english and punctuation and grammar um this could probably pass is something made by a human it even as a i would even say this fits the tone of a religious doctrine as well for for some kind of you know secret crazy cult or something like that uh very impressive uh next somebody turned uh basically it's like a reverse sql to natural language tool so it basically you give it an sql statement and it will generate it into a natural language kind of question very cool i can imagine gp gpt3 being used a lot in in dashboards and stuff just to in the future just to quickly uh query data and get back simple responses about the data next somebody was able to build a functioning react app by just simply describing it to gpt 3. a button that says add three dollars and a button that says withdraw five dollars then show me my balance so then he clicks generate and you can see here it's not only generated the layout but it's able to actually do the math and gives the final output as well so again he just gave it a description and it's functionally working and the design wise it's working as well it generated it entirely in react for him instantly wow [Music] uh next he was able to even get it to generate the google home page so take a look at this he's simply writing the google logo he's describing a search box that looks incorrectly spelt i think gpt3 is going to be fine with that too some buttons two buttons wow so goo it was able to grab the google logo it was able to create the search box and the two buttons and now it looks like he's even adjusting the padding in between the buttons in natural language and you can see the outline there's the two buttons right there really impressive stuff uh lastly this is a cool uh camera-based application so this user is taking a photo of a nutritional label and gpt3 is going to parse the ingredients [Music] uh it's gonna assign an emoji i guess to every single one of these or sorry every single one of these these are the ingredients down here and i guess it's even just going to give a thumbs up or thumbs down if that particular ingredient is healthy or unhealthy now you can see here the cane sugar is automatically got it down and gpt3 is generating a quick description of it as well so altogether that's pretty much all 14 that i had to share the key idea is again that gpt3 is the largest model ever built it's really impressive and one of the most impressive things about it is it's just observed so many patterns and has made so many predictions throughout the training process that it's able to do so many things out of the box and what's also interesting is is clearly you're able to just give it natural language very casually almost the way you would speak to a person and it's able to go off and do stuff with what you've entered things that it wasn't explicitly trained to do like generating summaries like generating layouts grabbing logos these are all just the uh examples of how powerful the model is and i'm really excited these are 14 things that i just found with the limited api access that's available it's going to be crazy to see where gpt3 will go when the api becomes publicly available and maybe even commercially available as well right so imagine various companies also using it as well either in their back end code front end code for various parts of their application it's going to be really exciting so uh i'm i'm really excited to check out the stuff with gpt3 if you like this video please make sure you like and subscribe uh and also my my email is just in the description so if you're working on something pretty cool with gpt3 uh and you'd love to share or you think it'd be uh it's it's pretty cool then send it my way and i'd be happy to talk about it as well uh and i'm also personally looking forward to trying to get uh gpt3 access as well so if i work on anything with gpt3 i'm also going to be sharing it here so make sure you subscribe so you can follow along and you too will probably just be blown away with how powerful uh gpt3 is and all the different possibilities anyways thanks so much for watching um [Music] [Music] [Music] [Music] [Music] [Music] you

Channel: Bakz T. Future

Views: 226,377

Keywords: gpt-3, openai, transformer

Id: G6Z_S6hs29s

Length: 14min 55sec (895 seconds)

Published: Mon Jul 20 2020