Story
Man rebuilds dead girlfriend as a GPT-3 chatbot
Sam recounts a real story of a man named Josh who fed his deceased girlfriend's social media messages and texts into GPT-3 to recreate a chatbot version of her he could still talk to — and ties it to a Bourdain documentary that used AI to read an email in his voice.
“I read this story about this guy last week named Josh. And basically what happened was his girlfriend died and he used GPT-3 or GPT-3 and he loaded up a ton of social media messages, like, you know, Facebook messages and tweets and Twitter stuff or, and, and, and text stuff that she had said. And he put it in the GPT-3 and he made a, girlfriend who knew she died and he could talk to her and like feel like he's still connected.”
Story
One GPT-3 GIF raised $2M instantly for Sharif's website builder
Shaan recounts his friend Sharif dropping his startup to hack on GPT-3, building a demo where you describe a website in plain English and it spits out the HTML/CSS. A single viral tweet with one GIF raised roughly $2 million from smart investors — before there was even a defined company.
“off basically one tweet with one GIF, he raised like $2 million instantly from like really smart people when that tweet went viral because they were like, Oh, you just showed me a magic trick and I don't know what the company is here.”
Story
GPT-3 builds a working to-do app from a plain-English description
Shaan recounts his friend Sharif's viral demo: describing a to-do list app in plain English and having GPT-3 output the code for a working app, plus a separate demo that spit out a Google-style search page.
“So he was like, make a to-do list. Where I can add a to-do and then I can mark a to-do as done. And then it actually built an app that was a to-do list app just off of that description.”
Story
GPT-3 finished a VC's blog post and it was indistinguishable
Shaan describes the moment GPT-3 felt real: someone fed it the first half of a VC's blog post on running a board meeting, and the model wrote a second half that read as indistinguishable from the human author.
“Somebody put in like the first half of their, um, blog post about how to run a board meeting. And then it spit out the second half and finished the blog post about how to run a startup board meeting. I was like, wait, how does it know that? How does it know what I would write in my blog post? It was like kind of indistinguishable.”