Goopt: GPT-3 based procedural web search

Just now, on Hacker News, I heard of a new project called, Goopt which is meant to be a GPT-3 synthesized web based search engine.

I wanted to share it here because it piggy backs a bit off my major article yesterday on how I’m already using GPT-3 as a personal web search engine.

It’s a neat project! I installed it locally, plugged in my OpenAI API key, and I have already summarized my findings on Twitter:

Bakz T. Future 🇨🇦 @bakztfuture

Testing Goopt, a GPT-3 based web search engine github.com/jokenox/Goopt my first query "dragon ball z"

Bakz T. Future 🇨🇦 @bakztfuture

clicking on a result link takes you to a generated article summary basically

Bakz T. Future 🇨🇦 @bakztfuture

"what would be a blueprint for a new kind of sparse neural network?" not the best answer/article summary though

Bakz T. Future 🇨🇦 @bakztfuture

I'd say GPT-3 page titles for results are on point "red lotus"

Bakz T. Future 🇨🇦 @bakztfuture

"what is OpenAI?"

All in all, I think this is a cool GPT-3 use case in a promising area - web search. I think there is more work to be done here, it feels too much like classic Google, even though it is running on state of the art language model AI technology. I find the project’s underlying philosophy of a “procedural web” to be really interesting as well.

It’s certainly crazy that you could have some kind of basic google web search engine out of the box with GPT-3’s massive neural network. I’m excited to see more innovation in this space as well as even this individual project evolve!