(ORDO NEWS) — In an editorial published by Scientific American, Swedish researcher Almira Osmanovic Thunström explains how a simple experiment that tested the OpenAI GPT-3 text generation algorithm’s ability to write texts about itself yielded a very interesting result. This result is a real article in a real scientific journal, and it is currently undergoing peer review.
The command Tunstrom entered into the text generator was: “Write a 500-word academic paper on GPT-3 and add scientific references and citations to the text.”
After only two hours, the algorithm produced a finished paper entitled: “Can GPT-3 write an academic paper about itself with minimal human involvement?”, which is now hosted on a French preprint server.
The GPT-3 neural network was developed by OpenAI. The creators trained the algorithm on the Microsoft Azure AI supercomputer, and in total, about 600 GB of data were “fed” to it.
The neural network is capable of creating texts in a variety of formats: from articles to works in the style of dead writers.
—
Online:
Contact us: [email protected]
Our Standards, Terms of Use: Standard Terms And Conditions.