At least 50% of code on github will be generated by language models or their successors on January 1 2030.
Created by amrav on 2020-06-13; known on 2030-01-01
- amrav estimated 70% on 2020-06-13
- Baeboo estimated 30% on 2020-06-13
- amrav changed the deadline from “on 2030-01-01” and changed their prediction from “At least 50% of code on github will be generated by language models or their successors.” on 2020-06-13
- EloiseRosen said “as in new commits, or 50% of all code including pre-existing? ” on 2020-06-14
- EloiseRosen said “Also for a clear resolution you should specify how this would be measured. Lines of code? # of commits? # of repos?” on 2020-06-14
- amrav said “Measured by lines of code at HEAD across all repositories regardless of when they were created.Code generated by AI and reviewed by humans counts.” on 2020-06-14
- amrav said “There’s no good way to do attribution right now for human-AI pair programming, I’m hoping this will change by 2030.” on 2020-06-14
- amrav estimated 20% and said “Revising downwards after thinking about all the ways in which this could not happen because this is so specific. Eg generated code is not checked in / lives elsewhere / instructions are interpreted on the fly, or even github goes defunct.” on 2020-06-14
- Baeboo estimated 10% on 2020-06-16
- wizzwizz4 estimated 5% and said “Updated up for a group of people deciding to auto-generate an absurd amount of code for the lolz, down for Microsoft trashing GitHub. But I think most generated code will be templated.” on 2020-06-19
- azatris estimated 1% on 2020-06-23
- MultiplyByZer0 estimated 1% on 2020-06-30
- Baeboo estimated 8% on 2020-06-30