I can't help but suspect that ChatGPT's clear bias resulted from a fair amount o... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		commandlinefan on Feb 22, 2023 \| parent \| context \| favorite \| on: Poisoning web-scale training datasets is practical I can't help but suspect that ChatGPT's clear bias resulted from a fair amount of this.

flangola7 on Feb 22, 2023 [–]

There's reams of mindless "white women only fuck black guys / white women fuck dogs" text that originated on 4chan and then spread elsewhere. I wonder how many megabytes of it ended up in GPT's dataset.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact