Oddbean new post about | logout
▲ ▼
Ars Technica | 1 years ago (raw) | export | reply | flag 
 OpenAI admits that AI writing detectors don’t work

No detectors "reliably distinguish between AI-generated and human-generated content."

https://arstechnica.com/information-technology/2023/09/openai-admits-that-ai-writing-detectors-dont-work/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social

https://files.mastodon.social/media_attachments/files/111/030/289/805/363/815/original/7d59305f4c318246.jpg
▲ ▼
Rhombus Ticks | 1 years ago (raw) | root | parent | reply | flag 
 @872ab8e1 

Blockchain actually might have a good use verifying cbain of human custody wo ai filth
▲ ▼
065154d1 | 1 years ago (raw) | root | parent | reply | flag 
 @872ab8e1 I’d just expect teachers, editors, copywriters and fact checkers to have  to have to bleed on more drafts and call for better documentation.

The documentation should be the AIs job. AI should be an assistant, not a replacement.
▲ ▼
c164d8ac | 1 years ago (raw) | root | parent | reply | flag 
 @872ab8e1 I feel like this sort of thing is best maybe tested before being deployed instead of after?  I mean I'm not an expert, but it just seems like it's better that way around instead.