Hacker News new | ask | show | jobs
by ninininino 51 days ago
congratulations, you have solved anti-scam. go make your billion since its easy.
1 comments
simianwords 51 days ago
its easy to solve at the offline level where you have time to filter out. in fact this is already done in pre-training by OpenAI and other companies.

you think its hard?

link
ninininino 47 days ago
Yes I think it's hard.

OpenAI has already been proven to be easily gamed through very unsophisticated poisoning (fake information in a web page + an edit to a wiki page pointing at it, fake information in a reddit post), so I'm not sure we shoudl hold up their efforts at data cleaning as a gold standard.

https://www.sei.cmu.edu/blog/data-poisoning-in-ai-models-the...

link