That’s interesting. If we know that, why not align in all languages?

For instance, while the model may reject a harmful request in English, it may comply in languages that have not been ‘aligned’. In layman’s terms, the training phase where we explain the model things like ‘whenever they ask you…
Anthropic’s Breakthrough: Understanding Frontier AI
1.3K
38
Ignacio de Gregorio
Anthony (Tony/Pcunix) Lawrence 👀
·Follow
Jun 5, 2024
--
That’s interesting. If we know that, why not align in all languages?
--
--
Written by Anthony (Tony/Pcunix) Lawrence 👀4.6K Followers
·142 Following
Retired Unix Consultant. I write tech and humor mostly but sometimes other things. See my Lists if your interests are specific.
Responses (1)
Help
Status
About
Careers
Press
Blog
Privacy
Terms
Text to speech
Teams