logoalt Hacker News

nottorpyesterday at 8:49 PM1 replyview on HN

> it apparently gets smarter once you uncensor it

Interesting, that has always been my intuition.


Replies

cluckindanyesterday at 10:28 PM

It makes sense. Guardrails and all other system-provided context tokens force activation of weights that would not otherwise activate. It’s just like telling a human not to think of a pink elephant and just provide numbers from the Fibonacci series or whatever.