"This model was produced using Hirundo's behavioral-unlearning approach to reduce censorship and geopolitical bias-related behavior while preserving non-target behavior."
I guess I never actually just chat with my Qwen models so I've never noticed any censorship or biased behavior. We just talk about the work, but this sounds like a good idea for other types of uses.
"This model was produced using Hirundo's behavioral-unlearning approach to reduce censorship and geopolitical bias-related behavior while preserving non-target behavior."
I guess I never actually just chat with my Qwen models so I've never noticed any censorship or biased behavior. We just talk about the work, but this sounds like a good idea for other types of uses.
i guess the point is, what's going on beneath the surface, and how is the alignment of the model potentially steering you without you realizing
[dead]
Can you do the same for frontier Chinese models?
I mean Orinth-1.5 35B has been out for a while and is considered a better model in that space. Not sure why they didn't start from there.
[flagged]