Interesting. I wonder if model distillation affected censoring in R1.

I think we might be talking about separate things. I tested with this 32B distilled model using llama-cpp.

Just to clarify - DeepSeek censors its hosted service. Self-hosted models aren’t affected.

I miss when phones rang because lightning struck.

I’m not sure. While it’s not for me, I know some rather bright individuals that enjoy the format.

I mostly hate that it’s such an advertising industry.

Not sure this makes sense. I think the window shifts right as people continue to vote right.

From the Wikipedia article about the Overton window:

The most common misconception is that lawmakers themselves are in the business of shifting the Overton window. That is absolutely false. Lawmakers are actually in the business of detecting where the window is, and then moving to be in accordance with it.