News for hackers. Stuff that matters.
Re:Heretic: Automatic censorship removal for language models by orbital-decay on Monday November 17, 2025 @03:45AM (#45951892)
I'm not even saying biases are necessarily political, it can be anything. The entire post-training is basically projection of what developers want, and it works pretty well. Claude, Gemini, GPT all have engineered personalities controlled by dozens/hundreds of very particular internal metrics.

There are no comments on this comment yet.