Heidelberg AI News

Article · heise online ·

heise+ | Abliteration erklärt: Wo KI ihr Gewissen hat und wie man es ausradiert

Wer weiß, wo das „Gewissen“ eines Sprachmodells sitzt, kann es mit mathematischen Operationen lahmlegen. Wir erläutern die Grundlagen.

Read at heise.de The opening of the article at heise.de.

In this story

A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.

Related stories

All stories

VideoDavid Ondrej

Abliterated Models, clearly explained (and how to run them)

The video explains how “abliterated” or uncensored open-weight models remove refusal behavior while retaining most of their knowledge and capability. It compares jailbreaks, fine-tuning, and weight editing, then covers practical ways to run these models locally, through APIs, or on rented GPUs for private development and defensive security testing.