Re:Astra can read and write weights to improve Qwen
by rumblefrog on Monday November 02, 2026 @02:26PM
(#50342257)
I wonder if the same technique can be applied to model abliteration? It would easily make any open weight models uncensored.
[ Parent
| on HN ]
Re:Astra can read and write weights to improve Qwen
by senderista on Monday November 02, 2026 @03:08PM
(#50342929)
It is already the same technique behind abliteration: identifying the direction associated with an observed behavior (refusal, for abliteration) and projecting it out.