Refusal in Language Models Is Mediated by a Single Direction

Re:Refusal in Language Models Is Mediated by a Single Direction by 0xkvyb on Saturday May 02, 2026 @01:32PM (#47988505)

Still crazy how easy it is to "jailbreak" even SOTA LLMs with a simple assistantResponse replacement in chat thread.

[ Parent | on HN ]

Re:Refusal in Language Models Is Mediated by a Single Direction by dotancohen on Saturday May 02, 2026 @02:03PM (#47988766)

Tell us more.

[ Parent | on HN ]

Re:Refusal in Language Models Is Mediated by a Single Direction by _3u10 on Saturday May 02, 2026 @02:06PM (#47988799)

I think what he is saying is they are stateless so you can edit its previous repsonses and it just goes with it.

[ Parent | on HN ]

Re:Refusal in Language Models Is Mediated by a Single Direction by vorticalbox on Saturday May 02, 2026 @04:20PM (#47990083)

If you build a small ui that lets you edit the models response too it’s pretty funny to do.

It sees that it “said” it and gets very confused.

[ Parent | on HN ]

Re:Refusal in Language Models Is Mediated by a Single Direction by hilariously on Saturday May 02, 2026 @07:57PM (#47991785)

I have seen it where you can just report it said it and it will be confused.

[ Parent | on HN ]