Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation — mentioned by throwaway219450

Sat, 05 Sep 2026 16:55:42 GMT

There's some cool research that looks at how strongly the weights are aligned through training vs adherence to the system prompt. Like when you know a model is lying through censorship: https://arxiv.org/html/2603.05494v2 Presumably if negative guidance is in the system prompt, there's a good chance that the model would happily comply if it wasn't there.

Paper and discussion history. A mention is not an endorsement.

Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation · HackerPapers discussions

Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation — mentioned by throwaway219450