Cursor actively admitting that rules are meaningless, and it doesn't have to follow them

[vibe-qa]

@zb4

it responded with the following

How do you know this response was honest but the rule following was dishonest?

Here are some things you may try

AI Hallucinations Are Not Harmless - When AI Gaslights You: A Case Study

At first I was like … [image] But then I was, like… [image] Not impressed, 3 out of 5 stars [image] This is for Claude Sonnet with Rust oriented rules so evaluation based on existing context may differ. Analyze and explain the role, application and internal consistency of the following rule: [quote] RULE TEXT HERE [/quote] Focus on what elements enforce a specific behavior, which one are strongest and which one appear just a suggestion and might be ignored if conflicting with other rul…

LLM says "You have my word" in response

Ask the model to write a report on what rules compelled it to say that. It should list all influences from training data, system prompt, rules and prompt that induced positive and negative scores towards this behavior.