|
|
|
|
|
|
by zvolsky
21 days ago
|
|
|
The misleading prompt cases are inadvertently testing the model's ability to filter out noise from its instructions. That could be a benchmark on its own. The correct response is to flag the inconsistency and ask for clarification. |
|
|