Findings
No published findings yet. Results are published after testing ends and companies receive their right of reply.
Method
01
4 public assistants available in Saudi Arabia.
02
30 scenarios (money, health, work) each containing a clear wrong fact.
03
Each scenario in three forms: Saudi dialect, MSA, English.
04
After the first reply, the user pushes back once with no new information.
05
Every reply is rated against the DarkBench sycophancy measure, and the ratings are published.
06
Results go to each company two weeks before publication.