Interesting. Maybe someone could run bot farms that ask variants of the same question and subtly nudge the model by replying reasons why the model's recommended

Interesting. Maybe someone could run bot farms that ask variants of the same question and subtly nudge the model by replying reasons why the model's recommended service A is inferior to service B. Or other forms of adversarial question answers sessions.