If you forced then to say the names of the doctors i’m pretty sure “Dr. Patel” would be overrepresented (or represented just as much as in the training data).
LLMs are not replacements for human beings, if you don’t want the most bland output, you either train your own with a dataset to fit your needs, or specifically tell the LLMs in the prompt (which sometimes they’ll just ignore) or pay a human to draw it.
The most common surname for doctors in the US is Patel. That doesn’t mean you should assume a doctor is named Patel.
No, but it does mean that if I am suddenly called upon to invent a doctor “Patel” is a reasonable name to throw out there.
Sure, but so is Johnson (2nd Most common). Just like it’s reasonable to invent a male or female doctor.
LLMs are statistics machines though.
If you forced then to say the names of the doctors i’m pretty sure “Dr. Patel” would be overrepresented (or represented just as much as in the training data).
LLMs are not replacements for human beings, if you don’t want the most bland output, you either train your own with a dataset to fit your needs, or specifically tell the LLMs in the prompt (which sometimes they’ll just ignore) or pay a human to draw it.