No it doesn't make sense you idiot. Random seeds aren't attached to any individual prompts and if by "trying multiple seeds" you mean setting some exact value, then you destroy any randomness in predictions and it will not lead to actual accurate results. Keep posting though, retard, because it makes it obvious you truly don't understand what the fuck you are talking about.
Seeds aren't attached to prompts, and I never made such a claim. They're attached to the output each time you make a query. The random seed behind the scene is the reason why identical queries at different times produce different responses.
They aren't attached to outputs. What the fuck are you talking about you retarded chimp? And if they are, how is taking specific values of these seeds going to guarantee increase in model accuracy and not just give you exact same outputs? These are probabilistic models, how is removing randomness doing anything other than destroying the models predictive power? Holy shit you are amazingly stupid.
Because the anomalous outputs are rare (this thing can pass the MBA remember). Instead of producing one output per prompt, behind the scenes they would produce multiple outputs per prompt. Then any prompts with outputs that didn't match the average output of the whole would have that anomalous output filtered out.
Love you
