Sampling doesn't make sense if a seed is set because each sampling will generate the exact same (potentially) problematic response. Each sampling strategy should do this for itself because there are scenarios where using a seed with sampling does make sense. Currently, the only case where this is true would be when a repair strategy is given.
We should also investigate setting "smarter" seeds, ie being able to set a single seed that then deterministically sets different seeds for future requests. This would make sampling more deterministic while still allowing for re-generating without repair.
Also investigate if there are other scenarios where this makes sense to implement.
Sampling doesn't make sense if a seed is set because each sampling will generate the exact same (potentially) problematic response. Each sampling strategy should do this for itself because there are scenarios where using a seed with sampling does make sense. Currently, the only case where this is true would be when a repair strategy is given.
We should also investigate setting "smarter" seeds, ie being able to set a single seed that then deterministically sets different seeds for future requests. This would make sampling more deterministic while still allowing for re-generating without repair.
Also investigate if there are other scenarios where this makes sense to implement.