remove some text we're unsure of from discussion

[elray1]

"All proper scoring rules for probabilistic forecasts have an explicit link to a loss function, thereby placing numerical value on different kinds of errors in specifying forecasts."