[SrslyJosh]

Where do you think the training data for AI judges would come from?

[vokep]

evaluations of historical judgements made and what the consequences were.

How to define good/bad consequences becomes the problem then, but thats the nature of it.

[sigfubar]

Such evaluations are useless as a training set because they inherit the biases of whoever wrote the evaluations. Of course, if fairness isn’t a desired property, feel free to carry on.

[103e]

It's AI judges, all the way down.