3 ms·Because they are doing it to compute quality metrics not to implement RLHF. It’s not training data.by _2d30 1y agoBecause they are doing it to compute quality metrics not to implement RLHF. It’s not training data.visarga 1y agoEvery decision they take based on evals influences the model._2d30 1y ago/"directly"/