Change Criteria Score Scale

[kargibora]

I found one research suggesting that 0-5 Likert scale for LLM-as-a-Judge framework is the one that is aligned with humans most. We can simply change our scale range based on that.

https://arxiv.org/pdf/2601.03444

[geoalgo]

Could we do an experiment to validate this? For instance making sure that one metric such as MAE-ELO is improved when doing that?

[kargibora]

Sounds good to me, I will try to use same samples with different range as a confirmation.