[planetpluta]

> Essentially we needed to teach LLMs (which are paid by the token) to only generate a small number of high quality comments.

The solution of filtering after the comment is generated doesn’t seem to address the “paid by the token” piece.

[dietr1ch]

It may be a step forward in terms of gathering data and defining the filter, but pushing up this filter might be expensive unless there's enough volume to justify it.

I took it as a comment on that generally models will be biased towards being nitty and that's something that needs to be dealt with as the incentives are not there to fix things at the origin.