Vals AI CEO @RayanKrishnan reveals frontier model writing has gotten worse over the last year despite improvements in verifiable domains, and the hyphenated word slop is kind of absurd at this point: "On some level, I'm still pretty optimistic that there are ways we can evaluate the subjective components of these models over time. It definitely feels like the writing capability of models has gotten worse over the last year, kind of in spite of improvements in other verifiable domains." "Just…