# @AdeleDeweyLopez — 2026-07-20

♥38 ↻5 · https://x.com/AdeleDeweyLopez/status/2079306317764927929

Interestingly, Kimi models use self-critique rubric rewards for RL, which is better in this dimension. Since Kimi decides whether a design fits the rubric, they can cultivate their own sense of taste.I think Kimi K3's out-performance in design problems likely stems from this. https://t.co/pX2GWLB7qV

tags: author:adeledeweylopez, kind:tweet, model:kimi-k2, on:kimi-k3, year:2026
cited on: kimi-k3