Pretty sad that they didn't even try our antislop sampler. Slop in LLM outputs is a choice: https://arxiv.org/abs/2510.15061