Santiago 抛出个有意思的问题：AI 回复为啥总是那几句开头？是提示词没写好，还是模型本身就这样？评论区吵起来了。
Santiago 在 X 上提出一个问题：大量 AI 生成的评论式回复总以 "What really got me was..." 这类固定短语开头。他给出了两种可能解释：一是用户提示词写得不够好，二是模型本身生成多样化回复的能力有限。这条推文引发了对 LLM 输出同质化原因的讨论。
Why do many AI-generated slop replies start with "What really got me was..."?
Is it because people don't know how to prompt these models, or because the models aren't good enough to generate better replies?
Astrolab 要把一台内置大语言模型的机器人送上月球南极找水冰，太空探测也开始用上 LLM 了，挺新鲜。
让 9 个模型互相出题、互相解题，结果最强的解题者反而不是最好的出题者，小模型出题更狠，挺有意思的视角。
arXiv 上这个 Quant-Arena 基准挺有意思：让 GPT-6 Astra 等 9 个模型做真实 A 股量化策略，结果没有一个跑赢等权市场，排行榜对外开放提交。
这篇论文实测了各大时间推理基准，发现微调只涨本数据集分数，成绩靠的是知识记忆而不是推理，做评测的人该看看。
The shift here is from optimising for one best answer to building an archive of diverse, reasoned recommendations — useful when a decision maker needs options, not rankings. Practitioners running discovery-style tasks such as M&A screening or research triage can reuse the released code with their own candidate pools. The authors note limits: results rest on LLM judges, and real deployments would need fresh data and human expert feedback.
