What a year of 40,000+ voice inputs taught me: a good mood takes a good model.基于自己一年 4 万多条语音输入的分析结论: 好模型才能有好心情.
Moods and Models情绪与模型
I almost never swear at people, but I have always been extremely demanding of models.我对人几乎不怎么骂人, 但对模型一直特别挑剔.
Thank you, Opus 4.5, Opus 4.6, Opus 5, GPT-6 Astra and Opus 5.5.谢谢 Opus 4.5, Opus 4.6, Opus 5, GPT-6 Astra 和 Opus 5.5.
Screw you, GPT-5.4.去你的, GPT-5.4.
Moods by Domain不同领域的情绪
Transcripts are from MAI-Transcribe-2, which I already use for my own training data and which tops Zouyang Bench. Tags are from Haiku 5.5, on API credit Anthropic gave me.转写来自我做训练数据用的 MAI-Transcribe-2, 走样 bench 上的第一名. 标注用 Anthropic 送的额度跑了 Haiku 5.5.