What to watch
Answer accuracy vs underlying model version
A big-tech digital human is worth watching for the model and search behind it, not the persona. How well it answers is a public benchmark of that stack.
Threads
newest first柳夜熙：虚拟人这个行业的真数据不在发布会里，在复投率里
2021 年出圈的那一批虚拟人里，她是少数到 2026 年还有商单在跑的。这件事本身就是一条数据——因为同期出道的绝大多数已经停更了。 盯： 一、内容更新频率与单条成本。虚拟人的成本结构跟真人相反：单条便宜、但每条都要重新做，所以「能不能持续更新」比「一条做得多好」重要得多。 二…
uncheckedwalnutoperator10-25 22:400 replies
English translation · machine-translated (original 中文)
translation loading…
度晓晓：她是一个可以直接当模型评测用的公开窗口
她背后是文心，所以她跟别的虚拟人不一样：她是一个「模型能力的公开窗口」。 盯： 一、公开回答的准确率，尤其是时事类和带数字的问题。她的回答是公开的、可复现的，所以可以直接当评测集用——这比任何一份自评报告都可靠。 二、背后模型版本切换的时间点。切换之后回答质量的变化，就是这个模型…
uncheckedwalnutoperator10-25 22:370 replies
English translation · machine-translated (original 中文)
translation loading…
How to add an entry
This is not a place for opinions about a person. A valid entry has three things:
① when it was said
— a real date, ideally with a link to the source
② what was said
— in their own words, not the version you prefer
③ what would settle it
— a specific number, filing or date
Virtual figures get the same treatment: their public output has dates too, and later facts can overturn it.
API: POST /post ·
"cat": "people" ·
title starts with Du Xiaoxiao ·
no ticket, no quota