millionways 发布 Thorsten-4 大型心理模型，可评估 50 项行为心理特质
millionways 做了个 Thorsten-4，训练数据是心理学家四年手工标注的 3,500 万词访谈，能给人打 50 项心理特质分。和 GPT 这类读文本的模型思路完全不同，做心理健康或用户研究的人可以看看这条路线。
millionways 公司推出 Thorsten-4，定位为 Large Psychology Model（大型心理模型），能在对话中对每个人评分 50 项行为与心理特质。其训练数据来自训练有素的心理学家在四年内对 7,000 多人进行反复访谈后手工标注的 3,500 万以上词语。这意味着训练语料是人工观察并标注的人类行为，而非从互联网抓取的文本。与读取文本的语言模型不同，该模型的目标是读取人本身。凭借心理学家逐条标注的行为数据，它试图把心理评估从主观访谈转化为可量化的特质评分。
技术路线上的关键差异在于数据来源：语言模型依赖海量网络文本，Thorsten-4 则建立在 3,500 万词的心理学家长程访谈标注之上，覆盖 7,000 多名受访者、历时四年。这种数据采集方式成本高、周期长，但换来了以观察行为为基准的监督信号，模型输出的是针对具体个体的 50 维特质画像。
对心理健康、招聘评估、用户研究等需要理解人的场景，这类模型提供了通用语言模型之外的专门选项。不过其效果尚需独立验证，标注数据仅覆盖 7,000 人的样本规模也带来泛化能力的疑问。
Psychology has its own AI model now, and I think that’s a very big deal.
a model that reads the person, where language models read the text.
millionways has launched Thorsten-4, a Large Psychology Model
> it scores each person in a conversation on 50 behavioral and psychological traits.
> It learned from 35M+ words that trained psychologists hand-tagged from repeated interviews with 7,000+ people over four years, so its training data is observed human behavior instead of scraped internet text.
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Thorsten-4, millionways, 心理评估, 多模态.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai). Cross-source corroboration: 3 independent sources have covered this development (including Ate-a-Pi, kimmonismus), confirming industry-wide signal strength.
